Rail lines are set for disruption in the week following the first May bank holiday as train drivers at 16 rail companies strike on different days.

Train strikes in May 2024: Full list of dates and lines affected | UK News | Sky News

Discover how the upcoming ASLEF strike on May 9 will impact Northern England&#x27;s train services and global travel, causing significant disruptions.

Everything you need to know about Northern England train strike - Travel And Tour World

There will be no Northern services across the North of England next Thursday (May 9) due to the latest industrial action by ASLEF.

No Northern train services across the North of England next Thursday due to the latest strike by ASLEF

The Calcutta High Court on Thursday disposed of a PIL seeking an extension of metro timings in Kolkata by directing the railway authorities to consider the petitioner&#x27;s prayer for an extension of the...

Plea In Calcutta High Court Seeks Extension Of Metro Service Timings, Says Last Train's Timing Being Earlier Than All Major Cities Causes Inconvenience To Commuters

A milestone achievement occurred as Punjab&#x27;s Kapurthala factory unveiled the inaugural Vande Bharat Metro coaches, heralding a new era in Indian rail development. Scheduled for July testing, these trains promise efficient travel with a 100-250 km range and expandable configuration.

rapid rail: First Vande Bharat Metro train set rolls out: Watch first-look video here - The Economic Times

Mumbai&#x27;s Harbour line trains see 30-40 min delays from speed restrictions after CSMT derailments. CR&#x27;s 10 km/h limit, aiming for 15 km/h, faces critic

Mumbai Harbour Line trains running late after two derailments | Mumbai News - Times of India

Mumbais Harbour Line Near CSMT faced service disruptions because there was an imposition of speed restriction at 10 kmph for local trains at the crossover point

Mumbai's Harbour Line Near CSMT Faces Disruption Following Train Derailments; Details | Railways News | Zee News

The main reason behind the rescheduling of the trains is due to non-interlocking work for the commissioning of doubling at Tapa-Rampura Phul stations

Multiple Trains Cancelled, Diverted By Northern Railways Due To Doubling Work, Check New Schedule - News18

Northern says nine passengers were on board and there are no reports of any &quot;serious&quot; injuries.

'No injuries' after Redcar level crossing crash - BBC News

The crash took place at the level crossing near Redcar Central Station

Train track drama after 'crossing failure' crash in Redcar sees drones drafted in for police probe - Teesside Live

