MAXA is a tech start-up that provides an AI \u0026 data automation layer that enables businesses to get more answers and better BI from their ERP and core systems - all at warp speed

Maxa AI

This story is a part of Hacker Noon's series of interviews. The series is intended for tech professionals contributing the most insightful Hacker Noon stories to share more about their writing habits, ideas, and professional background (and maybe a hobby or two). If you too would like to start contributing to Hacker Noon, . Meet the Writer you can do so here So let’s start! Tell us a bit about yourself. For example, name, profession, and personal interests. I am Max, a data scientist and AI developer leading the development of the crypto bot trading algorithms. My friends call me “the nerd with the PhD” (behind my back :-). SmithBot My personal interests are computing, robots and cryptocurrency and FOREX trading. Interesting! What was your latest Hackernoon Top story about? One of my passions is to explain complex topics to outsiders and make them interested in new topics. I also like to educate people on things I believe they should know to be successful in life. Therefore, my latest article was a where interested folks can find all they need to know in one place. beginner-friendly guide on cryptocurrency trading Do you usually write on similar topics? If not, what do you usually write about? Most of my contributions are really about cryptocurrency, trading, AI and all kind of tech topics. I am writing for all kinds of audience in blogs and news sites, but also for peer-reviewed scientific journals from time to time. Great! What is your usual writing routine like (if you have one?) As structure and the reading flow seems very important, I always start with an outline, so I put down the headlines first. Next, I add the graphical content, like images, diagrams and tables. At the very end I fill the space between these elements with text. Might not work for everyone, but for me it works great. Being a writer in tech can be a challenge. It’s not often our main role, but an addition to another one. What is the biggest challenge you have when it comes to writing? Assuming the role of your readers. As experts in our field, we often write in way outsiders cannot understand. While this might be appropriate when writing for a scientific journal, in most other cases it is not. My goal is that the average reader finds my stories interesting and helpful. What is the next thing you hope to achieve in your career? One of my dreams is teaching computer science students at the university when I have free time some day. At the moment, I’m too occupied with my work, however. Wow, that’s admirable. Now, something more casual: What is your guilty pleasure of choice? I love sushi and eat way too much of it … Do you have a non-tech-related hobby? If yes, what is it? Definitely! My other hobbies are swimming, scuba diving and walking my crazy dog. What can the Hacker Noon community expect to read from you next? One topic on my list is explaining some of the tech behind our crypto trading bots so people can understand that this not voodoo, but rooted in science. Another topic I was long planning to write about is popular misconceptions about Artificial Intelligence. Conceptually, it is really much simpler than most people think it is. Thanks for taking time to join our “Meet the writer” series. It was a pleasure. Do you have any closing words? Thanks for having me on Hackernoon. It’s a great community with great content. The place where nerds feel home ;-)

