Unisyn

unisyn

Computing has changed in recent years, with the demand for high-performance computing (HPC) rising every single day. Many advancements were made, with most notable breakthroughs in the form of artificial intelligence and supercomputing. Meanwhile, Linux has created an ecosystem that is super adept at supercomputing efficiency, offering a robust and flexible platform for engineers. Without a doubt, the capabilities of the Linux system have changed the landscape of techdev with its innovations. Why Linux is the Operating System of Choice for Supercomputers There are many benefits of Linux and there is more than one reason why Linux is the first choice for many engineers and researchers. The platform it offers is highly scalable and flexible, including tools to optimize for HPC tasks, such as automatic vectorization in GCC (GNU Compiler Collection), OpenMP (Open Multi-Processing), and OpenMPI (Open Message Passing Interface), all of which aid in the optimization of HPC applications. Another benefit of Linux is that, regardless of programming language, Linux has a large set of capabilities to support them, which makes it much easier for developers to utilize the platform and implement solutions that are tailored to the applications as needed. The open-source nature of Linux encourages both growth and innovation by relying on the wisdom of the crowd - having hundreds of thousands of users all over the world contributing to its advancement. It also fosters a sense of community, and true enough, there is a vibrant community of developers that support one another’s research, collaborating in unison for the betterment of tech. In today’s landscape, 90% of the world’s most sophisticated supercomputers are powered by Linux. Moreover, for more than 20 years, Linux systems have filled more than half of the spots in the top 500 list of the most powerful machines in the world. Even back in 2018, all of them have become Linux powered, and, at that time, two new systems entering the list were IBM machines based on the POWER9 family of processors, with newly implemented supercomputer features designed by Gabriel Gomes, in the form of “compiler-toolchain software for high-precision, high-performance, floating-point arithmetic integrated into the core math library”, as he recalls. Now almost a decade later, Linux is still at the forefront, pushing the envelope for supercomputing and setting the standard for open-source platforms. Current HPC trends and cyber security Technology is constantly being upgraded. Developers and programmers never rest in their quest to find better solutions. With the integration of AI and machine learning, more doors have opened to researchers, and it is really up to the imagination how they apply the technology to create new and innovative solutions. In current trends, libraries like TensorFlow and PyTorch, which are optimized for Linux environments, are taking over because they are also open-sourced and rely on developers who work on underlying optimizations that enhance the performance of these machine learning algorithms. Optimization is not the only concern for HPC, though. The large amounts of data it consumes, as well as the A.I. models produced by the processing of this data are highly valuable assets and thus an increasingly common target for cybercrime and cyber security. Linux is again well positioned in this field, since having so many users using the platform means that vulnerabilities are quickly picked up and dealt with. Independently of the technique used - be it code review, static or dynamic analysis, or even automated A.I. solutions - detecting vulnerable paths before threats have a chance to manifest themselves is the goal. “A proactive approach is always better, as the saying goes, prevention is better than cure”, explains Gomes, who has been working with tech since graduating from University of Campinas and has since worked with IT hotshots like IBM and Linux distributor, SUSE. According to him, “It is important to continue fortifying the security of supercomputing systems for obvious reasons, particularly for maintaining uptime and preventing data breaches.” In this scenario, live patching is a new and valuable feature for supercomputing as it doesn’t interfere with uptime. During his time at SUSE, Gomes led the development of libpulp, a framework that allows the live patching of user-space processes, which can be used to safely apply changes to long-running computer programs. Since HPC requires performance, changes can also be reverted if their cost is deemed too high. This feature enhances the efficiency of patch management in production systems, and can be installed on any Linux distribution, explains Gomes, who is a Debian Developer in the Debian project. The Future of High-Performance Computing There is a growing reliance on supercomputing for sensitive research, which means that the performance and security of these systems will be the highest priority for future developments. The industry will continue to rely heavily on performance and security researchers that are able to focus on operating systems, compilers, libraries and patching mechanisms development for HPC. There will be increased collaboration across HPC domains, bringing together A.I., machine learning, and big data analytics. This also means that resource management will play a crucial role, as supercomputing consumes more resources than traditional computers. The development of sophisticated scheduling algorithms and workload management tools within the Linux ecosystem will further enhance resource availability and enable systems to hit peak performances. A driving force in the realm of high-performance computing, the Linux ecosystem will continue to excel and provide the technology needed to achieve greater efficiencies and capabilities, pushing forward the boundaries of what is possible in the future of computing. Computing has changed in recent years, with the demand for high-performance computing (HPC) rising every single day. Many advancements were made, with most notable breakthroughs in the form of artificial intelligence and supercomputing. Meanwhile, Linux has created an ecosystem that is super adept at supercomputing efficiency, offering a robust and flexible platform for engineers. Without a doubt, the capabilities of the Linux system have changed the landscape of techdev with its innovations. Why Linux is the Operating System of Choice for Supercomputers Why Linux is the Operating System of Choice for Supercomputers There are many benefits of Linux and there is more than one reason why Linux is the first choice for many engineers and researchers. The platform it offers is highly scalable and flexible, including tools to optimize for HPC tasks, such as automatic vectorization in GCC (GNU Compiler Collection), OpenMP (Open Multi-Processing), and OpenMPI (Open Message Passing Interface), all of which aid in the optimization of HPC applications . HPC applications HPC applications Another benefit of Linux is that, regardless of programming language, Linux has a large set of capabilities to support them, which makes it much easier for developers to utilize the platform and implement solutions that are tailored to the applications as needed. The open-source nature of Linux encourages both growth and innovation by relying on the wisdom of the crowd - having hundreds of thousands of users all over the world contributing to its advancement. It also fosters a sense of community, and true enough, there is a vibrant community of developers that support one another’s research, collaborating in unison for the betterment of tech. In today’s landscape, 90% of the world’s most sophisticated supercomputers are powered by Linux. Moreover, for more than 20 years, Linux systems have filled more than half of the spots in the top 500 list of the most powerful machines in the world. Even back in 2018, all of them have become Linux powered, and, at that time, two new systems entering the list were IBM machines based on the POWER9 family of processors, with newly implemented supercomputer features designed by Gabriel Gomes, in the form of “compiler-toolchain software for high-precision, high-performance, floating-point arithmetic integrated into the core math library”, as he recalls. Now almost a decade later, Linux is still at the forefront, pushing the envelope for supercomputing and setting the standard for open-source platforms. 90% of the world’s most sophisticated supercomputers top 500 list top 500 list Current HPC trends and cyber security Technology is constantly being upgraded. Developers and programmers never rest in their quest to find better solutions. With the integration of AI and machine learning, more doors have opened to researchers, and it is really up to the imagination how they apply the technology to create new and innovative solutions. In current trends, libraries like TensorFlow and PyTorch, which are optimized for Linux environments, are taking over because they are also open-sourced and rely on developers who work on underlying optimizations that enhance the performance of these machine learning algorithms. Optimization is not the only concern for HPC, though. The large amounts of data it consumes, as well as the A.I. models produced by the processing of this data are highly valuable assets and thus an increasingly common target for cybercrime and cyber security. Linux is again well positioned in this field, since having so many users using the platform means that vulnerabilities are quickly picked up and dealt with. Independently of the technique used - be it code review, static or dynamic analysis, or even automated A.I. solutions - detecting vulnerable paths before threats have a chance to manifest themselves is the goal. “A proactive approach is always better, as the saying goes, prevention is better than cure”, explains Gomes, who has been working with tech since graduating from University of Campinas and has since worked with IT hotshots like IBM and Linux distributor, SUSE. According to him, “It is important to continue fortifying the security of supercomputing systems for obvious reasons, particularly for maintaining uptime and preventing data breaches.” In this scenario, live patching is a new and valuable feature for supercomputing as it doesn’t interfere with uptime. During his time at SUSE, Gomes led the development of libpulp, a framework that allows the live patching of user-space processes, which can be used to safely apply changes to long-running computer programs. Since HPC requires performance, changes can also be reverted if their cost is deemed too high. This feature enhances the efficiency of patch management in production systems, and can be installed on any Linux distribution, explains Gomes, who is a Debian Developer in the Debian project. The Future of High-Performance Computing There is a growing reliance on supercomputing for sensitive research, which means that the performance and security of these systems will be the highest priority for future developments. The industry will continue to rely heavily on performance and security researchers that are able to focus on operating systems, compilers, libraries and patching mechanisms development for HPC. security of these systems security of these systems There will be increased collaboration across HPC domains, bringing together A.I., machine learning, and big data analytics. This also means that resource management will play a crucial role, as supercomputing consumes more resources than traditional computers. The development of sophisticated scheduling algorithms and workload management tools within the Linux ecosystem will further enhance resource availability and enable systems to hit peak performances. A driving force in the realm of high-performance computing, the Linux ecosystem will continue to excel and provide the technology needed to achieve greater efficiencies and capabilities, pushing forward the boundaries of what is possible in the future of computing.

