Campus is on a mission to maximize access to a world-class education

Campus

campus

By , Founder and CEO at Washlava. on . Todd Belveal Originally published Quora Everything on a college campus today is an experience. Students don’t just go to a dingy basement gym to work out. They go to a multi-million dollar rec center complete with state-of-the-art equipment and a lazy river. Universities are competing for students, shelling out huge amounts of money to prove they have the best amenities. Lavish dining facilities, roomy dorms, and pricey athleisure at the bookstore are all there to attract students. But for all the structural and cosmetic improvements being made on campus, laundry has been overlooked entirely. The laundry facilities on every major college campus still look pretty much the same as they did 20 years ago. Small, white washing machines and stacks of cheap tumble dryers lined up against the walls. Some of the machines are almost the same age as the students. Universities are underestimating the importance of laundry by outsourcing it to old school equipment providers that don’t care. They figure it’s not a big deal to students, but they’re making a huge mistake. Anyone can see universities need to rethink laundry. It’s bizarre how forgotten an amenity laundry is on most campuses. At most campuses, this is how it works: A student walks out of her dorm room with her hamper, heads down the stairs to the basement of her brand new residence hall, and what greets her? A depressing sight. Rows of archaic laundry machines with coin slots — and maybe a card reader on the wall for her student ID card. The machines are old, the payment systems are faulty, and the process quickly becomes a bottleneck. Especially once students figure out which machines leave their clothes soaking wet. Most students are coming from homes with an LG, a Whirlpool, or some other machine with a digital display and six speeds for everything from wool sweaters to towels. Then they get to the university washing machines, and they have two options — gentle cycle or normal cycle. They just dropped $500 on overpriced school apparel at the bookstore, and now they’re supposed to wash it all in these antiquated machines. University laundry is undoubtedly ready for a change. Most universities haven’t raised the price of laundry in a decade or more. It’s still a dollar to do a load of wash. And why would they raise prices? The experience doesn’t justify it one bit. In reality, students aren’t going to care about a dollar increase if it comes with new machines and better technology. Technology they expect. No one notices the difference between one and two dollars when they’re using their iPhone to reserve a brand new washer and dryer. But universities aren’t making that upgrade, in part because they keep agreeing to ridiculous, decade-long contracts that serve no one but the laundry equipment provider. Once the contract has been signed, there’s no incentive for the provider to be helpful or provide good service. If a student has a problem with a machine, it’ll sit for weeks with a piece of paper duct taped to it that says “out of order.” It might even be useless all semester. These universities need to make a change, and they’re in a great position to do so, because they have the perfect clientele for an upgrade. Students are the best users for new laundry technology. College students do more laundry than anyone. There are only so many clothes they can bring with them from home. And every time their favorite shirt needs a wash, that’s another load. Every student today has a smartphone — or they should. So, why not capitalize on the ubiquity of those devices and go with a coinless, efficient system for laundry? Most students rarely have cash on them, let alone coins. People get change when they make purchases with cash, and students don’t make purchases with cash. They use university ID cards or they pay with their debit cards. Students expect to live in a tech-based world, and universities constantly advertise themselves as cutting edge. They claim to have the newest technology and do the most exciting research. Their students are primed to integrate tech into every part of their lives. Yet their laundry facilities are from 1995, if not earlier. You can’t deny the perks of the updated system. The mobile-based laundry app we’re developing at Washlava isn’t even comparable to the current situation. Students can use the app to reserve machines, lock them, pay for loads, and even learn how to do laundry on our newly-installed machines. Our goal is to take a chore and turn it into an experience — just like all the other experiences these universities are advertising. And it’s working. Our pilot program at the University of Florida in Gainesville was a huge success by every measure. Students loved how convenient the experience was. They knew if and when they could get a machine, they were notified when their laundry was done, and they could pay for it all with their phones. But it wasn’t just the students who loved the machines. The residence hall staff told us they made the laundry room more efficient, and the new machines were a draw on campus tours. Residence hall directors were actually fighting over who would get Washlava first, and the management company that operates the residence halls issued a letter recommendation for Washlava. Those are the reactions of people who’ve been dealing with a terrible product and experience for much too long. College students deserve better laundry, and when their universities finally realize this, we’re going to give it to them one campus at a time. By , Founder and CEO at Washlava. on . Todd Belveal Originally published Quora For more trending tech answers from , visit . Quora HackerNoon.com/quora

Laundry Room Logic: Updating (Extremely Outdated) Campus Technologies

**ZUG, Switzerland, November 21st, 2023/Chainwire/--**Future3 Campus, in partnership with The Open Network (TON) Foundation, has today announced the launch of The TON Bootcamp, an incubation program to foster mini-app development within TON’s Web3 ecosystem in Telegram. The 15 selected teams may receive support for integrating into TON’s Web3 ecosystem and stand a chance to get up to USD $500,000 in funding from Future3 Campus. Applications to be selected for The TON Bootcamp are open until December 18, 2023. The TON Bootcamp is a Web2.5 incubator, empowering developers to create mini-apps that bridge the Web2 experience with the decentralized benefits of Web3. The program will run for five months and offer selected participants a chance to receive significant funding for their projects. The Bootcamp is focused on supporting projects building real-world payment and gaming solutions to seamlessly integrate into TON's flourishing Web3 ecosystem in Telegram. Participants in the Bootcamp may take advantage of a spectrum of marketplace opportunities, including showcasing their projects on , enjoying the same visibility that has brought other TON-based applications success. Tapps.Center TON Foundation's team of seasoned blockchain professionals will offer hands-on mentorship and guidance, helping developers navigate the intricacies of integrating their solutions within Telegram's ecosystem. The selected projects may also receive the funding and support necessary to deploy their mini-apps in a system designed for the mass adoption of Web3 products. These resources include marketing amplification within TON’s ecosystem and priority access to an active audience of over 800 million monthly users on Telegram. The TON Bootcamp may also equip projects with mentorship from leaders at top blockchain investment institutions and office spaces in Hong Kong, Singapore, and Japan. Bootcamp participants may have the opportunity to present their products and services to an audience of entrepreneurs, industry leaders, and investors at Web3 Festival 2024’s demo day pitch and showcase events. The TON Bootcamp is an unprecedented opportunity for mini-app developers to transform their visions into reality. Builders worldwide are encouraged to seize this opportunity before applications close on December 18, 2023. Projects seeking to apply for the TON Bootcamp may click here: https://echo3.ucraft.site About The Open Network: The Open Network (TON) is a global, decentralized blockchain community focused on putting crypto in every pocket. By building the Web3 ecosystem in Telegram Messenger, TON's vision is to empower 800 million users to own their digital identity, data, and assets by 2028. Learn more at . https://ton.org/ About Future3 Campus The Future3 Campus is a Web3.0-oriented incubation platform powered by Wanxiang Blockchain Labs and HashKey Capital with a mission to promote and foster entrepreneurship and innovation in Web3.0. Focusing on Web3.0 Massive Adoption, DePIN and AI, the Future3 Campus kicks off a $50 million seed fund in Web3.0 incubation worldwide. For more information, please visit , or contact us at . http://www.future3.io/#/home contact@future3.io Contact TON Foundation ton@theagencypartnership.com This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program . here

Future3 Campus And TON Foundation Announce Bootcamp For Mini-App Builders In TON's Ecosystem

authentic nCent team fist-bump Here at nCent Labs we spend a great deal of time thinking about how to unlock collective talent and appropriately attribute value within communities. These are the building blocks of efficient incentive markets. Why We Need nGage Campus Founders The field of incentive design is both broad and interdisciplinary, touching , , and even . However, applications of incentive to is an emergent field of thought inquiry. Making enough progress to support market growth will require a whole new generation of independent thinkers. These are mainly long term issues and questions. Our “meta” solution is ! Economics Sociology Computer Science Biology design blockchain nGage We are currently laser focused on growing a founder network on college campuses. The goals for this project are: To promote an entrepreneurial “founder” driven view of technical development of blockchains in general and incentive markets on blockchains in particular To build an nCent Global Campus network that will between students across the globe, and connect mentors with practitioners implementing incentive compatible designs in the market facilitate connections and knowledge transfer To to be more inclusive of diverse backgrounds and interdisciplinary viewpoints in guiding the development of the blockchain community expand the pipeline of talent How Does nGage Work? We will get back to you with an invitation. There are special prizes for referring your friends to apply! Apply to be a Founder here . This requires you and a co-founder and 10 total members. Everyone in the group will be able to recruit new members with a unique link from our campus nGage page. The only requirement for validated members at this time is that they should submit their email on the Campus Founder page, and follow the nCent Nation telegram group ( ) Recruit your friends to join your chapter. https://t.me/ncent Post a photo of your first meeting and we will send you a box of swag. We will also cover the cost of pizza and soda for your meeting (restrictions may apply). You will also be connected with our nCent Worldwide Campus Founders Network. This will have special offers and deals on many products and services, educational and curriculum content, special dev seminars and classes, early access to pre-release tools, opportunities to be a hub for grants and other airdrops on campus, and potentially tickets to nCent sponsored meetups and events across the Globe. We also have travel grants available to help select students to attend our annual conference, Receive Swag and nCent Global Campus Resources! nSpire.network whiteboarding in progress (on blackboard) Why are We So Excited About nGage? We are fired up to get student attention about the power of incentive markets and - we know there is an entrepreneurial hunger out there and we are eager to what the community will build with our tools and platform nCent Even if there is no short term economic payoff, we think getting smart college kids involved with our platform will be mutually beneficial. This is our way of “paying it forward” and we could see a world where that positive karma results in numerous internships, startups and other opportunities for these talented students down the road We want to have numerous “use cases” ready to go when our main net launches. (Some of our current use cases include , , , and even ). We cannot possibly do it all alone. The fervor and community spirit of young people in College is unmatched and unstoppable — we want to be inspired by YOU! employee referral programs volunteering projects online auctions dating apps Learn more by emailing me directly at kk@ncnt.io or visiting campus.ncent.io . To stay in the nCent loop, hear me tweet and join our international telegram channel . — PS the .edu/student email address that refers the most people to sign up on our website will get extra swag (we are currently weighing sunglasses, water bottles, or notebooks, but if the referral count is >50 will likely be all 3!)