TrainFES

trainfes

In 2021, hosted the first-ever awards!! HackerNoon Startup of the Year The goal of this campaign was to give recognition to the hardworking startups who pulled through the pandemic -- growing and thriving (despite all of the challenges). More than 44,000 startups were nominated all across the globe! They were selected from over 4,553 cities, with a total number of * cough * “4 billion” votes. Insane, right? To see who is a winner in each respective city, visit and pick a city. The name at the top of the list is the winner. (Awards, with compromised reviews, are under review.) startups.hackernoon.com The will announce the winners per region in their respective cities. Startups Winners Announcement Series Apparently, South America startups did incredibly well in that regard. With so many, until now, uncrowned jewels, it’s time we present to you... * drumroll* 🥳 The South America Startups of the Year Winners! Santiago, Chile Chile is known for being the most stable country to do business in Latin America, ranking 13th in The Economist’s Business Environment Index. Moreover, according to the Global Innovation Index, Chile ranks first in Latin America for innovation and is recognised as Chilean Silicon Valley with Santiago being the central city. The winner is…. QCObjects It is the 1st world class open-source framework made in Latin-America. São Paulo, Brazil And the winner of Brazil’s vibrant financial center is... Innvo Digital Their mission is to create the best ecosystem for a technology professional to work and undertake while sharing it with our customers and society. São Carlos The absolute leader of Sao Carlos seems to be: Predify Predify is an online software for optimizing pricing proccess and improving profits with AI and Big Data Analytics. Belo Horizonte, Brazil With a more than noble goal, Belo Horizone’s winner is: CM Tecnologia CM Tecnologia is an online scheduling system for healthcare providers in Brazil. Porto Alegre, Brazil With the focus on data protection, Porto Alegre, Brazil’s winner is…. Privacy Tools Privacy compliance platform, based on AI/Blockchain, which helps global companies to keep compliant with the data protection requirements. Brasil, Brazil With the focus on EdTech, Brasil is bringing us: Awari Awari is a YC-W21 company focusing on education for the Tech Industry. The company is growing at a fast pace and has gathered experts from all leading tech companies into a full education experience to prepare the next cohort of tech talent. Quito The capital of Ecuador, and also the closest capital to the actual equator, as the winner brings us: Agrodemanda The most comprehensive agricultural directory in Ecuador. Rio De Janeiro, Brazil Rio de Janeiro has more than 300 startups and is ranked 3rd in Brazil in terms of startup ecosystem development. The city is famous for the startups in the spheres of education, energy & environment and foodtech. Their winner is…. Bybrand Bybrand is the simplest way to save time and avoid frustration when managing your employee's email signatures. Buenos Aires, Argentina Buenos Aires, Argentina is not only the cosmopolitan capital with a vibrant history. It’s also Latin America’s most important port. Which makes their winner actually a very logical choice. Please meet: Liftit Liftit is the leading technology platform to automate and execute last-mile truck deliveries across Latin America. Curitiba, Brazil With a more noble mission and sustainability in mind, Curitiba, Brazil, as the winner brings us... Discrepante is the marketplace for a more human and sustainable economy Discrepante.com Florianópolis, Brazil Would you really be surprised one of the Brazil’s winners is related to football training? We neither. Please meet... Fútbol Fit Club Football FitClub® is a fitness training firm that gives fitness training to improve like a professional football player. Medellín, Colombia And the first Colombian Startup Winner is is a data intelligence solution powered by a high quality team. Meet... LogComex Market intelligence for international trade in Latin America Montevideo, Uruguay The host of inaugural FIFA world cup of 1930 comes through when it comes to innovation even in 2021. I’d like to introduce you to: Drop Technologies First electronic hand antiperspirant device Lima, Peru Lima, Peru, has in 2021 delivered some amazing results when it comes to blockchain technology. Their winner is very well deserved: RSK Labs RSK is the first open-source smart contract platform São Carlos Do Pinhal Well this is a unique case. The absolute unanimous winner of São Carlos Do Pinhal is: brain4care brain4care solution enables monitoring intracranial compliance noninvasively anywhere Blumenau, Brazil We’re sensing a theme here. Another stellar EdTech winner is coming from Blumenau, Brazil. Please meet... Uniasselvi Uniasselvi is a Brazilian distance-learning company that offers more than 60 undergraduate and postgraduate courses, in person and distance. Bogotá, Colombia Startups of Bogotá, Colombia definitely showed they’re not to be underestimated. With a very noble mission, let me introduce you to: Risü Risü is an online platform that links online consumption to donation to social institutions. Campina, Brazil If there’s one typo of business we’d single out as taking the biggest hit out of the pandemic, it would have to be small businesses. Thankfully, the winner of Campina, Brazil, operates with small businesses in mind Superlógica Tecnologias Ltda. Superlogica is a management system company created for small service