Why Linux Leads the Charge in High-Performance Computing

The Case for Software Supply Chain Security. A couple of days ago, on Oct 22, 2024, the Securities and Exchange Commission fined millions of dollars for mislabeling the risks of the SolarWinds attack as “hypothetical.” Four companies Unisys, Avaya, Check Point, and Mimecast were fined $7 million. In 2020, the SolarWinds attack affected major organizations when malware gave backdoor access to its Orion software. Supply chain attacks have only grown in both quantity and magnitude since then. Though companies continue to downplay the impact of Supply Chain security, on average, less than 50% of the code base in your organization is written by your employees. Supply chain security controls mitigate risks coming from the “external” software that ships with your product. The Cure to Make Robust Software Supply Chains. Google has invested heavily in insider risk since 2010 and has been building multi-layered security controls to defend against advanced attacks. Around 2019/20, Google Chrome started developing a framework that applies to a broader community, starting with Chromium. Collaboration across the company led to the formation of the supply chain levels for software artifacts aka SLSA. Since then, Google has donated this project to OpenSSF, one of the largest open-source software foundations. SLSA version 1.0 was launched earlier this year and it provides an incremental solution to gradually improve security posture from development to deployment for the software supply chain. SLSA Threat Model Let’s take a look into how malicious actors exploit weaknesses throughout the process right from build systems to the artifact distribution level. The SLSA threat model states that issues like inadequate code review, misconfigured build systems, and unverified third-party code-introduced risks, etc. can cause significant security breaches. The successful attacks on SolarWinds where the hackers injected malicious code into the build process require strong controls. Knowing your dependencies and preventing unilateral access remains crucial for ensuring malware can’t be inserted. Applicability of SLSA in Real World Scenarios There are three major use cases for SLSA, First Party, Open Source Projects, and Vendors. In First Party, since there is no need to transfer trust across organizational boundaries, this makes it the easiest use case. SLSA can be used entirely within an organization. In the case of Open Source Projects, SLSA uses a canonical source repository approach where users of open-source software need to only trust a handful of build platforms instead of trusting thousands of developers who have upload permissions to many packages. SLSA would simply reject an upload if it was not from a canonical source repository. And finally, for users of vendor-provided software, SLSA would require a vendor to be certified by a third-party auditor. This would make it easier for consumers to trust vendor software products. Also, provenance is the main aspect of using SLSA, it would provide traceability and verifiability of where the software originated. The goal is not just to prevent tampering but also to trace who was involved and where the exact origin of the software was. Since it facilitates transparency, companies can easily adapt to regulatory restrictions while ensuring that their business operations are sound. Hardening the Chain of Trust The core of SLSA is provenance, which documents the origin and security of artifacts. SLSA can pinpoint tampering or any other malicious behavior using cryptographic validation and automated attestation. Since it verifies that each step from creation to distribution of the software is indeed secure, the chain of trust is therefore solidified. The majority of well-secured development environments apply the principles of provenance including Google. One of the innovators of the SLSA framework at Google is Akash Mukherjee. He discusses how Chrome ensures verifiability in its builds in this article Chromium Chronicle #23. The article focuses on the fundamentals of the framework and the four pillars that we see today in SLSA. Meeting Government Mandates With SLSA Another advantage of SLSA compared to other security libraries is its ability to meet U.S. government regulations. The National Institute Of Standards and Technology aka NIST plays a crucial role in cybersecurity by creating frameworks, guidelines, etc. like the Secure Software Development Framework (SSDF). SLSA is able to help companies meet these requirements. Following this framework will also align companies with the Executive Order 14028 directive which has a higher standard for cybersecurity along the entire software supply chain. Also in highly regulated defense, healthcare, and finance industries, SLSA is reliable enough to be the security blueprint for organizations that can help generate evidence of compliance even when it touches critical cybersecurity concerns. The Future of SLSA as a Blueprint for Secure Software Development But this journey does not stop here; it continues evolving the industry. With the flexibility in its structure, SLSA allows organizations to enhance their security step by step while preparing for future threats. As SBOMs become central to software development, experts focus on securing development practices in general, “SBOMs can help lift security health from the status quo, but frameworks like Supply-chain Levels for Software Artifacts (SLSA) and Secure Software Development Framework (SSDF) must be used to improve security as you’re building new packages.”, Akash Mukherjee, an early developer of the SLSA framework in his recent post on RSA conference. By integrating the best practices from SLSA with development insights gleaned from resources such as Chromium Chronicle, an organization is well-equipped to design secure software. At its core, SLSA revolves around provenance, providing cryptographic assurances and compliance with regulatory requirements, making it an extremely important tool for securing the global software ecosystem. The Case for Software Supply Chain Security. A couple of days ago, on Oct 22, 2024, the Securities and Exchange Commission fined millions of dollars for mislabeling the risks of the SolarWinds attack as “hypothetical.” Four companies Unisys, Avaya, Check Point, and Mimecast were fined $7 million. In 2020, the SolarWinds attack affected major organizations when malware gave backdoor access to its Orion software. Supply chain attacks have only grown in both quantity and magnitude since then. Though companies continue to downplay the impact of Supply Chain security, on average, less than 50% of the code base in your organization is written by your employees. Supply chain security controls mitigate risks coming from the “external” software that ships with your product. The Cure to Make Robust Software Supply Chains. The Cure to Make Robust Software Supply Chains. Google has invested heavily in insider risk since 2010 and has been building multi-layered security controls to defend against advanced attacks. Around 2019/20, Google Chrome started developing a framework that applies to a broader community, starting with Chromium. Collaboration across the company led to the formation of the supply chain levels for software artifacts aka SLSA. Since then, Google has donated this project to OpenSSF, one of the largest open-source software foundations. SLSA version 1.0 was launched earlier this year and it provides an incremental solution to gradually improve security posture from development to deployment for the software supply chain. SLSA version 1.0 SLSA version 1.0 SLSA Threat Model SLSA Threat Model Let’s take a look into how malicious actors exploit weaknesses throughout the process right from build systems to the artifact distribution level. The SLSA threat model states that issues like inadequate code review, misconfigured build systems, and unverified third-party code-introduced risks, etc. can cause significant security breaches. SLSA threat model SLSA threat model The successful attacks on SolarWinds where the hackers injected malicious code into the build process require strong controls. Knowing your dependencies and preventing unilateral access remains crucial for ensuring malware can’t be inserted. Applicability of SLSA in Real World Scenarios Applicability of SLSA in Real World Scenarios There are three major use cases for SLSA, First Party, Open Source Projects, and Vendors. In First Party, since there is no need to transfer trust across organizational boundaries, this makes it the easiest use case. SLSA can be used entirely within an organization. use cases use cases In the case of Open Source Projects, SLSA uses a canonical source repository approach where users of open-source software need to only trust a handful of build platforms instead of trusting thousands of developers who have upload permissions to many packages. SLSA would simply reject an upload if it was not from a canonical source repository. And finally, for users of vendor-provided software, SLSA would require a vendor to be certified by a third-party auditor. This would make it easier for consumers to trust vendor software products. Also, provenance is the main aspect of using SLSA, it would provide traceability and verifiability of where the software originated. The goal is not just to prevent tampering but also to trace who was involved and where the exact origin of the software was. Since it facilitates transparency, companies can easily adapt to regulatory restrictions while ensuring that their business operations are sound. Hardening the Chain of Trust Hardening the Chain of Trust The core of SLSA is provenance , which documents the origin and security of artifacts. SLSA can pinpoint tampering or any other malicious behavior using cryptographic validation and automated attestation. Since it verifies that each step from creation to distribution of the software is indeed secure, the chain of trust is therefore solidified. provenance provenance The majority of well-secured development environments apply the principles of provenance including Google. One of the innovators of the SLSA framework at Google is Akash Mukherjee. He discusses how Chrome ensures verifiability in its builds in this article Chromium Chronicle #23 . The article focuses on the fundamentals of the framework and the four pillars that we see today in SLSA. Chromium Chronicle #23 Chromium Chronicle #23 Meeting Government Mandates With SLSA Meeting Government Mandates With SLSA Another advantage of SLSA compared to other security libraries is its ability to meet U.S. government regulations. The National Institute Of Standards and Technology aka NIST plays a crucial role in cybersecurity by creating frameworks, guidelines, etc. like the Secure Software Development Framework ( SSDF ). SLSA is able to help companies meet these requirements. SSDF SSDF Following this framework will also align companies with the Executive Order 14028 directive which has a higher standard for cybersecurity along the entire software supply chain. Also in highly regulated defense, healthcare, and finance industries, SLSA is reliable enough to be the security blueprint for organizations that can help generate evidence of compliance even when it touches critical cybersecurity concerns. The Future of SLSA as a Blueprint for Secure Software Development The Future of SLSA as a Blueprint for Secure Software Development But this journey does not stop here; it continues evolving the industry. With the flexibility in its structure, SLSA allows organizations to enhance their security step by step while preparing for future threats. As SBOMs become central to software development, experts focus on securing development practices in general, “SBOMs can help lift security health from the status quo, but frameworks like Supply-chain Levels for Software Artifacts (SLSA) and Secure Software Development Framework (SSDF) must be used to improve security as you’re building new packages.” , Akash Mukherjee, an early developer of the SLSA framework in his recent post on RSA conference. “SBOMs can help lift security health from the status quo, but frameworks like Supply-chain Levels for Software Artifacts (SLSA) and Secure Software Development Framework (SSDF) must be used to improve security as you’re building new packages.” recent post recent post By integrating the best practices from SLSA with development insights gleaned from resources such as Chromium Chronicle, an organization is well-equipped to design secure software. At its core, SLSA revolves around provenance, providing cryptographic assurances and compliance with regulatory requirements, making it an extremely important tool for securing the global software ecosystem.