Introducing nGage: the nCent Labs Worldwide Campus Founder’s Program

I usually don’t write blogs on my personal experiences but this was one such experience that I feel like sharing with you all because it is really worth it. Ever since I entered engineering college, I was really excited about gaining a lot of practical knowledge apart from the regular learnings from the engineering curriculum. One fine day I was chatting with one of my friends & I told him about my keen desire to learn about startups and gain some practical experience. I also mentioned my love for meeting different people and listening to their stories. He suggested that I attend some workshops by and also intern with them if possible.So I spoke to , Founder of HelloMeets about interning with them. At that point she was hiring interns who could work for full time but being a college student that wasn’t possible for me. HelloMeets Sahiba Sethi Then one day, I was going through their website and I found the campus Ambassador Programme there. It really got me interested. They were offering the following to their campus ambassadors:- — Ambassadors could attend a few Meetups for free — Certificates — Opportunity to network with different people I immediately filled the form & the very next day received a call for an interview. After the interview I got selected as the Campus Ambassador. From sharing events on Social Media to handling an event on my own An interesting fact is that I used to just share events on social media and never attended a single meet up until one day when Sahiba texted me saying that I should attend the next meetup on . Those days as it is I was learning about social media marketing. So, I attended the event and it turned out to be really helpful for me. HelloMeets Content-based Marketing by Nalini Kher Also, I was the youngest one present in that workshop. Initially, I felt awkward, but when I interacted with all the attendees, I found them to be very interesting. This meetup boosted my confidence to talk to and network with people. Some key learnings from that event (October 2016)- I got to learn more about content marketing and speaker also gave us a sneak peak into marketing trends of 2016 We had a hands-on workshop session where I designed a campaign for Sheroes which was appreciated by all Then my exams came and I again got non-serious about HelloMeets. And honestly saying that was the time I missed some really awesome meetups like the one on Intermediate Photoshop & Wordpress Website Development. After that Sahiba arranged a Campus Ambassador Meet, where we discussed about an upcoming event — Hello Christmas Party (the only free event that HelloMeets does for its existing customers). It was a pool side party at Hotel Ashoka with food and live music. I got to meet and network with many entrepreneurs. That party had everything I loved and from that point, I started talking to Sahiba on a regular basis to learn anything that I could. Entry to the Christmas party was free for all the campus ambassadors. Once I asked her if I could volunteer at one of the Hiring Events. She agreed and asked me to come and manage the event. As she was going to be in Bangalore, I along with (Marketing, )& (who was a volunteer for the events that time) handled the event. It was a great experience for me in terms of learning how to talk to the startup founders, how to keep everyone happy and work with new people in a team. Ayushi Mohindra HelloMeets Vaibhav Gupta Then I got another chance to coordinate a hiring event. From there onwards I started my journey as Event Coordinator at HelloMeets. I am happy that I am connected with HelloMeets that believes in overall development of everyone, be it a campus ambassador or a full-time employee. Few months ago I saw the post on Facebook about an AI workshop by HelloMeets and I told Sahiba that I was very keen to attend it. She gave me the responsibility to manage the whole event on my own. That day due to some reason the workshop didn’t go well. We were not able to meet the expectations of the attendees. Learnings- I learned about a great topic like AI and that too for free !! :D I learned how to handle a situation where somethings go wrong People actually complained about their money being wasted as it was an expensive workshop. After that what Sahiba did was really surprising to me, she returned everyone’s money. I asked her why she did that and she said: “Money is important but when we ourselves know that we were not able to provide value to our customers then it is unethical to earn in that situation.” Some of my personal favorite meetups are the one on Product Management by Lakshay Pandey and Conversational UX by Neha Sharma. Reasons why I liked these meetups so much- Conversational UX- Learnt of a trendy way of UX designing and why conversations are important Product management- learnt how products are built and what all goes into it Then came our 2 weekends boot camp on UX designing. For this event, I managed and coordinated with the speakers and attendees on all the 4 days. Learnings- Being punctual- As we had this event in the morning I was supposed to reach even before the speakers!! It was quite a task for me. Handling the backend operations of an event Learning about User experiencing and why it plays such an important role in Design Getting feedbacks from the attendees and speakers This bootcamp was one of the best events organised by us with great speakers , , , and an amazing audience. Avisek Bhattacharya Ikshat Jain Anulal VS Riken Patel For details of UX bootcamp, you can read my blogs on the entire bootcamp event coming soon…. Now I feel that I can manage & organise workshops like these.I am also planning a startup related to workshop management for music, art, poetryn and Sahiba is advising and helping me build it. It has almost been a year at Hellomeets and now Sahiba has asked me to be the Community Head for Delhi. She is planning to expand to other cities and needs to focus on that so I have been given this new responsibility. It is a huge achievement for me that I am going to be the Community Head for Delhi very soon!! HelloMeets Blog credits: Naman Goel To join us as a Campus Ambassador, sign up here :) Some of our upcoming meetups: Investors & Founders Meetup — 8th July — Bengaluru GST-Opportunities & Challenges for Startups — 8th July — Delhi Augmented Reality Workshop — 8th July — Delhi The Art of Mobile App Testing — 16th July — Delhi

Journey From Campus Ambassador to Community Head, Delhi @HelloMeets

How I Implemented Access Approval in Our Open Source Project

আমি কীভাবে আমাদের ওপেন সোর্স প্রকল্পে অ্যাক্সেস অনুমোদন বাস্তবায়ন করেছি

Wie ich die Zugriffsgenehmigung in unserem Open-Source-Projekt implementiert habe

Cómo implementé la aprobación de acceso en nuestro proyecto de código abierto

Comment j'ai implémenté l'approbation d'accès dans notre projet Open Source

मैंने अपने ओपन सोर्स प्रोजेक्ट में एक्सेस स्वीकृति कैसे लागू की

オープンソース プロジェクトでアクセス承認を実装した方法

오픈 소스 프로젝트에서 액세스 승인을 구현한 방법

Como implementei a aprovação de acesso em nosso projeto de código aberto

Как я реализовал одобрение доступа в нашем проекте с открытым исходным кодом

Açık Kaynak Projemizde Erişim Onayını Nasıl Uyguladım

Cách tôi triển khai phê duyệt quyền truy cập trong dự án nguồn mở của chúng tôi

我如何在我们的开源项目中实施访问批准

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Conclusion, and References

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 9 DISCUSSIONS AND FUTURE WORKS Problem Scale. One limitation of LifeHD is the relative small problem scale (e.g., the image size of CIFAR-100 is restricted to 32x32) due to the essential difficulty of unsupervised lifelong learning problem, including single-pass non-iid data and no supervision. For the same reason, there remains a disparity in accuracy between unsupervised lifelong learning and fully supervised NNs, as substantiated by prior research [13, 54]. In order to scale LifeHD to more challenging applications such as self-driving vehicles, one possible direction is to leverage the pretrained foundation model as a frozen feature extractor in the HDnn framework, which we leave for future investigation. Hyperparameter Tuning. While we recognize that hyperparameters can influence the performance of LifeHD, such an issue is not exclusive to LifeHD, but has persistently been a challenge in machine learning research [7]. In LifeHD, the impact of hyperparameters can be mitigated through pre-deployment evaluation and component co-design. For example, encoding parameters such as 𝑄, 𝑃 can be tuned on similar health monitoring data sources prior to deployment. Meanwhile, the component of cluster HVs merging can increase LifeHD’s resiliency to the novelty detection threshold 𝛾, as a higher quantity of novel clusters can be merged in later stage of learning. Limitations of HDC. HDC serves as the fundamental core of LifeHD. While HDC shows promise with its notable lightweight design, it is burdened by several limitations that remain active areas of research. First, for complex datasets like audio and images, HDC requires a pretrained feature extractor (the HDnn encoding) which may not exist for certain applications. Moreover, akin to any other architecture, HDC vectors face capacity limitations determined by the dimension of HD space, encoding method, and potential noise levels in the input data [56]. Due to these factors, careful evaluation and sometimes manual feature engineering are required to successfully deploy HDC for new applications. Future Works. Although LifeHD focuses on single-device lifelong learning for classification tasks, the method can be extended for other types of tasks and learning settings, such as federated learning and reinforcement learning. We leave the investigation of these topics for future work. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 9 DISCUSSIONS AND FUTURE WORKS Problem Scale. One limitation of LifeHD is the relative small problem scale (e.g., the image size of CIFAR-100 is restricted to 32x32) due to the essential difficulty of unsupervised lifelong learning problem, including single-pass non-iid data and no supervision. For the same reason, there remains a disparity in accuracy between unsupervised lifelong learning and fully supervised NNs, as substantiated by prior research [13, 54]. In order to scale LifeHD to more challenging applications such as self-driving vehicles, one possible direction is to leverage the pretrained foundation model as a frozen feature extractor in the HDnn framework, which we leave for future investigation. Problem Scale. Hyperparameter Tuning. While we recognize that hyperparameters can influence the performance of LifeHD, such an issue is not exclusive to LifeHD, but has persistently been a challenge in machine learning research [7]. In LifeHD, the impact of hyperparameters can be mitigated through pre-deployment evaluation and component co-design. For example, encoding parameters such as 𝑄, 𝑃 can be tuned on similar health monitoring data sources prior to deployment. Meanwhile, the component of cluster HVs merging can increase LifeHD’s resiliency to the novelty detection threshold 𝛾, as a higher quantity of novel clusters can be merged in later stage of learning. Hyperparameter Tuning. Limitations of HDC. HDC serves as the fundamental core of LifeHD. While HDC shows promise with its notable lightweight design, it is burdened by several limitations that remain active areas of research. First, for complex datasets like audio and images, HDC requires a pretrained feature extractor (the HDnn encoding) which may not exist for certain applications. Moreover, akin to any other architecture, HDC vectors face capacity limitations determined by the dimension of HD space, encoding method, and potential noise levels in the input data [56]. Due to these factors, careful evaluation and sometimes manual feature engineering are required to successfully deploy HDC for new applications. Limitations of HDC. Future Works. Although LifeHD focuses on single-device lifelong learning for classification tasks, the method can be extended for other types of tasks and learning settings, such as federated learning and reinforcement learning. We leave the investigation of these topics for future work. Future Works. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Discussions and Future Works