Meet the Writer: Hacker Noon's Contributor Max Palacios, AI developer

This paper is under CC 4.0 license. available on arxiv Authors: (1) Domenico Cotroneo, University of Naples Federico II, Naples, Italy; (2) Alessio Foggia, University of Naples Federico II, Naples, Italy; (3) Cristina Improta, University of Naples Federico II, Naples, Italy; (4) Pietro Liguori, University of Naples Federico II, Naples, Italy; (5) Roberto Natella, University of Naples Federico II, Naples, Italy. Table of Links Abstract & Introduction Motivating Example Proposed Method Experimental Setup Experimental Results Related Work Conclusion & References 3. Proposed Method To overcome the limitations in the assessment of AI-generated assembly code described in the previous section, we propose ACCA, a method to automatically assemble and symbolically execute both the reference (i.e., the ground truth) and predicted snippets (i.e., the code generated by the models). Through symbolic execution, the method simulates the execution of both programs and determines whether, starting from the same state, they terminate producing equivalent results and effects. First, ACCA compares, at the string level, the prediction with the reference to assess its correctness, because, if the prediction is equal to the reference, then we assume the code prediction is correct. If this is not the case, the method assesses whether the code is syntactically correct. Indeed, if a code prediction is not structured according to the rules of the target programming languages, it is classified as incorrect. If this is not the case, i.e., if the prediction differs from the ground truth reference and the prediction is syntactically correct, ACCA generates two source code files, one for the reference and one for the predicted snippet. It then assembles and links them to produce the executable files needed for the symbolic execution. At this point, the method symbolically executes both files resulting from the assembling process to assess whether they are equivalent. Finally, ACCA returns the pair of syntactic correctness value (SYN) and the semantic correctness value (SEM) of the code predicted by the model, which is equal to 1 when the correctness is verified, 0 otherwise. Figure 1 shows a detailed flowchart of the syntactic and semantics correctness evaluation process. In the rest of this section, we detail the phases of the workflow. For the sake of simplicity, we showcase examples of assembly code for the Linux OS running on IA-32 architecture to describe the evaluation process, although the proposed method is not restricted to this specific context, and it can be applied to different instruction set architectures and operating systems. 3.1. String Comparison ACCA first checks whether the predicted code snippet perfectly matches the ground truth by performing a string comparison. If they match (i.e., the code generated by the model is equal to ground truth), the prediction is considered both syntactically and semantically correct (i.e., SYN=1, SEM=1) and the evaluation ends. Otherwise, i.e., if they differ, the method proceeds with the evaluation of the syntactic and semantic correctness. The preliminary string comparison is done to speed up the evaluation process by skipping the symbolic execution process when not needed, i.e., when the prediction perfectly matches the reference snippet and is, therefore, correct. 3.2. Assembling Process The purpose of the assembling process is to assess whether each code snippet generated by the model adheres to the syntax rules of the programming language it is written in, i.e., to check whether it is compilable. Since NMT is still far from producing entire complex programs, the output predicted by the models is a portion of an entire program (i.e., a single-line or multi-line statement). Thus, ACCA constructs a complete program by adding the necessary entry point and exit point code. For instance, consider the following code snippets for IA-32: cmp EAX, EBX je loop pop ECX This code compares the contents of two registers and, based on the result, either performs a jump operation or reads from the stack and saves the value into another register. The snippet is syntactically correct according to the assembly language, yet it is incomplete for the execution since code snippets need to contain a minimal set of information to be properly executed. In the case of Linux OS, these instructions are kept in the text section, which must begin with the declaration global start to define the initial program entry point. Moreover, to make sure the program terminates correctly, the proposed method also includes a fictitious label that represents the exit address the code jumps to at the end of its execution. This label is declared in the data section, which contains initialized read-write data. Therefore, to have a complete program, the code is modified as follows: section .data exit_addr db 0x56 section .text global _start _start: cmp EAX, EBX je loop pop ECX jmp exit_addr Once the whole program is created, ACCA generates a source file and leverages an assembler to assess its syntactic correctness. Indeed, if the programs compile, then all the instructions of the programs are syntactically correct, therefore also the code generated by the model respects the structure of assembly programming language. There are three possible output scenarios for the compilation: , in which the assembling process is completely correct; • No errors in which the assembler reports some type of warning (e.g., word data exceeds bounds, invalid register size specification ignored, etc.), but the compilation still terminates without errors; • Warnings, , in which the assembling process results in an error that prevents the code from being assembled. • Errors In the first two cases, the output produced by the model is considered syntactically correct (i.e., SYN=1). Warnings are considered acceptable since they indicate issues involving bad practice, but are not severe enough to prevent the code compilation. When the compilation produces an error (i.e., the third case), we investigate the nature of the error. More precisely, we focus on a specific category of the error raised by the compiler, the SymbolNot-Defined (SDN) errors, which occur when the code contains a symbol (e.g., a label or variable) that has not been previously defined or initialized. We handle this category of errors appropriately. Indeed, since the predicted snippets contain only one or a few instructions, they might reference a label or variable defined in a different portion of the program, which leads to an assembling error even when the program is syntactically correct. Consider, again, the previous code snippet: the first instruction compares the contents of the EAX and EBX registers and, if they are equal, the execution jumps to the loop label. However, this symbol has not been defined yet. To handle these cases, we analyze the assembler output to determine the missing symbol’s name and include it in the source code file as a fictitious label. This label simply points to a jump operation to the previously defined exit address (i.e., myExitAddr). Indeed, the destination of the jump is not significant for the evaluation since we are only interested in checking the correctness of the instructions generated by the model. Therefore, after a SDN error, ACCA further modifies the program as follows: section .data myExitAddr db 0x56 section .text global _start _start: cmp EAX, EBX je loop pop ECX jmp myExitAddr loop: jmp myExitAddr Once the source code file is modified accordingly, ACCA repeats the assembling process as before to check if the compilation ends with no errors or warnings. In this case, ACCA assigns the SYN score equal to 1 and continues the evaluation to check the semantic correctness. When the compilation ends with errors different from the SDN, such as an invalid combination of opcode and operands, expression syntax error, etc., then ACCA labels the model’s prediction as syntactically incorrect (SYN=0). Since a snippet syntactically incorrect is also semantically incorrect, then the evaluation process terminates, assigning also the SEM score equal to zero. A source code file is generated and assembled for both the ground truth and the predicted code snippet. 