How SLSA Can Secure Your Software Supply Chain

It’s a real shame that I couldn’t get a ticket to one of the landmark events that solidified London’s status as a hub of concentrated creative energy. This year, those who harness this energy – the ones who still burn with the desire to make the world a better place – used it to raise public awareness on a wide range of issues. At the London Design Festival, themes of synergy between nature and humanity took center stage, embodied in the work of Marjan van Aubel. Hot on the heels of her installation, the industrial design team at Diez Office encouraged an unspoken dialogue about ecological responsibility with their project Vert. In general, the program featured numerous innovative solutions, breaking down the boundaries of stereotypical thinking. And we crave that kind of intellectual stimulation! In this piece, I’ll talk about the playful, mesmerizing, light-hearted and literally warming design on display this year! London and design go hand in hand. It is part of our story. London Design Festival is a platform for hundreds of design stories to be told. Each of them speaks to an expanding audience hungry for design ideas and enjoying the quality and diversity of what’s on offer. It all confirms London’s status as the global center of design – says Ben Evans CBE. It’s hard to disagree with the words of this passionate advocate and activist for all-encompassing creativity, which can be found in everything that surrounds us. Today, we’ll spot it in glass, but in a broader sense, we’ll explore the role materials play in design – and in parallels. Parallels not only between the red oak beams of the exhibit at Chelsea College of Arts but also between the architecture of play and real life. The Sun, My Heart The energy within us has the power to create incredible things, especially when we learn to recognize it and channel it in the right direction – one where we create something extraordinary that captivates the world. Energy is the central theme in the work of Dutch designer Marjan van Aubel. Her creations can hardly be described as merely "sunny" although, in reality, each one of them is filled with light. Without this element, we can no longer imagine her work. Energy, as an aesthetic element, has become an integral part of every object crafted by her hands: from glassware, which reflects the flow of matter, to entire installations. We, in turn, have become vessels filled with her revelation: They were transparent and colored. It was the first time I saw that solar energy could be beautiful. From that moment, I started implementing it into different objects: glassware, tables, light installations, buildings. Now, I see all surfaces as a potential to harvest energy Speaking of installations, we can now explore her piece “The Sun, My Heart” a highlight of the LDF collection. The title becomes somewhat rhetorical when you consider van Aubel’s relationship with the primary source of life, with roots that delve deep into Buddhist philosophy. In the language of Vietnamese monks, it calls on us to "open our hearts to new connections". These connections come to life through 77 self-charging Sunne lamps (some of which represent nothing more than rays of light), symbolizing the complete solar cycle, accompanied by harmoniously selected sounds that echo the sun's vibrations. Working in unison, this masterpiece immerses the viewer in a state of total relaxation, stimulating the brain's theta waves. We receive the sun in such abundance. It’s obvious that we need to transition to solar energy, but the current approach is too technical. We shouldn’t only talk about numbers and efficiency. The sun is something holistic; we can feel it. We need beauty and design to make us want to change things. We need to feel it inside – the designer shares. Pavilions of Wonder A fresh perspective with a historical undertone, embodying cultural heritage – the true American dream, brought to life in a trilogy of works by designer Nina Tolstrup. This project has become a symbol of vivid imagination and aesthetics, driven by a spirit of fun and playfulness. The installation consists of three objects, harking back to the time when Palm Springs became the true capital of modernism in the USA, owing to its ability to "grow rapidly" with affordable housing after World War II. This area, dotted with buildings of the corresponding architectural style, would later become a prototype of inspiration for future designers. Among those designers is Nina Tolstrup, who infused the project with the same playful spirit reflected in Barbie’s DreamHouses. It is through her vibrant qualities, captured like film on a reel, that we recognize the essence of the designer. We looked at the evolution of the DreamHouse, informed by Greater Palm Springs architecture. We examined the visual language and iconic elements common to both Vert And here are the very same "parallels" made from red oak, presented in the collaborative work of the American Hardwood Export Council (AHEC), the Munich-based Diez Office, and Frankfurt's Office for Microclimatic Cultivation (OMC°C). This project is far from a cry for help, but rather a message of mindful care for the environment. Vert, which signifies something "green" is an undeniably bold statement, offering a tangible and effective solution to a global issue. It represents a solution through a small, carefully planned, and meticulously thought-out noble endeavor. It can't be achieved just by planting trees or through individual façade and rooftop greening projects, so additional solutions are urgently needed. We need fairly large structures to generate the necessary amount of shade and biomass – explains Nicola Stattmann, co-founder of OMC°C. The architectural installation consists of several triangular structures, each standing 10 meters tall. These are all covered with plants such as morning glory, cheese plant, black-eyed Susan, and night-blooming moonflower, which will act as "killers" of carbon dioxide released into the atmosphere. Interestingly, the modularity of the designed structure allows for Vert to be produced in series in the future. This once again highlights the brilliance of the minds behind the project, brought to life with the support of AHEC, who have long been advocating for the use of durable engineered materials. Red oak glulam's natural beauty, combined with its structural capabilities, makes it a promising choice both functionally and aesthetically – says Stefan Diez, Diez Office founder. The London Design Festival 2024 masterfully demonstrated the power of design in tackling global issues – from sustainable energy solutions to supporting biodiversity and reimagining urban spaces. Through innovative projects, the festival highlighted the crucial role of creativity in shaping a more sustainable and harmonious future. As these projects blur the lines between functionality and beauty, they invite us to envision a world where technology, nature, and human ingenuity coexist seamlessly, encouraging a more thoughtful and responsible approach to life. The London Design Festival week was one of the most anticipated events for us. Our fields are closely intertwined – that much is clear. We keep track of the latest trends to make our 3D visualizations even more realistic, filled with the smallest details that reflect the quality of bringing the 'imagined' to life. We’re interested in everything: from the materials often used in designers' projects to the very concepts behind these works, because this knowledge embodies our commitment to staying current. We don't want to fall behind. We aim to build trust with every client interaction, ensuring that the result meets and exceeds their needs, leaving them fully satisfied – says Alexander Kasperovich, CEO of 3D visualization and rendering studio, Studio57. It’s a real shame that I couldn’t get a ticket to one of the landmark events that solidified London’s status as a hub of concentrated creative energy. This year, those who harness this energy – the ones who still burn with the desire to make the world a better place – used it to raise public awareness on a wide range of issues. It’s a real shame that I couldn’t get a ticket to one of the landmark events that solidified London’s status as a hub of concentrated creative energy. This year, those who harness this energy – the ones who still burn with the desire to make the world a better place – used it to raise public awareness on a wide range of issues. At the London Design Festival , themes of synergy between nature and humanity took center stage, embodied in the work of Marjan van Aubel. Hot on the heels of her installation, the industrial design team at Diez Office encouraged an unspoken dialogue about ecological responsibility with their project Vert. In general, the program featured numerous innovative solutions, breaking down the boundaries of stereotypical thinking. And we crave that kind of intellectual stimulation! In this piece, I’ll talk about the playful, mesmerizing, light-hearted and literally warming design on display this year! London Design Festival London Design Festival London and design go hand in hand. It is part of our story. London Design Festival is a platform for hundreds of design stories to be