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 8 EVALUATION OF LIFEHDsemi AND LIFEHDa In this section, we compare LifeHDsemi and LifeHDa, our proposed extensions from LifeHD, with existing designs that are similar. Performance of LifeHDsemi. To evaluate LifeHDsemi in a lowlabel scenario, we compare it with SemiHD [22], which is the stateof-the-art HDC method for semi-supervised learning. We adapt SemiHD [22] for single-pass settings, introducing a pseudolabel assignment threshold. When the cosine similarity of an unlabeled sample to the nearest class hypervector surpasses the threshold, we assign that class as its pseudolabel. The sample is then employed to update the class hypervector in SemiHD. We explore various threshold values and choose the optimal result for comparison. Fig. 15 (a) compares LifeHDsemi and SemiHD [22] on ESC-50 and CIFAR100 across various labeling ratios 𝑟 < 0.01. The advantages of LifeHDsemi are most prominent when labels are limited, the weakly supervised scenario is LifeHDsemi’s major focus. LifeHDsemi improves ACC by up to 10.25% and 3.6% on ESC-50 and CIFAR-100 respectively. This outcome arises from the unsupervised nature of LifeHD, allowing it to autonomously organize prominent cluster HVs, especially when all samples from a class lack labels. As the labeling ratio increases, LifeHDsemi’s advantage over SemiHD diminishes, because more labels bolster SemiHD’s performance. Performance of LifeHDa. LifeHDa provides an interface to trade minimal performance loss for efficiency gains, by adaptively pruning out the insignificant dimensions. We compare LifeHDa with previous HDC works employing a fixed mask throughout training [25], and the results are presented in Fig. 15 (b) for CIFAR100, including ACC and training latency per batch on RPi 4B. Fixed masks negatively impact HDC learning, especially with smaller dimensions. Such masks fail to adapt to new hypervectors in classincremental streams, where less significant dimensions may become crucial later in training. LifeHDa addresses this issue by adjusting the mask upon novelty detection, leading to a degradation of only 0.71% in ACC and 4.5x efficiency gain compared to the complete LifeHD, using only 20% of the full HD dimension of LifeHD. The overhead of adaptively adjusting the mask is negligible when novelty detection occurs infrequently. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 8 EVALUATION OF LIFEHDsemi AND LIFEHDa In this section, we compare LifeHDsemi and LifeHDa, our proposed extensions from LifeHD, with existing designs that are similar. Performance of LifeHDsemi. To evaluate LifeHDsemi in a lowlabel scenario, we compare it with SemiHD [22], which is the stateof-the-art HDC method for semi-supervised learning. We adapt SemiHD [22] for single-pass settings, introducing a pseudolabel assignment threshold. When the cosine similarity of an unlabeled sample to the nearest class hypervector surpasses the threshold, we assign that class as its pseudolabel. The sample is then employed to update the class hypervector in SemiHD. We explore various threshold values and choose the optimal result for comparison. Fig. 15 (a) compares LifeHDsemi and SemiHD [22] on ESC-50 and CIFAR100 across various labeling ratios 𝑟 < 0.01. The advantages of LifeHDsemi are most prominent when labels are limited, the weakly supervised scenario is LifeHDsemi’s major focus. LifeHDsemi improves ACC by up to 10.25% and 3.6% on ESC-50 and CIFAR-100 respectively. This outcome arises from the unsupervised nature of LifeHD, allowing it to autonomously organize prominent cluster HVs, especially when all samples from a class lack labels. As the labeling ratio increases, LifeHDsemi’s advantage over SemiHD diminishes, because more labels bolster SemiHD’s performance. Performance of LifeHDsemi. Performance of LifeHDa. LifeHDa provides an interface to trade minimal performance loss for efficiency gains, by adaptively pruning out the insignificant dimensions. We compare LifeHDa with previous HDC works employing a fixed mask throughout training [25], and the results are presented in Fig. 15 (b) for CIFAR100, including ACC and training latency per batch on RPi 4B. Fixed masks negatively impact HDC learning, especially with smaller dimensions. Such masks fail to adapt to new hypervectors in classincremental streams, where less significant dimensions may become crucial later in training. LifeHDa addresses this issue by adjusting the mask upon novelty detection, leading to a degradation of only 0.71% in ACC and 4.5x efficiency gain compared to the complete LifeHD, using only 20% of the full HD dimension of LifeHD. The overhead of adaptively adjusting the mask is negligible when novelty detection occurs infrequently. Performance of LifeHDa. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Evaluation of LifeHDsemi & a

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Evaluation of LifeHD

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 6 VARIANTS OF LIFEHD While LifeHD is designed to cater to general IoT applications with streaming input and without supervision, real-world scenarios may vary. Some scenarios might have a few labeled samples in addition to the unlabeled stream, while others may require operation within strict power constraints. LifeHD offers extensibility to address these diverse needs. In this section, we introduce two software-based extensions: LifeHDsemi, which adds a separate processing path to manage labeled samples, and LifeHDa, which adaptively prunes the HDC model using masking to handle low-power scenarios. 6.1 LifeHDsemi 6.2 LifeHDa While HDC computation is typically lightweight, there may be instances of energy scarcity (e.g., when powered by a solar panel) that call for a balance between accuracy and power efficiency. Similar to neural networks, one approach is to prune the HDC model using a mask, retaining the most crucial HDC dimensions postencoding [25]. Dimension importance can be determined by aggregating all class hypervectors into one and sorting the values across all dimensions. Notably, direct reduction of the encoding dimension should be avoided, as it can degrade HDC’s expressive capability and end up with corruption. Fig. 7 (left) visually demonstrates the impact of masking in supervised HDC tasks: retaining the top 6000 bits incurs only a 3% accuracy loss compared to using the full 10K-bit precision. While the concept of masking has been employed in prior HDC studies [25], they are not directly applicable to LifeHD due to their offline training setting with iid data. With streaming non-iid data in LifeHD, the set of observed cluster HVs may only represent a subset of the potential classes, and the less significant bits could become crucial as new classes are introduced. We introduce LifeHDa, which enhances LifeHD through an adaptive masking approach applied to all cluster HVs in working and long-term memory. Let 𝐷𝑎 represent the target dimension for reduction. The rationale behind LifeHDa is depicted in Fig. 7 (right). Whenever an original LifeHD detects novelty, we temporarily revert to the full dimension 𝐷 for 2 batches, which is sufficient for LifeHD to consolidate new patterns in its memory. After these two batches, we assess the long-term memory cluster HVs by aggregating them and ranking the dimensions, and derive a mask retaining 𝐷𝑎 dimensions with the largest absolute values. This mask is then applied to the following batches of 𝜙 (𝑋) immediately after encoding, up until the next novelty is detected. Novelty detection is executed with the masked hypervectors. Importantly, LifeHDa can utilize the same novelty detection sensitivity as LifeHD since the most significant dimensions dominate the similarity check. In other words, the similarity results in LifeHDa using 𝐷𝑎 dimension are similar as using the full dimension. LifeHDa offers an adaptive HDC model pruning interface with minimal accuracy loss and overhead. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 6 VARIANTS OF LIFEHD While LifeHD is designed to cater to general IoT applications with streaming input and without supervision, real-world scenarios may vary. Some scenarios might have a few labeled samples in addition to the unlabeled stream, while others may require operation within strict power constraints. LifeHD offers extensibility to address these diverse needs. In this section, we introduce two software-based extensions: LifeHDsemi, which adds a separate processing path to manage labeled samples, and LifeHDa, which adaptively prunes the HDC model using masking to handle low-power scenarios. 6.1 LifeHDsemi 6.2 LifeHDa While HDC computation is typically lightweight, there may be instances of energy scarcity (e.g., when powered by a solar panel) that call for a balance between accuracy and power efficiency. Similar to neural networks, one approach is to prune the HDC model using a mask, retaining the most crucial HDC dimensions postencoding [25]. Dimension importance can be determined by aggregating all class hypervectors into one and sorting the values across all dimensions. Notably, direct reduction of the encoding dimension should be avoided, as it can degrade HDC’s expressive capability and end up with corruption. Fig. 7 (left) visually demonstrates the impact of masking in supervised HDC tasks: retaining the top 6000 bits incurs only a 3% accuracy loss compared to using the full 10K-bit precision. While the concept of masking has been employed in prior HDC studies [25], they are not directly applicable to LifeHD due to their offline training setting with iid data. With streaming non-iid data in LifeHD, the set of observed cluster HVs may only represent a subset of the potential classes, and the less significant bits could become crucial as new classes are introduced. We introduce LifeHDa, which enhances LifeHD through an adaptive masking approach applied to all cluster HVs in working and long-term memory. Let 𝐷𝑎 represent the target dimension for reduction. The rationale behind LifeHDa is depicted in Fig. 7 (right). Whenever an original LifeHD detects novelty, we temporarily revert to the full dimension 𝐷 for 2 batches, which is sufficient for LifeHD to consolidate new patterns in its memory. After these two batches, we assess the long-term memory cluster HVs by aggregating them and ranking the dimensions, and derive a mask retaining 𝐷𝑎 dimensions with the largest absolute values. This mask is then applied to the following batches of 𝜙 (𝑋) immediately after encoding, up until the next novelty is detected. Novelty detection is executed with the masked hypervectors. Importantly, LifeHDa can utilize the same novelty detection sensitivity as LifeHD since the most significant dimensions dominate the similarity check. In other words, the similarity results in LifeHDa using 𝐷𝑎 dimension are similar as using the full dimension. LifeHDa offers an adaptive HDC model pruning interface with minimal accuracy loss and overhead. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Variants of LifeHD