businesses. Cerqueira César, Brazil Swooping half the points from the entire City, this year’s winners of Cerqueira César are InfoPrice INFOPRICE is a business intelligence company focused on physical retail pricing. Governador Valadares, Brazil The same can be said for the winners here. Let’s hear it for: Grupo KPG Saas, Payment, Manufacture, Pharmaceutical, E-commerce, Marketplace, Logistics Caxias Do Sul Known for wine making and Italian cultural influence, Caxias Do Sul brings as an interesting winner. Studio 360 Studio 360 is a proptech focused on easing the purchase experience for real state companies, brokers, and costumers. Sorocaba From Sorocaba, we have an interesting concept coming. Looking forward to seeing it further develop. Let’s meet: Quinto Quinto is a social network that represents the collective opinion of society—to help democracy evolve through technology Viña Del Mar To be a venue-oriented business and survive through the pandemic definitely deserves some recognition. In the name of that, let’s meet: SonicPollen SonicPollen, based in Chile, is a web-based booking platform that connects bands with venues and organizers. Lages Learn more about high-growth startups in Lages, South America. and the winner is…. Qualiall Qualiall develops technologies for training in operational processes such as work security, equipment maintenance, and more. Córdoba, Argentina Another gem from Argentina brings us unique agricultural solutions. Please meet: Kilimo Kilimo is a Big Data solution for water management in agriculture. São José Dos Campos, Brazil Another great agricultural solution comes this time from Brasil. Meet São José Dos Campos Startup winner: TerraMagna TerraMagna is a fintech company that focuses on risk mitigation for agribusinesses. São Leopoldo, Brazil Event-oriented businesses that have survived in the previous years should definitely bottle their perseverance and start selling it. One of those is definitely this winner: Wedy Wedy offers a one-stop-social marketplace for weddings. Pelotas, Brazil Medical scheduling platforms are definitely something we need to see more of. So glad Pelotos winner comes from this particular niche. Fácil Consulta Fácil Consulta is a booking platform that allows people to schedule an appointment with a Specialist Physician in their area. Santa Maria Could you believe how versatile the startups of South America are when we are only now coming across a game development studio? Let’s meet them Imgnation Studios Imgnation Studios is a game development studio that focuses on virtual reality. Asunción, Paraguay A FinTech solution coming from Paraguay. Meet Asunción Startup of the Year winner Vision Banco Vision Banco is a solid and growing financial institution serving the financial needs of the micro, small and medium enterprises. Caracas Venezualan food delivery app that seemed to have your heart. Meet Caracas Startup of the Year winner Yummy Inc. First Food Delivery App in Venezuela Recife A suitable combo between gaming and education, Recife Startup of the Year winner brings revolutionary solutions. Let’s meet them. Joy Street Joy Street offers develops video games focused on educational needs. Rosario, Argentina The winner of Rosario Startup of the Year award for 2021 is... Gbot Gbot is an autonomous robot that controls weeds without using agrochemicals. We also collect data in fields and generate precision maps. Vila Olímpia, Brazil The winner of Vila Olímpia Startup of the Year award for 2021 is... Seekr Seekr operates in the Information Technology sector, providing tools for managing and monitoring of brands. Barueri The winner of Barueri Startup of the Year award for 2021 is.... Glic Brazilian mhealth (web/app) platform for diabetes management Brasília The winner of Brasília Startup of the Year award for 2021 is... Ticies Ticies is a social network that helps people plan where to go with their friends, relatives and acquaintances, in all cities of the world. Juiz De Fora The winner of Juiz De Fora Startup of the Year award for 2021 is... Qranio Qranio is a gamified mobile learning platform for enterprises that rewards learning and stimulates users to engage with educational content. La Paz The winner of La Paz Startup of the Year award for 2021 is... Jukumari Advice 360 photo based market place for responsible tour operators Temuco The winner of Temuco Startup of the Year award for 2021 is... GoFeels GoFeels is the first independent hospitality network in the world. We promote collaborative work among hotels around the world. Vitacura The winner of Vitacura Startup of the Year award for 2021 is... MeatMe MeatMe delivers meats and accessories to have a culinary experience. Barranquilla The winner of Barranquilla Startup of the Year award for 2021 is... Mandaos Mandaos is a digital platform that connects companies and people with a network of independent messengers Ceará The winner of Ceará Startup of the Year award for 2021 is... Caroá Caroá is a cooperative that sells 34 varieties of vegetables, organic fruits, and vegetables. Chapecó The winner of Chapecó Startup of the Year award for 2021 is... HortiAgro HortiAgro is a seed and fertilizer market. Chile Chico The winner of Chile Chico Startup of the Year award for 2021 is... Gleengo Gleengo connects children with their favorite toys, through a tiny device and an application based on text-to-voice. Goiânia The winner of Goiânia Startup of the Year award for 2021 is... Auvo Auvo is an external teams management software. Itá The winner of Itá Startup of the Year award for 2021 is... TISSUELAB