told. Each of them speaks to an expanding audience hungry for design ideas and enjoying the quality and diversity of what’s on offer. It all confirms London’s status as the global center of design London and design go hand in hand. It is part of our story. London Design Festival is a platform for hundreds of design stories to be told. Each of them speaks to an expanding audience hungry for design ideas and enjoying the quality and diversity of what’s on offer. It all confirms London’s status as the global center of design – says Ben Evans CBE. – says Ben Evans CBE. Ben Evans It’s hard to disagree with the words of this passionate advocate and activist for all-encompassing creativity, which can be found in everything that surrounds us. Today, we’ll spot it in glass, but in a broader sense, we’ll explore the role materials play in design – and in parallels. Parallels not only between the red oak beams of the exhibit at Chelsea College of Arts but also between the architecture of play and real life. Chelsea College of Arts The Sun, My Heart The Sun, My Heart The energy within us has the power to create incredible things, especially when we learn to recognize it and channel it in the right direction – one where we create something extraordinary that captivates the world. Energy is the central theme in the work of Dutch designer Marjan van Aubel . Her creations can hardly be described as merely "sunny" although, in reality, each one of them is filled with light. Without this element, we can no longer imagine her work. Marjan van Aubel Marjan van Aubel Energy, as an aesthetic element, has become an integral part of every object crafted by her hands: from glassware, which reflects the flow of matter, to entire installations. We, in turn, have become vessels filled with her revelation: They were transparent and colored. It was the first time I saw that solar energy could be beautiful. From that moment, I started implementing it into different objects: glassware, tables, light installations, buildings. Now, I see all surfaces as a potential to harvest energy They were transparent and colored. It was the first time I saw that solar energy could be beautiful. From that moment, I started implementing it into different objects: glassware, tables, light installations, buildings. Now, I see all surfaces as a potential to harvest energy Speaking of installations, we can now explore her piece “The Sun, My Heart” a highlight of the LDF collection. The title becomes somewhat rhetorical when you consider van Aubel’s relationship with the primary source of life, with roots that delve deep into Buddhist philosophy. In the language of Vietnamese monks, it calls on us to "open our hearts to new connections". These connections come to life through 77 self-charging Sunne lamps (some of which represent nothing more than rays of light), symbolizing the complete solar cycle, accompanied by harmoniously selected sounds that echo the sun's vibrations. Working in unison, this masterpiece immerses the viewer in a state of total relaxation, stimulating the brain's theta waves. We receive the sun in such abundance. It’s obvious that we need to transition to solar energy, but the current approach is too technical. We shouldn’t only talk about numbers and efficiency. The sun is something holistic; we can feel it. We need beauty and design to make us want to change things. We need to feel it inside We receive the sun in such abundance. It’s obvious that we need to transition to solar energy, but the current approach is too technical. We shouldn’t only talk about numbers and efficiency. The sun is something holistic; we can feel it. We need beauty and design to make us want to change things. We need to feel it inside – the designer shares. – the designer shares. Pavilions of Wonder Pavilions of Wonder A fresh perspective with a historical undertone, embodying cultural heritage – the true American dream, brought to life in a trilogy of works by designer Nina Tolstrup. This project has become a symbol of vivid imagination and aesthetics, driven by a spirit of fun and playfulness. The installation consists of three objects, harking back to the time when Palm Springs became the true capital of modernism in the USA, owing to its ability to "grow rapidly" with affordable housing after World War II. This area, dotted with buildings of the corresponding architectural style, would later become a prototype of inspiration for future designers. Among those designers is Nina Tolstrup , who infused the project with the same playful spirit reflected in Barbie’s DreamHouses . It is through her vibrant qualities, captured like film on a reel, that we recognize the essence of the designer. Nina Tolstrup Barbie’s DreamHouses Barbie’s DreamHouses We looked at the evolution of the DreamHouse, informed by Greater Palm Springs architecture. We examined the visual language and iconic elements common to both We looked at the evolution of the DreamHouse, informed by Greater Palm Springs architecture. We examined the visual language and iconic elements common to both Vert Vert And here are the very same "parallels" made from red oak, presented in the collaborative work of the American Hardwood Export Council (AHEC), the Munich-based Diez Office, and Frankfurt's Office for Microclimatic Cultivation (OMC°C). This project is far from a cry for help, but rather a message of mindful care for the environment. Vert, which signifies something "green" is an undeniably bold statement, offering a tangible and effective solution to a global issue. It represents a solution through a small, carefully planned, and meticulously thought-out noble endeavor. It can't be achieved just by planting trees or through individual façade and rooftop greening projects, so additional solutions are urgently needed. We need fairly large structures to generate the necessary amount of shade and biomass It can't be achieved just by planting trees or through individual façade and rooftop greening projects, so additional solutions are urgently needed. We need fairly large structures to generate the necessary amount of shade and biomass – explains Nicola Stattmann, co-founder of OMC°C. – explains Nicola Stattmann, co-founder of OMC°C. The architectural installation consists of several triangular structures, each standing 10 meters tall. These are all covered with plants such as morning glory, cheese plant, black-eyed Susan, and night-blooming moonflower, which will act as "killers" of carbon dioxide released into the atmosphere. Interestingly, the modularity of the designed structure allows for Vert to be produced in series in the future. This once again highlights the brilliance of the minds behind the project, brought to life with the support of AHEC, who have long been advocating for the use of durable engineered materials. Red oak glulam's natural beauty, combined with its structural capabilities, makes it a promising choice both functionally and aesthetically Red oak glulam's natural beauty, combined with its structural capabilities, makes it a promising choice both functionally and aesthetically – says Stefan Diez, Diez Office founder. – says Stefan Diez , Diez Office founder. Stefan Diez Stefan Diez The London Design Festival 2024 masterfully demonstrated the power of design in tackling global issues – from sustainable energy solutions to supporting biodiversity and reimagining urban spaces. Through innovative projects, the festival highlighted the crucial role of creativity in shaping a more sustainable and harmonious future. As these projects blur the lines between functionality and beauty, they invite us to envision a world where technology, nature, and human ingenuity coexist seamlessly, encouraging a more thoughtful and responsible approach to life. The London Design Festival week was one of the most anticipated events for us. Our fields are closely intertwined – that much is clear. We keep track of the latest trends to make our 3D visualizations even more realistic, filled with the smallest details that reflect the quality of bringing the 'imagined' to life. We’re interested in everything: from the materials often used in designers' projects to the very concepts behind these works, because this knowledge embodies our commitment to staying current. We don't want to fall behind. We aim to build trust with every client interaction, ensuring that the result meets and exceeds their needs, leaving them fully satisfied The London Design Festival week was one of the most anticipated events for us. Our fields are closely intertwined – that much is clear. We keep track of the latest trends to make our 3D visualizations even more realistic, filled with the smallest details that reflect the quality of bringing the 'imagined' to life. We’re interested in everything: from the materials often used in designers' projects to the very concepts behind these works, because this knowledge embodies our commitment to staying current. We don't want to fall behind. We aim to build trust with every client interaction, ensuring that the result meets and exceeds their needs, leaving them fully satisfied – says Alexander Kasperovich, CEO of 3D visualization and rendering studio, Studio57. – says Alexander Kasperovich, CEO of 3D visualization and rendering studio, Studio57.