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 5 LIFEHD In this section, we present the design of LifeHD, the first unsupervised HDC framework for lifelong learning in general edge IoT applications. Compared to operating in the original data space, HDC improves pattern separability through sparsity and high dimensionality, making it more resilient against catastrophic forgetting [52]. LifeHD preserves the advantages of HDC in computational efficiency and lifelong learning, while handling the input of unlabeled streaming data, which has not been achieved in previous work [18, 21, 22, 41, 65]. 5.1 LifeHD Overview Fig. 4 gives an overview of how LifeHD works. The first step is HDC encoding of data into hypervectors as described in Sec. 3. Training samples 𝑋 are organized into batches of size 𝑏𝑆𝑖𝑧𝑒 and input into an optional fixed NN for feature extraction (e.g. for images and sound) and the encoding module. The encoded hypervectors 𝜙 (𝑋) are input to LifeHD’s two-tier memory design inspired by cognitive science studies [5], consisting of working memory and long-term memory. This memory system intelligently and dynamically manages historical patterns, stored as hypervectors and referred to as cluster HVs. As shown in Fig. 4, the working memory is designed with three components: novelty detection, cluster HV update and cluster HV merge. 𝜙 (𝑋) is first input into novelty detection step (1). An insertion to the cluster HVs is made if a novelty flag is raised, otherwise 𝜙 (𝑋) updates the existing cluster HVs (2). The third component, cluster HV merge (3), retrieves the cluster HVs from long-term memory, and merges similar cluster HVs into a supercluster via a novel spectral clustering-based merging algorithm [59]. The interaction between working and long-term memory happens as commonly encountered cluster HVs are copied to long-term memory, which we call consolidation (4). Finally, when the size limit of either working or long-term memory is reached, the least recently used cluster HVs are forgotten (5). ecently used cluster HVs are forgotten (○5 ). All modules in LifeHD work collaboratively, making it adaptive and robust to continuously changing streams without relying on any form of prior knowledge. For example, in scenarios of distribution drift, LifeHD may generate new cluster HVs upon encountering drifted samples initially, which can later be merged into coarse clusters. This approach ensures that LifeHD can efficiently capture and retain historical patterns. In the following, we discuss more details about the major components of LifeHD: novelty detection (Sec. 5.2), cluster HV update (Sec. 5.3), and cluster HV merging (Sec. 5.4). We summarize the important notations used in this paper in Table 1. 5.2 Novelty Detection The initial novelty detection step (1 in Fig. 4) is crucial for identifying emerging patterns in the environment. SupposeM = {𝑚1, ...,𝑚𝑀} is the set of cluster HVs stored in the working memory. We gauge the "radius" of each cluster by tracking two scalars for each cluster HV 𝑖: 𝜇𝑖 and 𝜎ˆ𝑖 , which represent the mean cosine difference and standard difference between the cluster HV and its assigned inputs. Given 𝜙 (𝑋), we first identify the most similar cluster HV, denoted by 𝑗. LifeHD marks 𝜙 (𝑋) as “novel" if it substantially differs from its nearest cluster HV. Specifically, this dissimilarity is measured by comparing cos(𝜙 (𝑋),𝑚𝑗) with a threshold based on the historical distance distribution of cluster HV 𝑗: The hyperparameter 𝛾 fine-tunes the sensitivity to novelties. LifeHD recognizes new 𝜙 (𝑋) as prototypes and inserts them into the working memory. When reaching its size limit 𝑀, the working memory experiences forgetting (○5 in Fig. 4). The least recently used (LRU) cluster HV, represented by 𝐿𝑅𝑈 = argmin𝑀 𝑖=1 𝑝𝑖 , is replaced. Here 𝑝 corresponds to the latest batch index where the cluster HV was accessed. A similar forgetting mechanism is configured for the long-term memory, where the last batch accessed is marked with 𝑞. 5.3 Cluster HV Update If novelty is not detected, indicating that 𝜙 (𝑋) closely matches cluster HV 𝑗, we proceed to update the cluster HV and its associated information (2 in Fig. 4). This update process involves bundling 𝜙 (𝑋) with cluster HV 𝑚𝑗 , akin to how class hypervectors are updated as described in Sec. 3, and updated 𝜇𝑗 and 𝜎ˆ𝑗 with their moving average The hyperparameter 𝛼 adjusts the balance between historical and recent inputs, where a higher 𝛼 gives more weight to recent samples. Properly maintaining 𝜇𝑗 and 𝜎ˆ𝑗 is vital for tracking the “radius” of each cluster HV, affecting future novelty detection. We also increase the hit frequency ℎ𝑖𝑡𝑗 and refresh 𝑝𝑗 with current batch index 𝑖𝑑𝑥. ℎ𝑖𝑡𝑗 is further used to compared with a predetermined threshold ℎ𝑖𝑡𝑡ℎ to decide when a working memory cluster HV appears sufficiently frequently to be consolidated to long-term memory (4 in Fig. 4). 𝑝𝑗 determines forgetting as described in the previous section. With this lightweight approach, LifeHD continually records temporal cluster HVs from the environment, while the most prominent cluster HVs are transferred to long-term memory. 5.4 Cluster HV Merging Cluster HV merge (3 in Fig. 4) has the dual benefit of reducing memory use and of elucidating underlying similarity structure in the data. Intuitively, a group of cluster HVs can be merged if they are similar to each other and dissimilar from other cluster HVs. For instance, one might merge the cluster HVs for Bulldog and Chihuahua into a single “Dog” cluster HV, that remains distinct from the cluster HV for “Tabby Cat”. To merge the cluster HVs, we first construct a similarity graph defined over the cluster HVs from the long-term memory. The cluster HVs correspond to nodes, and a pair of cluster HVs are connected by an edge if they are sufficiently similar. We then merge the cluster HVs by computing a particular type of cut in the graph in a manner similar to spectral clustering [40]. This graph based formalism for clustering is able to capture complex types of cluster geometry and often substantially outperforms simpler approaches like K-Means [59]. We detail the steps of cluster HV merging in LifeHD below, while Fig. 5 offers an illustrative overview. Step 2: Decomposition. We compute the Laplacian𝑊 = 𝐷 −𝐴, where 𝐷 is the diagonal matrix in which 𝐷𝑖𝑖 = Í 𝑗 𝐴𝑖𝑗 . We then compute the eigenvalues 𝜆1, .., 𝜆𝐿, sorted in increasing order, and eigenvectors 𝜈1, ..., 𝜈𝐿 of 𝑊. Step 3: Grouping. We infer 𝑘 = max𝑖∈ [𝐿] 𝜆𝑖 ≤ 𝑔𝑢𝑏, and merge the cluster HVs by running K-Means on 𝜈1, ..., 𝜈𝑘 . The upperbound 𝑔𝑢𝑏 is a hyperparameter that adjusts the granularity of merging, with a smaller 𝑔𝑢𝑏 leading to smaller 𝑘 thus encouraging merging more aggressively. Our merging approach is formally grounded, as discussed in [59]. It is a well-known fact that the eigenvectors of 𝑊 encode information about the connected components of G. When G has 𝑘 connected components, the eigenvalues 𝜆1 = 𝜆2 = ... = 𝜆𝑘 = 0. To recover these components, K-Means clustering on 𝜈1, ..., 𝜈𝑘 can be employed, as explained in [59]. However, practical scenarios may have a few inter-component edges that should ideally be distinct. For instance, when the similarity threshold is imprecisely set, erroneous edges may appear in the graph, causing 𝜆1, ..., 𝜆𝑘 to be only approximately zero. Our merging approach is designed to handle this situation by introducing 𝑔𝑢𝑏. The cluster HV merging is evaluated every 𝑓𝑚𝑒𝑟𝑔𝑒 batches, where 𝑓𝑚𝑒𝑟𝑔𝑒 is a hyperparameter that controls the trade-off between merging performance and computational latency. Both 𝑔𝑢𝑏 and 𝑓𝑚𝑒𝑟𝑔𝑒 are analyzed in Sec. 7.8, along with other key hyperparameters in LifeHD. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 5 LIFEHD In this section, we present the design of LifeHD, the first unsupervised HDC framework for lifelong learning in general edge IoT applications. Compared to operating in the original data space, HDC improves pattern separability through sparsity and high dimensionality, making it more resilient against catastrophic forgetting [52]. LifeHD preserves the advantages of HDC in computational efficiency and lifelong learning, while handling the input of unlabeled streaming data, which has not been achieved in previous work [18, 21, 22, 41, 65]. 5.1 LifeHD Overview Fig. 4 gives an overview of how LifeHD works. The first step is HDC encoding of data into hypervectors as described in Sec. 3. Training samples 𝑋 are organized into batches of size 𝑏𝑆𝑖𝑧𝑒 and input into an optional fixed NN for feature extraction (e.g. for images and sound) and the encoding module. The encoded hypervectors 𝜙 (𝑋) are input to LifeHD’s two-tier memory design inspired by cognitive science studies [5], consisting of working memory and long-term memory. This memory system intelligently and dynamically manages historical patterns, stored as hypervectors and referred to as cluster HVs. As shown in Fig. 4, the working memory is designed with three components: novelty detection, cluster HV update and cluster HV merge. 𝜙 (𝑋) is first input into novelty detection step (1). An insertion to the cluster HVs is made if a novelty flag is raised, otherwise 𝜙 (𝑋) updates the existing cluster HVs (2). The third component, cluster HV merge (3), retrieves the cluster HVs from long-term memory, and merges similar cluster HVs into a supercluster via a novel spectral clustering-based merging algorithm [59]. The interaction between working and long-term memory happens as commonly encountered cluster HVs are copied to long-term memory, which we call consolidation (4). Finally, when the size limit of either working or long-term memory is reached, the least recently used cluster HVs are forgotten (5). ecently used cluster HVs are forgotten (○5 ). All modules in LifeHD work collaboratively, making it adaptive and robust to continuously changing streams without relying on any form of prior knowledge. For example, in scenarios of distribution drift, LifeHD may generate new cluster HVs upon encountering drifted samples initially, which can later be merged into coarse clusters. This approach ensures that LifeHD can efficiently capture and retain historical patterns. In the following, we discuss more details about the major components of LifeHD: novelty detection (Sec. 5.2), cluster HV update (Sec. 5.3), and cluster HV merging (Sec. 5.4). We summarize the important notations used in this paper in Table 1. 5.2 Novelty Detection The initial novelty detection step (1 in Fig. 4) is crucial for identifying emerging patterns in the environment. SupposeM = {𝑚1, ...,𝑚𝑀} is the set of cluster HVs stored in the working memory. We gauge the "radius" of each cluster by tracking two scalars for each cluster HV 𝑖: 𝜇𝑖 and 𝜎ˆ𝑖 , which represent the mean cosine difference and standard difference between the cluster HV and its assigned inputs. Given 𝜙 (𝑋), we first identify the most similar cluster HV, denoted by 𝑗. LifeHD marks 𝜙 (𝑋) as “novel" if it substantially differs from its nearest cluster HV. Specifically, this dissimilarity is measured by comparing cos(𝜙 (𝑋),𝑚𝑗) with a threshold based on the historical distance distribution of cluster HV 𝑗: The hyperparameter 𝛾 fine-tunes the sensitivity to novelties. LifeHD recognizes new 𝜙 (𝑋) as prototypes and inserts them into the working memory. When reaching its size limit 𝑀, the working memory experiences forgetting (○5 in Fig. 4). The least recently used (LRU) cluster HV, represented by 𝐿𝑅𝑈 = argmin𝑀 𝑖=1 𝑝𝑖 , is replaced. Here 𝑝 corresponds to the latest batch index where the cluster HV was accessed. A similar forgetting mechanism is configured for the long-term memory, where the last batch accessed is marked with 𝑞. 5.3 Cluster HV Update If novelty is not detected, indicating that 𝜙 (𝑋) closely matches cluster HV 𝑗, we proceed to update the cluster HV and its associated information (2 in Fig. 4). This update process involves bundling 𝜙 (𝑋) with cluster HV 𝑚𝑗 , akin to how class hypervectors are updated as described in Sec. 3, and updated 𝜇𝑗 and 𝜎ˆ𝑗 with their moving average The hyperparameter 𝛼 adjusts the balance between historical and recent inputs, where a higher 𝛼 gives more weight to recent samples. Properly maintaining 𝜇𝑗 and 𝜎ˆ𝑗 is vital for tracking the “radius” of each cluster HV, affecting future novelty detection. We also increase the hit frequency ℎ𝑖𝑡𝑗 and refresh 𝑝𝑗 with current batch index 𝑖𝑑𝑥. ℎ𝑖𝑡𝑗 is further used to compared with a predetermined threshold ℎ𝑖𝑡𝑡ℎ to decide when a working memory cluster HV appears sufficiently frequently to be consolidated to long-term memory (4 in Fig. 4). 𝑝𝑗 determines forgetting as described in the previous section. With this lightweight approach, LifeHD continually records temporal cluster HVs from the environment, while the most prominent cluster HVs are transferred to long-term memory. 5.4 Cluster HV Merging Cluster HV merge (3 in Fig. 4) has the dual benefit of reducing memory use and of elucidating underlying similarity structure in the data. Intuitively, a group of cluster HVs can be merged if they are similar to each other and dissimilar from other cluster HVs. For instance, one might merge the cluster HVs for Bulldog and Chihuahua into a single “Dog” cluster HV, that remains distinct from the cluster HV for “Tabby Cat”. To merge the cluster HVs, we first construct a similarity graph defined over the cluster HVs from the long-term memory. The cluster HVs correspond to nodes, and a pair of cluster HVs are connected by an edge if they are sufficiently similar. We then merge the cluster HVs by computing a particular type of cut in the graph in a manner similar to spectral clustering [40]. This graph based formalism for clustering is able to capture complex types of cluster geometry and often substantially outperforms simpler approaches like K-Means [59]. We detail the steps of cluster HV merging in LifeHD below, while Fig. 5 offers an illustrative overview. Step 2: Decomposition. We compute the Laplacian𝑊 = 𝐷 −𝐴, where 𝐷 is the diagonal matrix in which 𝐷𝑖𝑖 = Í 𝑗 𝐴𝑖𝑗 . We then compute the eigenvalues 𝜆1, .., 𝜆𝐿, sorted in increasing order, and eigenvectors 𝜈1, ..., 𝜈𝐿 of 𝑊. Step 2: Decomposition. Step 3: Grouping. We infer 𝑘 = max𝑖∈ [𝐿] 𝜆𝑖 ≤ 𝑔𝑢𝑏, and merge the cluster HVs by running K-Means on 𝜈1, ..., 𝜈𝑘 . The upperbound 𝑔𝑢𝑏 is a hyperparameter that adjusts the granularity of merging, with a smaller 𝑔𝑢𝑏 leading to smaller 𝑘 thus encouraging merging more aggressively. Our merging approach is formally grounded, as discussed in [59]. It is a well-known fact that the eigenvectors of 𝑊 encode information about the connected components of G. When G has 𝑘 connected components, the eigenvalues 𝜆1 = 𝜆2 = ... = 𝜆𝑘 = 0. To recover these components, K-Means clustering on 𝜈1, ..., 𝜈𝑘 can be employed, as explained in [59]. However, practical scenarios may have a few inter-component edges that should ideally be distinct. For instance, when the similarity threshold is imprecisely set, erroneous edges may appear in the graph, causing 𝜆1, ..., 𝜆𝑘 to be only approximately zero. Our merging approach is designed to handle this situation by introducing 𝑔𝑢𝑏. The cluster HV merging is evaluated every 𝑓𝑚𝑒𝑟𝑔𝑒 batches, where 𝑓𝑚𝑒𝑟𝑔𝑒 is a hyperparameter that controls the trade-off between merging performance and computational latency. Both 𝑔𝑢𝑏 and 𝑓𝑚𝑒𝑟𝑔𝑒 are analyzed in Sec. 7.8, along with other key hyperparameters in LifeHD. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: LifeDH