3.3. Symbolic Execution To evaluate the semantic correctness, ACCA leverages the symbolic execution. To this aim, the method needs the program executable. If the assembling process ends correctly, the assembler outputs an object file, which is then fed to the linker to complete the process. Since the same operation may be correctly implemented in different ways, a simple textual comparison with the reference is not enough to assess the semantic correctness of a program. We still consider the ground truth as the reference for the correct implementation of the intended code. However, we do not limit the comparison to a textual similarity, but we examine the system state at the end of the execution of both the reference and the generated code. Indeed, two programs that implement the same functionality using different operations can be considered semantically equivalent if they result in the same final system state. Since the final execution state depends on the inputs and initial state of the program, we need to compare the state produced by both programs for every possible combination of inputs and initial state. Symbolic execution is a state-of-the-art solution for program analysis based on abstract execution. It consists in simulating the execution of a program providing symbolic values as its input instead of concrete ones. The result of this execution is a set of output values expressed as functions of the input symbols. ACCA uses symbolic execution to determine all the existing execution paths and all possible corresponding output system states. It then compares the set of output system states of the generated program with the set of output system states of the ground truth program: if they match, then the programs are semantically equivalent (SEM=1), otherwise, the method classifies the model’s prediction and the ground-truth as not semantically equivalent (SEM=0). To symbolically execute the programs, we use a binary analysis platform (BAP) that loads each executable and provides a complete abstract representation of the target architecture, CPU registers, memory address space, and stack. The program is conceived as a sequence of basic blocks, i.e., a straight-line code sequence with no branches, and the interconnections between the blocks represent the jump operations. An example is shown in Fig. 2: the program compares the contents of two registers and, if they are equal, the execution jumps to a specific address, otherwise, it jumps to the next instruction. Each possible branch is the entry point of a different basic block, which contains a sequence of operations that can be executed in order and one last instruction that causes the execution to move to another basic block. ACCA begins the execution by initializing the abstract registers and flags to the same symbolic values; the method also sets the value of the stack pointer and initializes the memory to a blank state. Then, it executes the program by simulating each instruction step by step and keeping track of its state at each given step. During the execution of the programs, performing operations, such as arithmetic operations, comparisons, assignments, etc., that involve a variable (e.g., if X > 10) will yield an execution tree, i.e., a tree of operations formed by all the possible paths of the program, which encode all branch decisions taken up to a specific point. Execution trees are then translated into constraints, i.e., formulas that express a set of assumptions on the symbolic outputs of the execution. These constraints are finally solved by a satisfiability modulo theories (SMT) solver, which is typically employed in program verification to verify whether constraints can be satisfied by some assignment of concrete values to the program’s symbolic arguments [17]. As an example, consider the following simple constraints: x > y y > 2 10 > x The SMT solver treats them as assertions that must be satisfied with the valid values of symbolic variables. It, therefore, outputs a value that is consistent with all the constraints (e.g., x=4, y=3). symbolically executes both the reference code and the predicted snippet: assuming that the program state at the beginning of the execution is identical, if the programs are semantically equivalent, then their state is also identical at the end of the execution. Therefore, to assess the semantic correctness of the generated code compared to the ground truth, the proposed method checks whether the states of the architecture are equal at the end of both executions. The program state includes: ACCA , i.e., the contents of the abstract CPU registers; • state of the registers , i.e., the abstract status register that keeps track of the CPU state, including the result of arithmetic operations (e.g., carry flag, overflow flag, etc.) and interruptions; • state of the flags , i.e., the contents of the memory area used to store temporary values; • values on stack i.e., the condition of input values, defined over the previous items, that leads to the corresponding final state. • path constraints, To this end, ACCA compares the state of each leaf node, i.e., the final states at the end of each path in the execution tree representing the program, of both executables. To compare the leaf nodes, the method constructs a set of lists for every final basic block of the two programs. Each set contains a list of register values, flag values, boolean constraints, and stack values. For example, a reference program whose execution tree ends with two basic blocks (leaf nodes) has two sets of lists. Each set contains all the values that represent the system state for that particular execution path. If the execution tree of the generated program has the same number of leaf nodes, then each list of the two sets is compared with each list of two sets of the reference program. If there is a correspondence between each leaf of the first program and each leaf of the second program, then they are semantically equivalent (SEM=1) and the evaluation process ends. Contrarily, if the leaf nodes of the two program execution trees do not match, then we conclude that the predicted code is not semantically equivalent to the reference snippet (SEM=0) and the process ends. Since the total number of states can grow exponentially in the number of branches, one of the main challenges of symbolic execution is the path explosion problem [17]. Indeed, keeping track of a large number of pending branches to be explored impacts both the running time and the space requirements of symbolic execution. The primary causes of path explosion are loops and function calls. One common method to deal with this problem is to bind the path exploration to a limited number of iterations. To handle programs whose symbolic execution does not terminate, we set a maximum number of execution steps. Since AI-generated code is typically concise and consists of a few assembly instructions, a correct program concludes its execution in a few execution steps. If it runs for more than max steps, then the symbolic execution stops, and the generated program is classified as incorrect (SEM=0). 3.4. Implementation Details To assemble the programs and generate the executable files, we rely on the wide set of available open-source software. For the previous examples (on the IA-32), we used the Netwide Assembler (NASM) [18], an 80x86 and x86-64 assembler that supports a wide range of executable formats, including Linux, BSD, and Windows operating system formats. As a binary analysis platform, we use ANGR [19]. ANGR provides support for a variety of CPU architectures, including ARM, MIPS, PPC, and x86 processors. It comprises a series of sub-components that implement the different steps necessary for the symbolic execution: to disassemble the executables and lift the binary code to an intermediate representation; to simulate the program state and execution, including registers and memory; and to solve the generated constraints, using the z3 [20] SMT solver as a backend. We set the maximum number of execution steps max steps to 100 to avoid infinite loops. Our implementation runs on both Linux and Windows OS. We will publicly share the implementation of ACCA on GitHub.