Fostering Biodiversity and Solar Energy Design: What Made the 2024 London Design Festival Memorable?

It's one o'clock in the morning. My phone starts going crazy. I groan and look over. Sure enough, it's another PagerDuty call. I am on-call this week, and it's the third time in as many days that I have been woken up for the same reason. Hundreds of our customers have complained in unison, for three days in a row, that they have purchased event tickets from us and arrived at the venue only to find that their tickets have already been redeemed. They are being denied entry, and they are understandably very angry. Initial Investigation Showed Some Very Odd Behavior. Customers had purchased tickets, and then a matter of minutes later, their tickets would be redeemed. This was far in advance of the event itself, so it was clearly unexpected. And it wasn't spread evenly across our customer base; the issue was heavily focused within some select groups of university students. We already knew that we had a fundamental issue with our redemption system; it was possible to scan the QR code in the ticket order confirmation email yourself like the door staff would do at the venue, which would redeem the ticket. Unfortunately, the bigger fix for this was still a few weeks away. Even so, why would large numbers of customers within specific university circles redeem or accidentally redeem their own tickets before the event?! None of our early theories made sense, and the issue kept happening. We Went Digging. Eventually, We Found the Issue. The automated virus scanner on specific university .edu email systems were hitting all links in our confirmation emails to check them for viruses, including programmatically scanning all QR codes. Therefore this one process, completely invisible to my team, was automatically redeeming our customers' event tickets before the customer even received the email. Never, Ever Update Persisted Data on Page-Load. By making the ticket QR code hit a webpage that immediately redeems the ticket on page-load, it meant that anything visiting that link, whether human or machine, would inadvertently update our database to mark the ticket as redeemed. There are thousands of unseen automated visitors to our websites that we don't think about most of the time. Here are some examples: Browsers prefetching resources. This is a big one. Let's say you visit https://myblog.com, and there are prominent links on that page to/blog-post-1 and /blog-post-2 . It is very likely that your browser will already be prefetching those pages in the background. If those pages mutate data on page-load, for example, to track visits, just being on the homepage is enough to trigger it, even though you might never click on the links. Google crawlers. Another big one. We recently saw a large number of errors that came from Google crawlers visiting pages and then triggering unauthenticated calls to APIs. Email scanners. Very common in schools, governments, large corporations, and more. Internet archive machines The simple lesson here is to avoid things happening automatically on page-loads as much as possible. Besides the issues that we experienced, unexpected side-effects like this are bad UX. If the action is "Redeem", make it happen when you click a button that says "Redeem". I hope that this was useful - subscribe now to get my Dev Diary in your inbox! It's one o'clock in the morning. My phone starts going crazy. I groan and look over. Sure enough, it's another PagerDuty call. I am on-call this week, and it's the third time in as many days that I have been woken up for the same reason. Hundreds of our customers have complained in unison, for three days in a row, that they have purchased event tickets from us and arrived at the venue only to find that their tickets have already been redeemed . They are being denied entry, and they are understandably very angry. tickets have already been redeemed very angry. Initial Investigation Showed Some Very Odd Behavior. Customers had purchased tickets, and then a matter of minutes later, their tickets would be redeemed. This was far in advance of the event itself, so it was clearly unexpected. And it wasn't spread evenly across our customer base; the issue was heavily focused within some select groups of university students. We already knew that we had a fundamental issue with our redemption system; it was possible to scan the QR code in the ticket order confirmation email yourself like the door staff would do at the venue, which would redeem the ticket. Unfortunately, the bigger fix for this was still a few weeks away. Even so, why would large numbers of customers within specific university circles redeem or accidentally redeem their own tickets before the event?! None of our early theories made sense, and the issue kept happening. We Went Digging. Eventually, We Found the Issue. The automated virus scanner on specific university .edu email systems were hitting all links in our confirmation emails to check them for viruses, including programmatically scanning all QR codes . automated virus scanner .edu programmatically scanning all QR codes Therefore this one process, completely invisible to my team, was automatically redeeming our customers' event tickets before the customer even received the email. before the customer even received the email. Never, Ever Update Persisted Data on Page-Load. By making the ticket QR code hit a webpage that immediately redeems the ticket on page-load, it meant that anything visiting that link, whether human or machine, would inadvertently update our database to mark the ticket as redeemed. anything There are thousands of unseen automated visitors to our websites that we don't think about most of the time. Here are some examples: Browsers prefetching resources. This is a big one. Let's say you visit https://myblog.com, and there are prominent links on that page to/blog-post-1 and /blog-post-2 . It is very likely that your browser will already be prefetching those pages in the background. If those pages mutate data on page-load, for example, to track visits, just being on the homepage is enough to trigger it, even though you might never click on the links. Browsers prefetching resources. This is a big one. Let's say you visit https://myblog.com , and there are prominent links on that page to /blog-post-1 and /blog-post-2 . It is very likely that your browser will already be prefetching those pages in the background. If those pages mutate data on page-load, for example, to track visits, just being on the homepage is enough to trigger it, even though you might never click on the links. Browsers prefetching resources. https://myblog.com /blog-post-1 /blog-post-2 just being on the homepage Google crawlers. Another big one. We recently saw a large number of errors that came from Google crawlers visiting pages and then triggering unauthenticated calls to APIs. Google crawlers. Another big one. We recently saw a large number of errors that came from Google crawlers visiting pages and then triggering unauthenticated calls to APIs. Google crawlers. Email scanners. Very common in schools, governments, large corporations, and more. Email scanners. Very common in schools, governments, large corporations, and more. Email scanners. Internet archive machines Internet archive machines Internet archive machines The simple lesson here is to avoid things happening automatically on page-loads as much as possible. Besides the issues that we experienced, unexpected side-effects like this are bad UX. If the action is "Redeem", make it happen when you click a button that says "Redeem". I hope that this was useful - subscribe now to get my Dev Diary in your inbox!

Dev Diary #3: The Sneaky Issue That Was Silently Ruining Our Database

Is Your Apache Ni-Fi Ready for Production?

Countering Mainstream Bias via End-to-End Adaptive Local Learning: Conclusion and References