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 4 PROBLEM DEFINITION Before diving into our method, we first rigorously formulate the unsupervised lifelong learning problem using streaming sources, driven by real-world IoT applications. Streaming Data. To represent continuously changing environment, we assume a well-known class-incremental model in lifelong learning, in which new classes emerge in a sequential manner [46]. We also allow data distribution shift within one class. This setting models a scenario in which a device is continuously sampling data while the surrounding environment may change implicitly over time, e.g., the self-driving vehicle as shown in Fig. 1. We require that all samples appear only once (i.e., single-pass streams). Learning Protocol. Our goal is to build a classification algorithm that maps X → Y. For evaluation, we use the common evaluation protocol in state-of-the-art lifelong learning works [13, 14, 54], in which we construct an iid dataset E = {(𝑋𝑘 , 𝑦𝑘 )} for periodic testing, by sampling labeled examples from each class in a manner that preserves the overall (im)balance between the classes. Note, that even when one class has not appeared in the training data stream, it is always included in E. Hence E is a global view of all classes that can potentially exist in the environment. Unsupervised Clustering Accuracy. Since we do not give class labels or the total number of classes during training, the predicted label can be different from the ground-truth label. Therefore, for evaluation metric, we cannot adopt the simple prediction accuracy that requires exact label matching. Instead, we employ a widely used clustering metric known as unsupervised clustering accuracy (ACC) [63], which mirrors the conventional accuracy evaluation but within an unsupervised context. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 4 PROBLEM DEFINITION Before diving into our method, we first rigorously formulate the unsupervised lifelong learning problem using streaming sources, driven by real-world IoT applications. Streaming Data. To represent continuously changing environment, we assume a well-known class-incremental model in lifelong learning, in which new classes emerge in a sequential manner [46]. We also allow data distribution shift within one class. This setting models a scenario in which a device is continuously sampling data while the surrounding environment may change implicitly over time, e.g., the self-driving vehicle as shown in Fig. 1. We require that all samples appear only once (i.e., single-pass streams). Streaming Data. Learning Protocol. Our goal is to build a classification algorithm that maps X → Y. For evaluation, we use the common evaluation protocol in state-of-the-art lifelong learning works [13, 14, 54], in which we construct an iid dataset E = {(𝑋𝑘 , 𝑦𝑘 )} for periodic testing, by sampling labeled examples from each class in a manner that preserves the overall (im)balance between the classes. Note, that even when one class has not appeared in the training data stream, it is always included in E. Hence E is a global view of all classes that can potentially exist in the environment. Learning Protocol. Unsupervised Clustering Accuracy. Since we do not give class labels or the total number of classes during training, the predicted label can be different from the ground-truth label. Therefore, for evaluation metric, we cannot adopt the simple prediction accuracy that requires exact label matching. Instead, we employ a widely used clustering metric known as unsupervised clustering accuracy (ACC) [63], which mirrors the conventional accuracy evaluation but within an unsupervised context. Unsupervised Clustering Accuracy. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Problem Definition