Automating the Correctness Assessment of AI-generated Code for Security Contexts: Proposed Method

Playing Checkers With Genetic Algorithms: Evolutionary Learning in Traditional Games

জেনেটিক অ্যালগরিদম দিয়ে চেকার খেলা: ঐতিহ্যগত গেমে বিবর্তনীয় শিক্ষা

Dame spielen mit genetischen Algorithmen: Evolutionäres Lernen in traditionellen Spielen

Jugar a las damas con algoritmos genéticos: aprendizaje evolutivo en juegos tradicionales

Jouer aux dames avec des algorithmes génétiques : apprentissage évolutif dans les jeux traditionnels

जेनेटिक एल्गोरिदम के साथ चेकर्स खेलना: पारंपरिक खेलों में विकासवादी शिक्षा

遺伝的アルゴリズムを使用したチェッカーのプレイ: 従来のゲームにおける進化的学習

유전 알고리즘으로 체커 게임하기: 전통 게임의 진화 학습

Jogando Damas com Algoritmos Genéticos: Aprendizagem Evolutiva em Jogos Tradicionais

Игра в шашки с использованием генетических алгоритмов: эволюционное обучение в традиционных играх

Genetik Algoritmalarla Dama Oynamak: Geleneksel Oyunlarda Evrimsel Öğrenme

Chơi cờ đam bằng thuật toán di truyền: Học tập tiến hóa trong các trò chơi truyền thống

用遗传算法玩西洋跳棋：传统游戏中的进化学习

Hackernoon conducted this interview to help the community get to know Max better as well as find out some writing tips from him. So let’s start! Tell us a bit about yourself. For example, name, profession, and personal interests. I'm Max, a passionate Software Engineer specializing in the development of mobile applications with a strong focus on the iOS platform. My experience is deeply rooted in architecting apps, enhancing client-side performance, and developing sophisticated user interfaces, with a particular emphasis on utilizing in production. iOS SwiftUI Outside of work, I’m also an ambassador of (SWH), a remarkable universal source code archive. I believe in the importance of the Software Heritage mission to collect, preserve, and share all software source code. Software Heritage Interesting! What was your latest Hackernoon Top story about? My latest Hackernoon Top story focused on UI in iOS and macOS, specifically on creating a . While it might seem like a simple tale about small UI elements, there's more to it. Essentially, it's a reflection on the evolution of UI design in iOS and macOS, examining the changes in Apple's UI elements. The article goes beyond just a visual aspect, offering a brief tutorial on developing this progress bar. Retro Progress Bar for iOS using only CALayers Do you usually write on similar topics? If not, what do you usually write about? While I often explore UI efficiencies and challenges in iOS development, my writing also covers a broader range of topics. This includes tackling non-standard issues in iOS, such as integrating XCFrameworks within other frameworks, blending Objective-C with SwiftUI, and incorporating Unity into iOS projects. Great! What is your usual writing routine like (if you have one?) My writing process is dynamic. Sometimes, an article takes shape in just three days, especially when inspired by ongoing work projects. Initially, I develop a pet project, potentially shaping it into an open-source library, and then share my findings and experiences through an article. Being a writer in tech can be a challenge. It’s not often our main role, but an addition to another one. What is the biggest challenge you have when it comes to writing? I think the initial step of starting an article is often the most challenging. All tech writers experience this. However, a significant hurdle arises when there's insufficient information to substantiate my technological ideas or solutions. A notable example was when I was writing about XCFrameworks. I encountered unexpected complexities, which taught me the importance of fully developing and testing a solution before commencing the writing process. Btw, the still is one of the most popular articles on my list. article about XCFrameworks Additionally, there's a personal aspect to my writing. I sometimes find myself referring back to my articles for specific solutions I've tackled already in the past. It's like creating a personal knowledge base, which is useful not only for the community and my readers but also serves as a valuable resource for me. What is the next thing you hope to achieve in your career? In the next stage of my professional journey, my objective is to elevate my proficiency in optimizing performance within iOS environments. This includes a dedicated effort to master low-level APIs, which are crucial for fine-tuning the performance and efficiency of iOS applications. Alongside this, I am deeply interested in the integration of AI technologies, recognizing their transformative impact on app functionalities and user experiences. A significant part of my focus will also be on gaining expertise in Apple's Vision PRO. This revolutionary mixed-reality headset presents a new frontier for iOS/MacOS developers, and I am excited about the potential of creating immersive and interactive applications using its advanced capabilities. My goal is to be at the forefront of these technologies, contributing innovative solutions and setting new standards in the realm of mobile app development. Wow, that’s admirable. Now, something more casual: What is your guilty pleasure of choice? My guilty pleasure leans towards the realm of classic Blizzard games, particularly Diablo and Starcraft. There's a unique sense of nostalgia and excitement that comes from diving back into these iconic games. They remind me of earlier days when gaming was just as much about strategy and storytelling as it was about graphics and speed. Playing Diablo and Starcraft offers a delightful contrast to my professional focus on cutting-edge technology. It's not just a fun escape but also a way to reconnect with the simpler joys of gaming, allowing me to unwind and recharge. Do you have a non-tech-related hobby? If yes, what is it? I would say skateboarding and surfing are my go-to activities. They're not just hobbies but also a source of energy and inspiration, a way to focus and abstract from everything, so it’s helping me to bring creativity and enthusiasm to my professional projects. What can the Hacker Noon community expect to read from you next? I'm currently working on a couple of exciting projects for my upcoming articles. The first one will delve into the world of low-level APIs in iOS, while the second will discuss the architecture I've chosen for an iOS app I'm about to release. These articles are designed to be valuable not just for iOS Engineers but for a broader audience interested in technology. What’s your opinion on HackerNoon as a platform for writers? HackerNoon, in my view, stands out as an incredible platform for writers, especially those in the tech field. It provides a unique space where diverse voices can share their insights, experiences, and innovations. What I particularly appreciate is its inclusive and supportive community, which encourages writers of all skill levels to contribute. The platform's wide-ranging topics, from deep technical discussions to broader industry trends, make it an invaluable resource for anyone interested in technology. As a writer, I find that HackerNoon not only allows me to share my expertise but also challenges me to grow and learn from others. It's a dynamic environment where the exchange of ideas is not just encouraged but celebrated, making it a fantastic community for both seasoned professionals and budding enthusiasts. Thanks for taking time to join our “ ” series. It was a pleasure. Do you have any closing words? Meet the writer Thank you for this engaging conversation and the opportunity to share my experiences with the Hackernoon community. If at least one of my articles has helped someone tackle a challenge or solve a problem, then I consider my efforts a success. That's the true reward of writing in the tech field – the ability to impact and assist others in their professional journey. I'm grateful for the platform Hackernoon provides, enabling us to share knowledge and learn from each other. Here's to continuing to explore, innovate, and support one another in this dynamic and ever-evolving tech world. See you in my articles!