Table of Links Abstract and 1 Introduction 2 Preliminaries 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.3 Ablation Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 6 Conclusion, Acknowledgements, and References 4.3 Ablation Study Next, we aim to investigate the effectiveness of different components in the proposed framework, including the proposed adaptive loss-driven gate module, the adaptive weight module, the gap mechanism in the adaptive weight module, and the loss change mechanism in the adaptive weight module. Adaptive Loss-Driven Gate. To verify the effectiveness of the proposed adaptive loss-driven gate module in the proposed TALL, we compare the MoE component with the adaptive loss-driven gate (denoted as LMoE, which is the TALL model without the adaptive weight module) to a conventional MoE (denoted as MoE) with the standard multilayer perceptron (MLP) as the gate learning from the dataset. By comparing LMoE and MoE, we can justify the effect of the proposed adaptive loss-driven gate. The results are present in Table 3, where we also include the result of MultVAE as a baseline. From the table, we can observe that MoE produces better performance for niche users compared to MultVAE but worse results for other users. This is caused by the MLP-based gate model in MoE, which cannot precisely allocate gate values to expert models to ensemble a strong customized model for different users. Conversely, we can see that even without the adaptive weight module, the LMoE can deliver greatly higher utilities for all types of users compared to MultVAE and MoE, showing the strong capability of the proposed adaptive loss-driven gate module in distributing gate values across expert models. And this result also demonstrates the efficacy of the proposed LMoE module in TALL in terms of addressing the discrepancy modeling problem. Adaptive Weight. To address the unsynchronized learning problem, we develop the adaptive weight module to dynamically adjust the learning paces of different users. To verify the effectiveness of the proposed adaptive weight, we compare the complete TALL algorithm (with both the loss-driven MoE module and the adaptive weight module) to the loss-driven MoE module (LMoE). The results on ML1M are shown in Table 4, from which we can see that TALL outperforms LMoE for all types of users, manifesting the effectiveness of the proposed adaptive weight method. Furthermore, we explore the effectiveness of two special mechanisms, the gap mechanism and the loss change mechanism within the adaptive weight module, in the subsequent sections. Gap Mechanism. To avoid the unstable loss problem at the initial training stage, we propose to have a gap for the adaptive weight method, i.e., we wait for a certain number of epochs at the initial training stage until the loss is stable and then apply the adaptive weight method. To verify the effectiveness of such a gap strategy, we compare the complete TALL (with a full version of the adaptive weight module including both the gap mechanism and the loss change mechanism) to a variation of TALL with the adaptive weight module without the gap mechanism. The comparison is presented in Table 4 as LMoE+LC vs. TALL. We can observe that the gap mechanism does have a significant influence on the model performance that the model with the gap mechanism (TALL) delivers better utilities for all types of users than the model without the gap mechanism (LMoE+LC). Loss Change Mechanism. Last, we aim to verify the effectiveness of the proposed loss change mechanism in the adaptive weight module. The goal of loss change is to counter the scale diversity problem when applying the adaptive weight module. Here, we compare the complete TALL (with a full version of the adaptive weight module including both the gap mechanism and the loss change mechanism) to a variation of TALL (LMoE+gap+L) with the adaptive weight module that uses original loss as introduced in Section 3.2. From the comparison result shown in Table 4, we see that the model using loss change (TALL) performs better than the model using original loss for calculating weights (LMoE+gap+L). TALL outperforms LMoE+gap+L for all types of users. In sum, by a series of comparative analyses, we show that the proposed adaptive loss-driven gate module, adaptive weight module, the gap mechanism in adaptive weight, and the loss change mechanism in adaptive weight are effective and play imperative roles in the proposed TALL framework. Authors:
(1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA;
(2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA;
(3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA;
(4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA;
(5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries 2 Preliminaries 3 End-to-End Adaptive Local Learning 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.2 Debiasing Performance 4.3 Ablation Study Ablation 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 5 Related Work 6 Conclusion, Acknowledgements, and References 6 Conclusion, Acknowledgements, and References 4.3 Ablation Study Next, we aim to investigate the effectiveness of different components in the proposed framework, including the proposed adaptive loss-driven gate module, the adaptive weight module, the gap mechanism in the adaptive weight module, and the loss change mechanism in the adaptive weight module. Adaptive Loss-Driven Gate. To verify the effectiveness of the proposed adaptive loss-driven gate module in the proposed TALL, we compare the MoE component with the adaptive loss-driven gate (denoted as LMoE, which is the TALL model without the adaptive weight module) to a conventional MoE (denoted as MoE) with the standard multilayer perceptron (MLP) as the gate learning from Adaptive Loss-Driven Gate. the dataset. By comparing LMoE and MoE, we can justify the effect of the proposed adaptive loss-driven gate. The results are present in Table 3, where we also include the result of MultVAE as a baseline. From the table, we can observe that MoE produces better performance for niche users compared to MultVAE but worse results for other users. This is caused by the MLP-based gate model in MoE, which cannot precisely allocate gate values to expert models to ensemble a strong customized model for different users. Conversely, we can see that even without the adaptive weight module, the LMoE can deliver greatly higher utilities for all types of users compared to MultVAE and MoE, showing the strong capability of the proposed adaptive loss-driven gate module in distributing gate values across expert models. And this result also demonstrates the efficacy of the proposed LMoE module in TALL in terms of addressing the discrepancy modeling problem. Adaptive Weight. To address the unsynchronized learning problem, we develop the adaptive weight module to dynamically adjust the learning paces of different users. To verify the effectiveness of the proposed adaptive weight, we compare the complete TALL algorithm (with both the loss-driven MoE module and the adaptive weight module) to the loss-driven MoE module (LMoE). The results on ML1M are shown in Table 4, from which we can see that TALL outperforms LMoE for all types of users, manifesting the effectiveness of the proposed adaptive weight method. Furthermore, we explore the effectiveness of two special mechanisms, the gap mechanism and the loss change mechanism within the adaptive weight module, in the subsequent sections. Adaptive Weight. Gap Mechanism . To avoid the unstable loss problem at the initial training stage, we propose to have a gap for the adaptive weight method, i.e., we wait for a certain number of epochs at the initial training stage until the loss is stable and then apply the adaptive weight method. To verify the effectiveness of such a gap strategy, we compare the complete TALL (with a full version of the adaptive weight module including both the gap mechanism and the loss change mechanism) to a variation of TALL with the adaptive weight module without the gap mechanism. The comparison is presented in Table 4 as LMoE+LC vs. TALL. We can observe that the gap mechanism does have a significant influence on the model performance that the model with the gap mechanism (TALL) delivers better utilities for all types of users than the model without the gap mechanism (LMoE+LC). Gap Mechanism Loss Change Mechanism. Last, we aim to verify the effectiveness of the proposed loss change mechanism in the adaptive weight module. The goal of loss change is to counter the scale diversity problem when applying the adaptive weight module. Here, we compare the complete TALL (with a full version of the adaptive weight module including both the gap mechanism and the loss change mechanism) to a variation of TALL (LMoE+gap+L) with the adaptive weight module that uses original loss as introduced in Section 3.2. From the comparison result shown in Table 4, we see that the model using loss change (TALL) performs better than the model using original loss for calculating weights (LMoE+gap+L). TALL outperforms LMoE+gap+L for all types of users. Loss Change Mechanism. In sum, by a series of comparative analyses, we show that the proposed adaptive loss-driven gate module, adaptive weight module, the gap mechanism in adaptive weight, and the loss change mechanism in adaptive weight are effective and play imperative roles in the proposed TALL framework. Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. Authors: Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv