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 3 BACKGROUND ON HDC Hyperdimensional Computing (HDC) is an emerging paradigm for information processing from the cognitive-neuroscience literature [24]. In HDC, all computation is performed on low-precision and distributed representations of data that accord naturally with highly parallel and low-energy hardware. The encoding function 𝜙 : X → H embeds data from its ambient representation into HD-space. In general, encoding should preserve some meaningful notion of similarity between input points in the sense that 𝜙 (𝑥) · 𝜙 (𝑥 ′ ) ≈ 𝑘 (𝑥, 𝑥′ ), where 𝑘 is some similarity function of interest on X. In this paper, we use spatiotemporal encoding for time series sensor data, and HDnn for more complex data, such as images, which we explain in the following. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 3 BACKGROUND ON HDC Hyperdimensional Computing (HDC) is an emerging paradigm for information processing from the cognitive-neuroscience literature [24]. In HDC, all computation is performed on low-precision and distributed representations of data that accord naturally with highly parallel and low-energy hardware. The encoding function 𝜙 : X → H embeds data from its ambient representation into HD-space. In general, encoding should preserve some meaningful notion of similarity between input points in the sense that 𝜙 (𝑥) · 𝜙 (𝑥 ′ ) ≈ 𝑘 (𝑥, 𝑥′ ), where 𝑘 is some similarity function of interest on X. In this paper, we use spatiotemporal encoding for time series sensor data, and HDnn for more complex data, such as images, which we explain in the following. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Background on HDC

Authors:
(1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu);
(2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu);
(3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx);
(4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu);
(5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction 2 Related Work 3 Background on HDC 4 Problem Definition 5 LifeDH 6 Variants of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 2 RELATED WORK Lifelong and On-Device Learning. Lifelong learning (or continual learning) is a large and active area of research in the broader machine learning community. Catastrophic forgetting is a major challenge in lifelong learning, and refers to a commonly observed empirical phenomenon in which updating certain machine learning models with new data severely degrades their ability to perform previously learned tasks [36]. Previous works proposed techniques such as dynamic architecture [31, 49], regularization by penalizing important weights [28, 66], knowledge distillation from past models [14] and experience replay using a memory buffer [35, 58]. The lifelong learning literature has examined a wide range of problem settings, ranging from the fully supervised case, in which tasks and class labels are provided, and the fully unsupervised case without any labels and prior knowledge [13, 57]. However, all of these works are based on deep NNs and require backpropagation, which is problematic for resource-constrained devices. Neurally-inspired lightweight algorithms have recently been proposed for lifelong learning applications. FlyModel [52] and SDMLP [8] use sparse coding and associative memory for lifelong learning. However, both approaches assume full supervision. STAM [54] is an expandable memory architecture for unsupervised lifelong learning, using layered receptive fields and a two-tier memory hierarchy. It learns via online centroid-based clustering pipeline, novelty detection and memory updates. Nevertheless, the memory in STAM is solely dedicated to image storage, while our LifeHD additionally emphasizes merging past patterns into coarse groups and shows more effective learning performance. Recent works optimize the resource usage of on-device training via pruning and quantization [34, 45], tuning partial weights [9, 47], memory profiling and optimization [15, 61, 64], as well as growing the NN on the fly [67]. All these works optimize training given resource constraints and do not focus on lifelong learning. They are orthogonal to the contribution of LifeHD which focuses on adaptive and continual training. LifeHD can be further optimized by combining with such techniques. Hyperdimensional Computing. HDC has garnered substantial interest from the computer hardware community as an energyefficient and low-latency approach to learning, and has been successfully applied to problems such as human activity recognition [27], voice recognition [23], image recognition [11, 65], to name a few. The large majority of literature on HDC has focused on using the technique to perform supervised classification tasks. Among the limited literature for weakly-supervised learning with HDC, HDCluster [21] enabled unsupervised clustering in HDC with a new algorithm that is similar to K-Means. SemiHD [22] is a semi-supervised learning framework using HDC with iterative self-labeling. Hyperseed [41], C-FSCIL [18] and FSL-HD [65] adopted HDC or similar vector symbolic architectures (VSA) for unsupervised or few-shot learning. All above works did not consider the lifelong aspect and used offline training on a static dataset. To the best of the authors’ knowledge, LifeHD is the first work that designs and deploys lifelong learning in edge IoT applications especially with zero or minimal amount of labels. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Authors: Authors: (1) Xiaofan Yu, University of California San Diego, La Jolla, California, USA (x1yu@ucsd.edu); (2) Anthony Thomas, University of California San Diego, La Jolla, California, USA (ahthomas@ucsd.edu); (3) Ivannia Gomez Moreno, CETYS University, Campus Tijuana, Tijuana, Mexico (ivannia.gomez@cetys.edu.mx); (4) Louis Gutierrez, University of California San Diego, La Jolla, California, USA (l8gutierrez@ucsd.edu); (5) Tajana Šimunić Rosing, University of California San Diego, La Jolla, USA (tajana@ucsd.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Related Work 2 Related Work 3 Background on HDC 3 Background on HDC 4 Problem Definition 4 Problem Definition 5 LifeDH 5 LifeDH 6 Variants of LifeHD 6 Variants of LifeHD 7 Evaluation of LifeHD 7 Evaluation of LifeHD 8 Evaluation of LifeHD semi and LifeHDa 8 Evaluation of LifeHD semi and LifeHDa 9 Discussions and Future Works 9 Discussions and Future Works 10 Conclusion, Acknowledgments, and References 10 Conclusion, Acknowledgments, and References 2 RELATED WORK 2 RELATED WORK Lifelong and On-Device Learning. Lifelong learning (or continual learning) is a large and active area of research in the broader machine learning community. Catastrophic forgetting is a major challenge in lifelong learning, and refers to a commonly observed empirical phenomenon in which updating certain machine learning models with new data severely degrades their ability to perform previously learned tasks [36]. Previous works proposed techniques such as dynamic architecture [31, 49], regularization by penalizing important weights [28, 66], knowledge distillation from past models [14] and experience replay using a memory buffer [35, 58]. The lifelong learning literature has examined a wide range of problem settings, ranging from the fully supervised case, in which tasks and class labels are provided, and the fully unsupervised case without any labels and prior knowledge [13, 57]. However, all of these works are based on deep NNs and require backpropagation, which is problematic for resource-constrained devices. Lifelong and On-Device Learning. Neurally-inspired lightweight algorithms have recently been proposed for lifelong learning applications. FlyModel [52] and SDMLP [8] use sparse coding and associative memory for lifelong learning. However, both approaches assume full supervision. STAM [54] is an expandable memory architecture for unsupervised lifelong learning, using layered receptive fields and a two-tier memory hierarchy. It learns via online centroid-based clustering pipeline, novelty detection and memory updates. Nevertheless, the memory in STAM is solely dedicated to image storage, while our LifeHD additionally emphasizes merging past patterns into coarse groups and shows more effective learning performance. Recent works optimize the resource usage of on-device training via pruning and quantization [34, 45], tuning partial weights [9, 47], memory profiling and optimization [15, 61, 64], as well as growing the NN on the fly [67]. All these works optimize training given resource constraints and do not focus on lifelong learning. They are orthogonal to the contribution of LifeHD which focuses on adaptive and continual training. LifeHD can be further optimized by combining with such techniques. Hyperdimensional Computing. HDC has garnered substantial interest from the computer hardware community as an energyefficient and low-latency approach to learning, and has been successfully applied to problems such as human activity recognition [27], voice recognition [23], image recognition [11, 65], to name a few. The large majority of literature on HDC has focused on using the technique to perform supervised classification tasks. Among the limited literature for weakly-supervised learning with HDC, HDCluster [21] enabled unsupervised clustering in HDC with a new algorithm that is similar to K-Means. SemiHD [22] is a semi-supervised learning framework using HDC with iterative self-labeling. Hyperseed [41], C-FSCIL [18] and FSL-HD [65] adopted HDC or similar vector symbolic architectures (VSA) for unsupervised or few-shot learning. All above works did not consider the lifelong aspect and used offline training on a static dataset. To the best of the authors’ knowledge, LifeHD is the first work that designs and deploys lifelong learning in edge IoT applications especially with zero or minimal amount of labels. Hyperdimensional Computing. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Related Work