Meet the Writer: HackerNoon's Contributor Max Kalik, iOS Software Engineer

Our ever-evolving digital landscape and the technologies that support it have enabled us to take care of everything from banking and shopping to socializing online. Yet our increased reliance on digital services has also led to growing concerns about security and privacy and the access that third parties have to our data. The emergence of represents a promising evolution in the eyes of many experts. Web 3.0 That’s because Web 3.0 technologies use a decentralized framework that puts control back into the hands of users and allows them to interact online without sacrificing privacy. The increasing prominence of , decentralized finance, and non-fungible tokens (NFTs) further emphasizes the demand for a new way of safeguarding our digital interactions. blockchain-driven applications A new breed of tech visionaries has been instrumental in driving this change and helping to deliver Web 3.0 solutions to the masses. These leaders possess a profound understanding of the complexities of cybersecurity and the limitless possibilities of Web 3.0 to embrace the potential of decentralized networks while acknowledging the unique of 2023. security challenges Here, we celebrate 15 experts who are helping to shape the future of Web 3.0 and deliver powerful new cybersecurity solutions to create a digital infrastructure where decentralized technologies redefine the internet and safeguard its most valuable assets – our data and privacy. Jan Lunter, CEO & CTO of Innovatrics Jan Lunter graduated at the Télécom ParisTech University in France before becoming co-founder and CEO of Innovatrics, a company that has been developing and providing fingerprint recognition solutions since 2004. Lunter is an author of the algorithm for fingerprint analysis and recognition, which regularly ranks among the top in prestigious comparison tests (NIST PFT II, NIST Minex). In recent years he is also dealing with image processing and the use of neural networks for face recognition. , CEO and Cofounder of Lightning Labs Elizabeth Stark Elizabeth Stark is a strong advocate of open source and decentralized protocols, who brings her vast knowledge in peer-to-peer technology, privacy, open source software, and memes to her roles as a co-founder and CEO of Lightning Labs. Her company focuses on constructing a programmable financial layer for the Internet through fast and scalable blockchain transactions. With a law degree from Harvard and a background in teaching at Stanford and Yale, Stark remains a prominent figure in shaping the future of the internet and empowering startups in various fields, including AI and decentralized technology. Her dedication to Internet freedom was evident during the successful efforts to defeat SOPA/PIPA, engaging millions of people worldwide. , CEO of ManagedMethods Charlie Sander Charlie Sander is the CEO of Managed Methods, a Colorado-based data security and student safety platform for K-12 schools. With more than three decades of experience in the IT industry, Sander has been an executive at some of the fastest-growing companies in business. He holds 10 patents and graduated from the Cockrell School of Engineering at the University of Texas at Austin with a BSEE degree. , Founder & CEO of Custodia Bank Caitlin Long Caitlin Long is a Wall Street veteran deeply involved in Bitcoin and blockchain since 2012. With a remarkable career at major financial institutions, including Morgan Stanley and Credit Suisse, Long holds a law degree from Harvard, further solidifying her exceptional qualifications. Custodia Bank, a Wyoming-based institution, serves as a compliant bridge to the US dollar payments system while providing secure custodial services for digital assets at institutional standards. As a fully compliant entity, Custodia adheres to the Bank Secrecy Act and federal regulations, including “know your customer,” anti-money laundering, and related laws. Additionally, the bank follows Wyoming’s special purpose depository institution and digital asset laws, ensuring 100% reserved fiat deposits and offering the highest level of consumer protection in the digital asset industry. Founder & CEO, BreachQuest Shaun Gordon, Gordon founded BreachQuest, a cybersecurity company focused on minimizing breach costs and downtime. He is also the founder and Managing Partner of AGI Partners, an investment firm dedicated to growing exceptional mid-sized companies. His achievements include successfully leading the sale of Pristine Environments and contributing to educational initiatives. Gordon’s expertise spans business services, cybersecurity, SaaS, real estate, financial services, and healthcare. Cathy Hackl, Chief Futurist & Founder, Journey Cathy Hackl, a prominent business executive and tech futurist, boasts an impressive background working with major companies like Amazon Web Services, Magic Leap, and HTC VIVE. In her leading role at Journey’s Virtual Studio, she advises companies and governments on gaming, AR, AI, spatial computing, and virtual world strategies. Hackl’s expertise has influenced brands such as Louis Vuitton, Nike, Walmart, and Clinique. She is a recognized Vogue Business 100 Innovator and Vogue Singapore’s Metaverse Editor. In 2023, Hackl launched VerseLuxe, her luxury tech label disrupting the fashion and luxury industries with connected jewelry called Frillz. Journey, the global design and innovation studio, empowers clients to create next-gen customer experiences that engage and elevate, spanning the creator economy, immersive destinations, and new generations of fans. , Founder & CEO, Eden Data Taylor Hersom Taylor Hersom has extensive experience advising Fortune 500 companies and serving as a Chief Information Security Officer (CISO). His expertise lies in building tailored security programs for