Countering Mainstream Bias via End-to-End Adaptive Local Learning: Ablation Study

Table of Links Abstract and 1 Introduction 2 Preliminaries 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.3 Ablation Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 6 Conclusion, Acknowledgements, and References 3.2 Synchronized Learning via Adaptive Weight After addressing the discrepancy modeling problem, another core cause of the mainstream bias is the unsynchronized learning problem – the learning difficulties vary for different users, and users reach the performance peak during training at different speeds (refer to Figure 1 for an example). Thus, a method to synchronize the learning paces of users is desired. We devise an adaptive weight approach to achieve learning synchronization so that the dilemma of performance trade-off between mainstream users and niche users can be overcome. Adaptive Weight. The fundamental motivation of the proposed method lies in linking the learning status of a user to the loss function of the user at the current epoch. A high loss for a user signifies ineffective learning by the model, necessitating more epochs for accurate predictions. Conversely, a low loss indicates successful modeling, requiring less or even no further training. Moreover, we can use a weight in the loss function to control the learning pace for the user: a small weight induces slow updating, and a large weight incurs fast updating. Hence, we propose to synchronize different users by applying weights to the objective function based on losses users get currently – a user with a high loss should receive a large weight, and vice versa. We aim to achieve this intuition by solving the following optimization problem: Loss Change Mechanism and Gap Mechanism. With the proposed adaptive weight approach, we take a solid step toward learning synchronization. However, two critical issues remain unaddressed. First, the scale of the loss function is innately different across different users. Usually, mainstream users possess a lower loss value than niche users because the algorithm can achieve better utility for mainstream users who are easier to model. Due to this scale diversity problem, computing weights by exact values of the loss can lead to the undesired situation that mainstream users get overly low weights and niche users get overly high weights, disturbing the learning process. On the other hand, the loss is not always stable, especially at the early stage of training. This unstable loss problem can deteriorate the efficacy of the problem adaptive weight method too. At last, since the losses are excessively unstable at the initial stage of training and the proposed adaptive weight module heavily relies on the stability of the loss value, the proposed method cannot perform well at the initial stage of training. Hence, we propose to have a gap at the beginning for our adaptive weight method. That is, we do not apply the proposed adaptive weight method to the framework at the first T epochs. Since at the early stage of the training, all users will be at a fast learning status (consider the first 50 epochs in Figure 1), as there is no demand for learning synchronization. After T epochs of ordinary training, when the learning procedure is more stable and the loss is more reliable, we apply the adaptive weight method to synchronize the learning for different users. And this time synchronization is desired and plays an important role. The gap window T is a hyper-parameter and needs to be predefined. Authors:
(1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA;
(2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA;
(3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA;
(4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA;
(5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries 2 Preliminaries 3 End-to-End Adaptive Local Learning 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.2 Debiasing Performance 4.3 Ablation Study Ablation 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 5 Related Work 6 Conclusion, Acknowledgements, and References 6 Conclusion, Acknowledgements, and References 3.2 Synchronized Learning via Adaptive Weight After addressing the discrepancy modeling problem, another core cause of the mainstream bias is the unsynchronized learning problem – the learning difficulties vary for different users, and users reach the performance peak during training at different speeds (refer to Figure 1 for an example). Thus, a method to synchronize the learning paces of users is desired. We devise an adaptive weight approach to achieve learning synchronization so that the dilemma of performance trade-off between mainstream users and niche users can be overcome. Adaptive Weight. The fundamental motivation of the proposed method lies in linking the learning status of a user to the loss function of the user at the current epoch. A high loss for a user signifies ineffective learning by the model, necessitating more epochs for accurate predictions. Conversely, a low loss indicates successful modeling, requiring less or even no further training. Moreover, we can use a weight in the loss function to control the learning pace for the user: a small weight induces slow updating, and a large weight incurs fast updating. Hence, we propose to synchronize different users by applying weights to the objective function based on losses users get currently – a user with a high loss should receive a large weight, and vice versa. We aim to achieve this intuition by solving the following optimization problem: Adaptive Weight. Loss Change Mechanism and Gap Mechanism. With the proposed adaptive weight approach, we take a solid step toward learning synchronization. However, two critical issues remain unaddressed. First, the scale of the loss function is innately different across different users. Usually, mainstream users possess a lower loss value than niche users because the algorithm can achieve better utility for mainstream users who are easier to model. Due to this scale diversity problem , computing weights by exact values of the loss can lead to the undesired situation that mainstream users get overly low weights and niche users get overly high weights, disturbing the learning process. On the other hand, the loss is not always stable, especially at the early stage of training. This unstable loss problem can deteriorate the efficacy of the problem adaptive weight method too. Loss Change Mechanism and Gap Mechanism. scale diversity problem unstable loss problem At last, since the losses are excessively unstable at the initial stage of training and the proposed adaptive weight module heavily relies on the stability of the loss value, the proposed method cannot perform well at the initial stage of training. Hence, we propose to have a gap at the beginning for our adaptive weight method. That is, we do not apply the proposed adaptive weight method to the framework at the first T epochs. Since at the early stage of the training, all users will be at a fast learning status (consider the first 50 epochs in Figure 1), as there is no demand for learning synchronization. After T epochs of ordinary training, when the learning procedure is more stable and the loss is more reliable, we apply the adaptive weight method to synchronize the learning for different users. And this time synchronization is desired and plays an important role. The gap window T is a hyper-parameter and needs to be predefined. Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. Authors: Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv

Countering Mainstream Bias via End-to-End Adaptive Local Learning: Adaptive Weight

Table of Links Abstract and 1 Introduction 2 Preliminaries 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.3 Ablation Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 6 Conclusion, Acknowledgements, and References 3 End-to-End Adaptive Local Learning To debias, we propose the end-To-end Adaptive Local Learning (TALL) framework, shown in Figure 2. To address the discrepancy modeling problem, this framework integrates a loss-driven Mixture-of-Experts module to adaptively provide customized models for different users by an end-to-end learning procedure. To address the unsynchronized learning problem, the framework involves an adaptive weight module to synchronize the learning paces of different users by adaptively adjusting weights in the objective function. Authors:
(1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA;
(2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA;
(3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA;
(4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA;
(5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries 2 Preliminaries 3 End-to-End Adaptive Local Learning 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.2 Debiasing Performance 4.3 Ablation Study Ablation 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 5 Related Work 6 Conclusion, Acknowledgements, and References 6 Conclusion, Acknowledgements, and References 3 End-to-End Adaptive Local Learning To debias, we propose the end- T o-end A daptive L ocal L earning ( TALL ) framework, shown in Figure 2. To address the discrepancy modeling problem, this framework integrates a loss-driven Mixture-of-Experts module to adaptively provide customized models for different users by an end-to-end learning procedure. To address the unsynchronized learning problem , the framework involves an adaptive weight module to synchronize the learning paces of different users by adaptively adjusting weights in the objective function. T A L L TALL discrepancy modeling loss-driven Mixture-of-Experts unsynchronized learning problem adaptive weight Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. Authors: Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv

Countering Mainstream Bias via End-to-End Adaptive Local Learning: Adaptive Local Learning

Table of Links Abstract and 1 Introduction 2 Preliminaries 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.3 Ablation Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 6 Conclusion, Acknowledgements, and References 2 Preliminaries Mainstream Bias. Typically, a recommender system is evaluated by averaging the utility over all users (such as NDCG@K), which essentially conceals the performance differences across different types of users. Previous work [39] formalizes the mainstream bias as the recommendation performance difference across users of different levels of mainstreamness. In this work, we follow the problem setting of [39] to measure the mainstream level for a user by calculating the average similarity of the user to all others: the more similar the user is to the majority, the more mainstream she is. Debiasing Goal. In terms of the goal of debiasing, while addressing the issues of discrepancy modeling and unsynchronized learning benefits all user types, it is inappropriate to expect equalized utility across all users, which possibly encourages decreasing the utility for mainstream groups. Hence, in this work, we follow the Rawlsian Max-Min fairness principle of distribute justice [29]. To achieve fairness, this principle aims to maximize the minimum utility of individuals or groups, ensuring that no one is underserved. So, to counter the mainstream bias, we aim to improve the recommendation utility of niche users while preserving or even enhancing the performance of mainstream users. Authors:
(1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA;
(2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA;
(3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA;
(4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA;
(5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries 2 Preliminaries 3 End-to-End Adaptive Local Learning 3 End-to-End Adaptive Local Learning 3.1 Loss-Driven Mixture-of-Experts 3.1 Loss-Driven Mixture-of-Experts 3.2 Synchronized Learning via Adaptive Weight 3.2 Synchronized Learning via Adaptive Weight 4 Debiasing Experiments and 4.1 Experimental Setup 4 Debiasing Experiments and 4.1 Experimental Setup 4.2 Debiasing Performance 4.2 Debiasing Performance 4.3 Ablation Study Ablation 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 4.4 Effect of the Adaptive Weight Module and 4.5 Hyper-parameter Study 5 Related Work 5 Related Work 6 Conclusion, Acknowledgements, and References 6 Conclusion, Acknowledgements, and References 2 Preliminaries Mainstream Bias. Typically, a recommender system is evaluated by averaging the utility over all users (such as NDCG@K), which essentially conceals the performance differences across different types of users. Previous work [39] formalizes the mainstream bias as the recommendation performance difference across users of different levels of mainstreamness. In this work, we follow the problem setting of [39] to measure the mainstream level for a user by calculating the average similarity of the user to all others: the more similar the user is to the majority, the more mainstream she is. Mainstream Bias. the recommendation performance difference across users of different levels of mainstreamness. Debiasing Goal. In terms of the goal of debiasing, while addressing the issues of discrepancy modeling and unsynchronized learning benefits all user types, it is inappropriate to expect equalized utility across all users, which possibly encourages decreasing the utility for mainstream groups. Hence, in this work, we follow the Rawlsian Max-Min fairness principle of distribute justice [29]. To achieve fairness, this principle aims to maximize the minimum utility of individuals or groups, ensuring that no one is underserved. So, to counter the mainstream bias, we aim to improve the recommendation utility of niche users while preserving or even enhancing the performance of mainstream users. Debiasing Goal. Rawlsian Max-Min fairness improve the recommendation utility of niche users while preserving or even enhancing the performance of mainstream users. Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. Authors: Authors: (1) Jinhao Pan [0009 −0006 −1574 −6376], Texas A&M University, College Station, TX, USA; (2) Ziwei Zhu [0000 −0002 −3990 −4774], George Mason University, Fairfax, VA, USA; (3) Jianling Wang [0000 −0001 −9916 −0976], Texas A&M University, College Station, TX, USA; (4) Allen Lin [0000 −0003 −0980 −4323], Texas A&M University, College Station, TX, USA; (5) James Caverlee [0000 −0001 −8350 −8528]. Texas A&M University, College Station, TX, USA. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv

Stories, News & Mentions about UNISYN

Unisyn’s Stories on HackerNoon