Lifelong Intelligence Beyond the Edge using Hyperdimensional Computing: Abstract and Introduction

The first qualifier round of the FutureTech World Cup was held on July 10, 2024, at the Dubai AI Campus


10 startups pitched their groundbreaking ideas to an esteemed panel of five jury members


Momint, Nettle, s.r.o., and Orbii AI qualify for the grand finale set to take place during the Dubai AI & Web3 Festival 2024. The first Qualifier round of FutureTech World Cup, one of the new initiatives launched by Dubai AI & Web3 Festival & Trescon to recognise and support the most promising emerging tech start-ups from around the world, was hosted successfully by the Dubai AI Campus on July 10, 2024. Ten pre-qualified global startups pitched their groundbreaking ideas to a panel of five esteemed jury members with three startups - Momint, Nettle and Orbii - advancing to the grand finale set to be held during the Dubai AI & Web3 Festival on September 11 & 12, 2024. Momint: Recognised for its innovative approach and impactful solutions for accelerating clean energy.


Nettle, s.r.o.: Praised for its goal of redefining customer engagement with AI-Powered Digital Humans for the omnichannel age


Orbii AI: Honoured for its model of harnessing better data and intelligent AI models to help lenders make faster and more accurate credit decisions. "The competition provides a platform for innovators to showcase ideas, gain exposure, connect with stakeholders, collaborate with peers, and access capital, while also driving advancements in AI, Web3, and global digital economies.”, said Naveen Bharadwaj, Group CEO, Trescon. The distinguished panel of jury for the first qualifier included: Hasan Fardan Al Fardan, CEO, Al Fardan Exchange
Ian Arden, General Partner, Mempool Ventures
Jonathan Doerr, Partner MENAP, Antler
Ahmed Refaie, CEO & Global CIO, G.P.J. Global Partners Investments
Sarah Abuagela, Partner & Head of Investment, Ceras Ventures Serial Angel Investor & Chairman of Trescon, Mohammed Saleem said, “The FutureTech World Cup is designed to celebrate and elevate those who dare to believe they can transform the world of AI and Web3. Our goal is to support competent startups and founders from around the world with access to connections and investments that will help scale their businesses while ensuring they are solving real-world challenges”. All start-ups qualifying to the grand finale are offered fully sponsored Start-up Pods at the Dubai AI & Web3 Festival and the penultimate winner receives fully sponsored one-year Dubai AI Campus License, Trescon credits worth USD 10,000 to be used for future events, access to mentorship by the Jury members, and networking opportunities with the industry’s finest, in addition to the coveted title of ‘FutureTech World Cup Champion’, which will help elevate their start-up’s overall journey. The second qualifier round is expected to take place in August and the pre-qualification has already begun. Interested startups can apply for free on https://dubaiaiweb3festival.com/future-tech-world-cup/ About FutureTech World Cup The FutureTech World Cup marks a pioneering endeavour to reshape the world of AI & Web3. This event kicks off with a qualifier round, spotlighting start-ups from around the globe. Held in partnership with leading organisations, these competitions provide a launching pad for revolutionary AI and Web3 concepts. Standout startups advance to the grand finale hosted at the Dubai AI & Web3 Festival. Here, global investors, venture capitalists, angel investors, and the finest minds in AI and Web3 gather to witness start-ups pitch groundbreaking ideas and compete for multiple investment funds. More than just a competition, the FutureTech World Cup cultivates connections, propels innovation, and fuels the future of AI and Web3. Join us at the Dubai AI & Web3 Festival to inspire, challenge, and connect with the future of technology. Website: FutureTech World Cup Shadi Dawi Director of PR, Media Partnerships, & Communications Global & MEA shadi@tresconglobal.com +971 55 498 4989 Join us and be a part of the technological revolution that is shaping our future. This story was distributed as a release by ZEX MEDIA under HackerNoon’s Brand As An Author Program. Learn more about the program here. The first qualifier round of the FutureTech World Cup was held on July 10, 2024, at the Dubai AI Campus 10 startups pitched their groundbreaking ideas to an esteemed panel of five jury members Momint, Nettle, s.r.o., and Orbii AI qualify for the grand finale set to take place during the Dubai AI & Web3 Festival 2024. The first qualifier round of the FutureTech World Cup was held on July 10, 2024, at the Dubai AI Campus The first qualifier round of the FutureTech World Cup was held on July 10, 2024, at the Dubai AI Campus The first qualifier round of the FutureTech World Cup was held on July 10, 2024, at the Dubai AI Campus 10 startups pitched their groundbreaking ideas to an esteemed panel of five jury members 10 startups pitched their groundbreaking ideas to an esteemed panel of five jury members 10 startups pitched their groundbreaking ideas to an esteemed panel of five jury members Momint, Nettle, s.r.o., and Orbii AI qualify for the grand finale set to take place during the Dubai AI & Web3 Festival 2024. Momint, Nettle, s.r.o., and Orbii AI qualify for the grand finale set to take place during the Dubai AI & Web3 Festival 2024. Momint, Nettle, s.r.o., and Orbii AI qualify for the grand finale set to take place during the Dubai AI & Web3 Festival 2024. The first Qualifier round of FutureTech World Cup, one of the new initiatives launched by Dubai AI & Web3 Festival & Trescon to recognise and support the most promising emerging tech start-ups from around the world, was hosted successfully by the Dubai AI Campus on July 10, 2024. Ten pre-qualified global startups pitched their groundbreaking ideas to a panel of five esteemed jury members with three startups - Momint, Nettle and Orbii - advancing to the grand finale set to be held during the Dubai AI & Web3 Festival on September 11 & 12, 2024. Momint: Recognised for its innovative approach and impactful solutions for accelerating clean energy. Nettle, s.r.o.: Praised for its goal of redefining customer engagement with AI-Powered Digital Humans for the omnichannel age Orbii AI: Honoured for its model of harnessing better data and intelligent AI models to help lenders make faster and more accurate credit decisions. Momint: Recognised for its innovative approach and impactful solutions for accelerating clean energy. Momint : Recognised for its innovative approach and impactful solutions for accelerating clean energy. Momint Nettle, s.r.o.: Praised for its goal of redefining customer engagement with AI-Powered Digital Humans for the omnichannel age Nettle, s.r.o. : Praised for its goal of redefining customer engagement with AI-Powered Digital Humans for the omnichannel age Nettle, s.r.o. Orbii AI: Honoured for its model of harnessing better data and intelligent AI models to help lenders make faster and more accurate credit decisions. Orbii AI : Honoured for its model of harnessing better data and intelligent AI models to help lenders make faster and more accurate credit decisions. Orbii AI "The competition provides a platform for innovators to showcase ideas, gain exposure, connect with stakeholders, collaborate with peers, and access capital, while also driving advancements in AI, Web3, and global digital economies.”, said Naveen Bharadwaj, Group CEO, Trescon. "The competition provides a platform for innovators to showcase ideas, gain exposure, connect with stakeholders, collaborate with peers, and access capital, while also driving advancements in AI, Web3, and global digital economies.”, said Naveen Bharadwaj, Group CEO, Trescon. The distinguished panel of jury for the first qualifier included: Hasan Fardan Al Fardan, CEO, Al Fardan Exchange Ian Arden, General Partner, Mempool Ventures Jonathan Doerr, Partner MENAP, Antler Ahmed Refaie, CEO & Global CIO, G.P.J. Global Partners Investments Sarah Abuagela, Partner & Head of Investment, Ceras Ventures Hasan Fardan Al Fardan , CEO, Al Fardan Exchange Hasan Fardan Al Fardan CEO, Al Fardan Exchange Ian Arden , General Partner, Mempool Ventures Ian Arden General Partner, Mempool Ventures Jonathan Doerr , Partner MENAP, Antler Jonathan Doerr Partner MENAP, Antler Ahmed Refaie , CEO & Global CIO, G.P.J. Global Partners Investments Ahmed Refaie CEO & Global CIO, G.P.J. Global Partners Investments Sarah Abuagela , Partner & Head of Investment, Ceras Ventures Sarah Abuagela Partner & Head of Investment, Ceras Ventures Serial Angel Investor & Chairman of Trescon, Mohammed Saleem said, “The FutureTech World Cup is designed to celebrate and elevate those who dare to believe they can transform the world of AI and Web3. Our goal is to support competent startups and founders from around the world with access to connections and investments that will help scale their businesses while ensuring they are solving real-world challenges”. Serial Angel Investor & Chairman of Trescon, Mohammed Saleem said, “The FutureTech World Cup is designed to celebrate and elevate those who dare to believe they can transform the world of AI and Web3. Our goal is to support competent startups and founders from around the world with access to connections and investments that will help scale their businesses while ensuring they are solving real-world challenges”. All start-ups qualifying to the grand finale are offered fully sponsored Start-up Pods at the Dubai AI & Web3 Festival and the penultimate winner receives fully sponsored one-year Dubai AI Campus License, Trescon credits worth USD 10,000 to be used for future events, access to mentorship by the Jury members, and networking opportunities with the industry’s finest, in addition to the coveted title of ‘FutureTech World Cup Champion’, which will help elevate their start-up’s overall journey. The second qualifier round is expected to take place in August and the pre-qualification has already begun. Interested startups can apply for free on https://dubaiaiweb3festival.com/future-tech-world-cup/ https://dubaiaiweb3festival.com/future-tech-world-cup/ About FutureTech World Cup About FutureTech World Cup The FutureTech World Cup marks a pioneering endeavour to reshape the world of AI & Web3. This event kicks off with a qualifier round, spotlighting start-ups from around the globe. Held in partnership with leading organisations, these competitions provide a launching pad for revolutionary AI and Web3 concepts. Standout startups advance to the grand finale hosted at the Dubai AI & Web3 Festival. Here, global investors, venture capitalists, angel investors, and the finest minds in AI and Web3 gather to witness start-ups pitch groundbreaking ideas and compete for multiple investment funds. More than just a competition, the FutureTech World Cup cultivates connections, propels innovation, and fuels the future of AI and Web3. Join us at the Dubai AI & Web3 Festival to inspire, challenge, and connect with the future of technology. Website: FutureTech World Cup FutureTech World Cup Shadi Dawi Shadi Dawi Director of PR, Media Partnerships, & Communications Global & MEA shadi@tresconglobal.com +971 55 498 4989 Join us and be a part of the technological revolution that is shaping our future. This story was distributed as a release by ZEX MEDIA under HackerNoon’s Brand As An Author Program. Learn more about the program here. This story was distributed as a release by ZEX MEDIA under HackerNoon’s Brand As An Author Program. Learn more about the program here. This story was distributed as a release by ZEX MEDIA under HackerNoon’s Brand As An Author Program. Learn more about the program here. This story was distributed as a release by ZEX MEDIA under HackerNoon’s Brand As An Author Program. Learn more about the program here . here