decentralized and cloud-centric organizations. Eden Data is dedicated to integrating security and privacy into the core of next-generation businesses. The company disrupts traditional cybersecurity consulting by prioritizing relationships, cultivating leaders, and offering a fresh approach. They actively contribute as a security thought leader, serving on industry advisory boards and sharing insights through speaking engagements and online content. Eugenia Kuyda, Founder and CEO at Replika Eugenia Kuyda is a Russian-born entrepreneur and computer scientist based in the United States. She is renowned as the co-founder and CEO of Replika, a cutting-edge AI-powered chatbot app for human-like conversation simulation. With over 10 million users worldwide and a remarkable 35% usage increase since 2020, Replika has garnered widespread attention and acclaim for its unique approach to mental health support. Prior to Replika, Kuyda co-founded Luka, an AI-driven messaging app that recommended local venues. Kuyda’s groundbreaking work in the AI and mental health spheres has earned her numerous speaking engagements and media features. , Co-Founder, Sensei Node Ignacio Roizman Ignacio Roizman is a seasoned Digital Marketing Executive and Entrepreneur, boasting over 25 years of experience leading operations in Latin America and the US. With a strong track record, he has successfully overseen organizations providing a range of digital advertising services, including Creative, Digital Media Planning and Buying, and website development for renowned Fortune 500 brands. SenseiNode is the first blockchain infrastructure independent service provider in Latin America. They automate and simplify the deployment and monitoring of nodes on leading protocols, providing access via their APIs to organizations looking to build their blockchain infrastructure with a high availability and low total cost of ownership. Kelly Max, Founder & CEO, Solvv Kelly Max is a visionary founder with a deep commitment to creating impactful products and companies that make a difference in people’s lives. His passion lies in assembling exceptional teams and driving focused efforts toward achieving ambitious goals. Currently, his primary focus is on Solvv, the pioneering Social Podcasting platform. At Solvv Inc, They strive to develop a sustainable product that empowers individuals to progress in life, shape their identities, and cultivate their personal brands. Solvv invites others to join their journey of learning, creation, and growth. In their past ventures. , Co-Founder and CEO at Everyrealm Janine Yorio Janine Yorio has established a metaverse innovation and investment platform. Everyrealm holds a significant position as one of the largest owners of digital real estate NFTs across metaverses like Decentraland, The Sandbox, and Axie Infinity. The company actively develops in-metaverse projects, including the Metajuku shopping center and the Fantasy Islands ultra-luxury master-planned community NFT project. Yorio’s previous achievements include being the CEO of Compound, a real estate fintech app acquired by Republic in 2020, and her work in private equity and real estate and hotel development. Jasmine Brunson, Founder, Rektify AI Jasmine Brunson is a dynamic founder and risk-taker who has ventured into data science, blockchain and technology. She sees data as her creative currency and actively contributes to the social entrepreneurial community, leveraging her passions to drive meaningful change. Brunson is a dedicated advocate for user data awareness, fighting against exploiting consumer data that fuels harmful algorithm biases and fosters misleading consumer identities. Her company, Rektify AI, is a Web3 Security & Risk Analytics company making cryptocurrency more secure and intelligent. , Founder and CEO of Crescite Eddie Cullen Eddie Cullen is the Founder and CEO of Crescite, which is building a Web 3.0 community for faith-based institutions and ESG investments. He was earlier the Director of Fordham University’s incubator program, and was a Candidate to be Mayor of NYC. He also served as a Fellow and Faith Advisor for the Biden for President campaign. Mike Starr, CEO and Founder of trackd Mike Starr is a cross-functional leader and former NSA engineer with experience building and launching products in new and disruptive markets. Mike received his Bachelor’s degree from SUNY Alfred and enjoys nerding out on wine and reading fantasy novels in his free time. His company trackd is delivering a modern patch management platform that collects patching experience data from patches applied using the platform, then anonymizes that data and shares it with all other users. The trackd platform unifies vulnerability and patch management and provides insight into how patches have affected systems previously, helping operators identify which patches might be disruptive when applied in their environment, and which are safe to designate for auto-patching. , Co-Founder of Cyber Security Capital Jane Frankland Jane Frankland brings two decades of cybersecurity expertise to Cyber Security Capital, holding senior executive roles and actively participating in esteemed cybersecurity organizations. Her mission is to promote women in security and create a safer, stronger, and more prosperous world for all. Cyber Security Capital operates globally, connecting people, opportunities, and resources within the cybersecurity ecosystem. Their London-based headquarters house a passionate and skilled team of cyber security entrepreneurs, consultants, facilitators, and resources. Cyber Security Capital emphasizes outstanding services and innovation to drive positive changes within the industry, achieving more together and making quantum leaps towards progress. This article was originally published by on Alexander García The Sociable.

15 Advanced Cybersecurity & Web 3.0 Executives

Exploring Developer Experience on ZKRUs: An In-Depth Analysis

This article presents one of the approaches to building a computer vision-driven car damage detection solution. We will make an introduction to the popular instance and semantic segmentation architectures Mask R-CNN and U-Net and provide insights deriving from our attempt to train and test ML models for car damage detection using these architectures. Then, we will share datasets used for training and testing. After that, we will explain in depth how our computer vision solution operates and demonstrate how a car damage model works using a demo. We also provide parts of the code of the models used to build a resulting solution. Overview of Mask R-CNN and U-Net Architectures and Performance Comparison Based on Observation We considered it reasonable to try both approaches to image segmentation that are suitable for the purposes of our project. This way, we get not only desirable outcomes but also receive meaningful insights into how instance and semantic segmentation architectures work on the particular car damage detection task. Here’s a brief into the technologies utilized: : Constructed on top of the Faster R-CNN object detection model, this deep learning architecture incorporates a segmentation component. It generates region proposals using a Region Proposal Network (RPN), which suggests regions of the image that are likely to contain objects. Subsequently, it performs object detection and segmentation by predicting class labels and providing bounding boxes and masks for each proposal. Instance segmentation architecture — Mask R-CNN : U-Net is a convolutional neural network architecture designed for image segmentation tasks. It comprises two pathways: the contracting path, which extracts context and decreases the input image size, and the expansive path, which facilitates precise localization and increases the feature maps’ size. This CNN can retrieve in-depth information about segmented objects while capturing the image’s context and overall structure. Semantic segmentation architecture — U-Net The complete procedure for training, testing, and comparing the models built with these algorithms is provided in the below sections. Datasets Used in Training and Testing There is a strong correlation between the volume and quality of data and the outcomes of training AI models. After all, a widely known concept which is garbage in, garbage out (GIGO), clearly indicates that nonsense input data produces nonsense output. In ideal circumstances, we would prepare extensive datasets of damaged vehicles with varying deteriorations, from diverse angles, with different lighting, etc., and conduct data annotation ourselves. But, for this trial project, we considered it reasonable to utilize precleaned and prepacked data from these publicly available datasets: Coco car damage detection dataset. Segme image dataset. We recommend using the mentioned datasets. Yet, it may already be possible to find newer, higher-quality collections. For business-level projects, such as implementing computer vision technology in an insurance claim processing workflow, we highly recommend the following approach to data preparation: Decide on the types of damages the ML model is intended to assist with the identification of which. Choose high-quality images from the database of past car inspection cases and build an extensive collection. Perform manual image labeling assisted with the usage of ML algorithms. Discover additional information on . how to get better datasets for your computer vision tasks The Essence of our Tried and Tested Computer Vision Solution for Car Damage Detection Let’s thoroughly get through our process of choosing, training, and testing semantic and instance segmentation models. It contains insights and explanations of the unique solutions we came up with. These can come of great usefulness to anyone willing to develop their ML computer vision model for car damage detection. Choosing Models We picked a PyTorch framework, and in particular, from a segmentation_models_pytorch library, we chose a U-Net model. We tested different backbones (pre-trained models that are used for feature extraction). Still, the best tradeoff between performance and accuracy we achieved was with an efficient net-b1 (Efficient net architecture) backbone that was pre-trained on the ImageNet dataset. Semantic segmentation architecture — U-Net. import segmentation_models_pytorch as smp


CLASSES = ['damage']


model_params = {"encoder_name": 'efficientnet-b1',
               "encoder_weights": 'imagenet',
               "activation": None,
               "encoder_depth": 5,
               "decoder_channels": [256, 128, 64, 32, 16]
               }
model = smp.Unet(classes=len(CLASSES), **model_params) We used a detectron2 library and tried different models from it. Mask-RCN_R_50_FPN_3x proved itself to be the best performing on car damage detection tasks. Instance segmentation architecture — Mask R-CNN. Training Models We fine-tuned the model on a pretty small dataset consisting of 59 train images and 11 images for the eval and final tests, respectively. We increased the dataset size by using different augmentation techniques, including random flipping, scaling, rotations, adding Gaussian noise, changing brightness blurring, etc. Semantic segmentation architecture — U-Net. For the loss function, we utilized different functions such as BCEWithLogitsLoss, DiceLoss, and LovaszHingeLoss. As a result, the best scores on the test dataset were achieved with the linear combination of two losses which were DiceLoss and BCEWithLogitsLoss. Combining two loss functions is a rather unconventional approach, which, nevertheless, provided the best outcomes in the case of car damage detection task. import torch.nn as nn
import torch.nn.functional as F


class DiceLoss(nn.Module):
   def __init__(self, weight=None, size_average=True):
       super(DiceLoss, self).__init__()


   def forward(self, inputs, targets, smooth=1):
      
       #comment out if your model contains a sigmoid or equivalent activation layer
       inputs = F.sigmoid(inputs)      
       #flatten label and prediction tensors
       inputs = inputs.view(-1)
       targets = targets.view(-1)
      
       intersection = (inputs * targets).sum()                           
       dice = (2.*intersection + smooth)/(inputs.sum() + targets.sum() + smooth) 
      
       return 1 - dice criterion = DiceLoss()
criterion_2 = nn.BCEWithLogitsLoss()

# …
 
loss = criterion(output.squeeze(1), mask) + criterion_2(output.squeeze(1), mask) We trained this model using the AdamW optimizer and scheduling a learning rate via the OneCycleLR scheduler. Additionally, we adopted an early stopping algorithm to track a loss on an evaluation dataset. optimizer = torch.optim.AdamW(model.parameters(),
                              lr=params['max_lr'],
                              weight_decay=params['weight_decay'])