Dubai AI and Web3 Festival and Trescon Launch FutureTech World Cup to Support Global Start-ups

দুবাই এআই এবং ওয়েব 3 ফেস্টিভ্যাল এবং ট্রেসকন গ্লোবাল স্টার্ট-আপগুলিকে সমর্থন করার জন্য ফিউচারটেক ওয়ার্ল্ড কাপ চালু করেছে

Dubai AI und Web3 Festival sowie Trescon veranstalten FutureTech World Cup zur Unterstützung globaler Start-ups

Dubai AI y Web3 Festival y Trescon lanzan la Copa Mundial FutureTech para apoyar a las empresas emergentes globales

Le Dubai AI and Web3 Festival et Trescon lancent la FutureTech World Cup pour soutenir les start-ups mondiales

दुबई एआई और वेब3 फेस्टिवल और ट्रेसकॉन ने वैश्विक स्टार्ट-अप को समर्थन देने के लिए फ्यूचरटेक वर्ल्ड कप लॉन्च किया

ドバイAI＆Web3フェスティバルとトレスコンが世界のスタートアップを支援するフューチャーテックワールドカップを開催

두바이 AI 및 Web3 페스티벌과 Trescon, 글로벌 스타트업 지원 위해 FutureTech 월드컵 개최

Dubai AI e Web3 Festival e Trescon lançam FutureTech World Cup para apoiar startups globais

Дубайский фестиваль искусственного интеллекта и Web3 и компания Trescon запускают чемпионат мира по технологиям FutureTech для поддержки глобальных стартапов

Dubai AI ve Web3 Festivali ve Trescon, Küresel Startup'ları Desteklemek İçin FutureTech Dünya Kupasını Başlatıyor

Lễ hội Dubai AI và Web3 và Trescon ra mắt FutureTech World Cup để hỗ trợ các công ty khởi nghiệp toàn cầu

迪拜 AI 和 Web3 节与 Trescon 联合推出未来科技世界杯，助力全球初创企业

Axion Processors: Google's First Arm-based CPUs

A biography generated by an AI? Indeed. So I decided I’d write my biography, or a sort of an edited autobiography of some particular episodes. Then while on a break from scribing away I ran up on this article about an AI that generates “artificial” web sites based upon a verbal prompt. This page is the intro to the site and you can use it to create your own web pages. A final link is this more or less explanatory page about the synthesizer. So I decided I’d ask it to write a biographical piece for me. Here then is the AI generated biography: Adventures in Nerdspace: A Geek's Synthetic Biography Greetings, fellow traveler of the cybersphere! Welcome to a glimpse into the life and times of a proud geek, synthesized for your digital delectation. Strap on your VR headset, plug in your neurolink, and jack into this thrilling tale of nerdly exploits across the meatspace/cyberspace divide. Chapter 1: Birth of a Nerd Born with a keyboard in one hand and a polyhedral die in the other, it was clear from an early age that I was destined for geekdom. While other babies were content chewing on rattles and blowing raspberries, I could be found attempting to dismantle my crib to build a rudimentary 8-bit computer from the parts. My first word was "404" and my first crush was on Ada Lovelace. Chapter 2: Nerd School Daze Ah, school - that venerable institution of standardized mediocrity. While my peers were busy trying to climb the social hierarchy, I was more interested in climbing the high score tables of classic arcade games. Lunch breaks were spent huddled around a Game Boy, trading Pokémon and debating Kirk vs Picard. Prom night? More like Lan party night! Chapter 3: Code, Hack, Sleep, Repeat University was where I truly blossomed into a world-class nerd. Fueled by ramen and Jolt cola, I spent countless caffeine-soaked nights lost in the glorious world of code, hacking everything from toasters to the campus mainframe. My dorm room became a sanctuary of geekiness, where soldering irons outnumbered textbooks and arguments over Linux distros raged into the wee hours. Chapter 4: The Geek Inherits the Earth In the real world, it turns out that being an unbridled nerd is actually a superpower. Armed with my elite tech skills and utterly un-self-conscious enthusiasm, I set out to disrupt industry after industry with my quirky yet brilliant innovations. Did I take over the world? Not quite. But I did amass a sizable collection of Funko Pops and a dedicated following on Github! Chapter 5: To Infinity and Beyond! And so the adventure continues, one line of code and obscure pop-culture reference at a time. From the depths of the darknet to the halls of Comic-Con, this geek's synthetic life story is still being written. Will I eventually transcend meatspace to become a digital demigod living in the cloud? Will I find love with a cyborg soulmate? Will I finally get a high score on Donkey Kong? Tune in next time to find out! Same nerd-time, same nerd-channel. The End...? Here is the biography presented as screen captures. Return to the Geek-Space Continuum link shown in the page above. Hopefully you found this entertaining and informative. As always Comments, Criticisms, and Suggestions are welcome. God bless all. A biography generated by an AI? Indeed. So I decided I’d write my biography, or a sort of an edited autobiography of some particular episodes. Then while on a break from scribing away I ran up on this article about an AI that generates “artificial” web sites based upon a verbal prompt. This page is the intro to the site and you can use it to create your own web pages. A final link is this more or less explanatory page about the synthesizer. So I decided I’d ask it to write a biographical piece for me. article article page page explanatory page explanatory page Here then is the AI generated biography: Adventures in Nerdspace: A Geek's Synthetic Biography Greetings, fellow traveler of the cybersphere! Welcome to a glimpse into the life and times of a proud geek, synthesized for your digital delectation. Strap on your VR headset, plug in your neurolink, and jack into this thrilling tale of nerdly exploits across the meatspace/cyberspace divide. Chapter 1: Birth of a Nerd Born with a keyboard in one hand and a polyhedral die in the other, it was clear from an early age that I was destined for geekdom. While other babies were content chewing on rattles and blowing raspberries, I could be found attempting to dismantle my crib to build a rudimentary 8-bit computer from the parts. My first word was "404" and my first crush was on Ada Lovelace. Chapter 2: Nerd School Daze Ah, school - that venerable institution of standardized mediocrity. While my peers were busy trying to climb the social hierarchy, I was more interested in climbing the high score tables of classic arcade games. Lunch breaks were spent huddled around a Game Boy, trading Pokémon and debating Kirk vs Picard. Prom night? More like Lan party night! Chapter 3: Code, Hack, Sleep, Repeat University was where I truly blossomed into a world-class nerd. Fueled by ramen and Jolt cola, I spent countless caffeine-soaked nights lost in the glorious world of code, hacking everything from toasters to the campus mainframe. My dorm room became a sanctuary of geekiness, where soldering irons outnumbered textbooks and arguments over Linux distros raged into the wee hours. Chapter 4: The Geek Inherits the Earth In the real world, it turns out that being an unbridled nerd is actually a superpower. Armed with my elite tech skills and utterly un-self-conscious enthusiasm, I set out to disrupt industry after industry with my quirky yet brilliant innovations. Did I take over the world? Not quite. But I did amass a sizable collection of Funko Pops and a dedicated following on Github! Chapter 5: To Infinity and Beyond! And so the adventure continues, one line of code and obscure pop-culture reference at a time. From the depths of the darknet to the halls of Comic-Con, this geek's synthetic life story is still being written. Will I eventually transcend meatspace to become a digital demigod living in the cloud? Will I find love with a cyborg soulmate? Will I finally get a high score on Donkey Kong? Tune in next time to find out! Same nerd-time, same nerd-channel. The End...? Here is the biography presented as screen captures. Return to the Geek-Space Continuum link shown in the page above. Return to the Geek-Space Continuum Hopefully you found this entertaining and informative. As always Comments, Criticisms, and Suggestions are welcome. God bless all.

Stories, News & Mentions about CAMPUS

Campus’s Stories on HackerNoon