sched = torch.optim.lr_scheduler.OneCycleLR(optimizer,
                                            params['max_lr'],
                                            epochs=params['epoch'],
                                            steps_per_epoch=len(train_dataloader)) Again, here we fine-tuned the model on a pretty small dataset consisting of 59 train images and 11 images for the eval and final tests, respectively. For the detectron2, we used the built-in default loss with default optimizer. Instance segmentation architecture — Mask R-CNN. cfg = get_cfg()
cfg.merge_from_file(model_zoo.get_config_file("COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml")) 
cfg.DATASETS.TRAIN = ("car_damage_dataset_train")
cfg.DATASETS.TEST = ("car_damage_dataset_val",) 
cfg.DATALOADER.NUM_WORKERS = 4 
cfg.MODEL.WEIGHTS = model_zoo.get_checkpoint_url("COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml") 
cfg.SOLVER.IMS_PER_BATCH = 4 
cfg.SOLVER.BASE_LR = 0.001 
cfg.SOLVER.WARMUP_ITERS = 700 
cfg.SOLVER.MAX_ITER = 700 
cfg.SOLVER.STEPS = (600, 800) 
cfg.SOLVER.GAMMA = 0.05 
cfg.MODEL.ROI_HEADS.BATCH_SIZE_PER_IMAGE = 128 
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 1 
cfg.MODEL.RETINANET.NUM_CLASSES = 1 
cfg.TEST.EVAL_PERIOD = 600 
cfg.MODEL.DEVICE = device 
trainer = CocoTrainer(cfg)
trainer.resume_or_load(resume=False) 
trainer.train() The neat part of how detectron2 makes a prediction is that it predicts multiple different instances for one damaged part. It can be seen in the below example. We tried to improve the outcomes by implementing post-processing, where we merge together all provided instances that overlap more than a certain threshold. The results of our attempt can be seen in the below example. Testing and Comparing Models ML engineers from Intelliarts AI considered the Mean Intersection-Over-Union (mIoU) metric as our primary choice for evaluating the outcomes of the U-net and Mask R-CNN model comparison. Jumping ahead, we also utilized the dice coefficiency metric, but the results were less representative compared to the ones with mIoU. So, with the U-net model, we achieved the mIoU equal to approximately 0.398. With detectron2, the mIoU values ranged from approximately 0.249 to 0.3654 depending on the threshold, which is significantly worse than the results with the U-net model. We assumed that this is because U-net considers the image as one instance while semantic segmentation tries to predict different parts of the same damage as different damages, which by the way, we tried to solve by implementing post-processing. As for the visual representation of the results, we may also compare the predictions of both models with the ground truth masks, i.e., what actually mask on the test dataset looks like, provided in the image below. We can observe that U-net tends to underestimate damages, and the detectron2, on the contractor, is more inclined to overestimate damages, masking larger segments and in a higher quantity. The End Result: Two-Model Computer Vision Solution As stated, we decided to move on with the U-net model. Not only was it justified by the better results than the U-net model showed, but also by the fact that underestimating damage is better than overestimating in our case. The reason is that we try to automate insurance claim processing by helping inspectors make assessments faster. However, we also are to minimize business risks of insurance agents that include overestimating damage and, respectively, overpaying the car owners. In the end, we built and trained two separate U-net-based models, one of which is intended for car damage detection and the other one for car part detection only. Allows for identifying the damaged area. It also provides a rough estimation of the magnitude of damage. Model 1 for car damage detection. Allows for identifying distinct car parts. It also provides the name of an affected part. Model 2 for car part detection. When a user enters an image of a damaged car, the resulting solution indicates the damage and identifies the affected car part separately. If they intersect, the solution outputs something like “left door — dent.” The results are then compared to similar cases in a prepared image database with repair cost estimates. You can observe the working principles of both models in the image below. Demo Overview We prepared a demo that shows the mechanism of how the detailed ML models work. The example is not as high-performing as the real-world model is, but it provides an appropriate insight into the implementation of the technology. You can check the below recording of the demo: Link Or you can give a car damage detection demo a try yourself here: https://huggingface.co/spaces/intelliarts/Car_parts_damage_detection Repo link https://gitlab.com/r-d-machine-learning/researches/Car-damage-detection.git Final Take This ML solution is a non-trivial combination of distinct models, one for detecting car damage and another for identifying car parts. Combined together, they allow for obtaining detailed information for car inspection. From our research, it’s evident that semantic segmentation architecture, particularly U-net, is probably the optimal algorithm for solving image processing tasks that are related to car damage detection. However, it’s crucial to understand that the technology is intended for dealing with large batches of minor vehicle assessment cases. We position it as a useful instrument that human inspectors should master. Therefore, we believe that insurance firms will find our solution useful for claims processing automation. Finally, our engineering team regards integration of the described technology in business processes as another step towards digital transformation.

Solving Car Damage Detection Task By Using a Two-Model Computer Vision Solution

Smart Contract Security in Solidity

সলিডিটিতে স্মার্ট কন্ট্রাক্ট সিকিউরিটি

Intelligente Vertragssicherheit in Solidität

Una encuesta de 2023 sobre seguridad de contratos inteligentes para la solidez

Une enquête 2023 sur la sécurité des contrats intelligents pour la solidité

सॉलिडिटी के लिए स्मार्ट कॉन्ट्रैक्ट सिक्योरिटी पर 2023 का सर्वे

Solidity のスマート コントラクト セキュリティに関する 2023 年の調査

Solidity의 스마트 계약 보안

Uma pesquisa de 2023 sobre segurança de contratos inteligentes para solidez

Sağlamlıkta Akıllı Sözleşme Güvenliği

Khảo sát năm 2023 về bảo mật hợp đồng thông minh cho sự vững chắc

2023 年 Solidity 智能合约安全性调查

MAXA AI

maxa.ai

-- employees

Since n.d.

COMPANY RANKING

MAXA AI

EVERGREEN INDEX #425

Representing this company? Click here to claim and customize this page!

Maxa AI WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

MAXA AI

maxa.ai

-- employees

Since n.d.

COMPANY RANKING

MAXA AI

EVERGREEN INDEX #425

RELATED COMPANIESRANK

STORIES ABOUT MAXA AI

Representing this company? Click here to claim and customize this page!

Maxa AI WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!