Tissuelab is a research and development center focused on biotechnology, and systems for reconstructive musculoskeletal surgery. Jardim Paulista The winner of Jardim Paulista Startup of the Year award for 2021 is... DiviCom DiviCom is a digital video networking company that designs, integrates, and markets complete systems. João Pessoa The winner of João Pessoa Startup of the Year award for 2021 is... YupiPlay YupiPlay develops and offers an educational gaming platform. La Reina The winner of La Reina Startup of the Year award for 2021 is... INTI-TECH INTI-TECH is the fastest growing renewable energy in Chile and the world's leading unconventional source. Leme The winner of Leme Startup of the Year award for 2021 is... EcoStart EcoStart focuses on manufacturing sustainable electric bicycles, fans, and accessories. Loja The winner of Loja Startup of the Year award for 2021 is... Enerwi Corporation Keeping people connected and secured through wireless charging novel solutions Manizales The winner of Manizales Startup of the Year award for 2021 is... Porter Metrics Porter Metrics has developed an automated solution for generating social media reports in minutes. Nova Lima The winner of Nova Lima Startup of the Year award for 2021 is... Biomimetic Solutions Biomimetic Solutions is an innovative tissue engineering company. Paraná The winner of Paraná Startup of the Year award for 2021 is... Laura Laura is a provider of artificial intelligence and predictive analytics platform for hospitala and clinics to prevent damage & reduce costs. Salvador The winner of Salvador Startup of the Year award for 2021 is... Sanar Sanar is a platform that focuses on being the doctor's guide and support, offering specialized content that helps in their development. Santa Catarina The winner of Santa Catarina Startup of the Year award for 2021 is... DrPOCKET Sunchales The winner of Sunchales Startup of the Year award for 2021 is... Eolo Pharma Eolo Pharma focuses on developing drugs targeting the molecular and cellular basis of inflammation-related diseases Valparaiso The winner of Valparaiso Startup of the Year award for 2021 is... Invitta Invitta transforms plastic waste into a high-performance construction material for manufacturing sustainable urban furniture. Vila Velha The winner of Vila Velha Startup of the Year award for 2021 is... OneGrid OneGrid is a developer of hardware and software solutions focused on energy management. Indaiatuba The winner of Indaiatuba Startup of the Year award for 2021 is... DWUL Energy Renewable Energy Campinas, Brazil The winner of Campinas Startup of the Year award for 2021 is... Ativy Ativy is an IT company specialized in cloud computing, data centers, IT outsourcing, service desk, and IT management. Las Condes, Chile The winner of Las Condes Startup of the Year award for 2021 is... # MuvPass Subscription services that allows it's users unlimited access to fitness classes in the best gyms and studios in Latin America. Pinheiros, Brazil The winner of Pinheiros Startup of the Year award for 2021 is... GuiaBolso GuiaBolso is a personal finance platform that empowers people to take control of financial well-being. Rio, Brazil The winner of Rio Startup of the Year award for 2021 is... EmCasa EmCasa is a vertically-integrated platform that helps customers buy and sell a home in Latin America, starting in Brazil. Guayaquil The winner of Guayaquil Startup of the Year award for 2021 is... Sabroso Online platform for LatAm's enterprises to order low-cost meals Jundiaí The winner of Jundiaí Startup of the Year award for 2021 is... Sintecsys Sintecsys offers a fast and effective detection, reducing losses and costs with fire brigades. Minas Gerais The winner of Minas Gerais Startup of the Year award for 2021 is... Rizoflora Biotechnology Rizoflora Biotechnology is a producer of biological product designed for nematodes control. São Caetano Do Sul The winner of São Caetano Do Sul Startup of the Year award for 2021 is... Wevo 1 iPaaS in Latam - Enabling companies to be truly digital companies with APIs, systems and data integrated into a single platform Canoas The winner of Canoas Startup of the Year award for 2021 is... ContaÁgil ContaÁgil facilitates and reduces bureaucracy in accounting routines of companies. Chile The winner of Chile Startup of the Year award for 2021 is... Unight Unight is the future of nightlife social media. Israelândia The winner of Israelândia Startup of the Year award for 2021 is... Matchpool Matchpool is a platform that creates human connections. La Plata The winner of La Plata Startup of the Year award for 2021 is... Pixivus Pixivus offers a community-based application that enables its users to collect, store, and share pictures of events. Novo Hamburgo The winner of Novo Hamburgo Startup of the Year award for 2021 is... Devorando Devorando offers an online food ordering system and menus available in restaurants based in Brazil. Pernambuco The winner of Pernambuco Startup of the Year award for 2021 is... Redu Redu is a distant learning platform which is built on a social networking structure. Providencia The winner of Providencia Startup of the Year award for 2021 is... # TrainFES TrainFES is a medical Devices for the rehabilitation of people with paralysis. Ribeirão Prêto The winner of Ribeirão Prêto Startup of the Year award for 2021 is... AllyHub Ally is a sales distribution channel technology that connects Schools and Service Providers with Education Agencies. Santa Cruz De La Sierra The winner of Santa Cruz De La Sierra Startup of the Year award for 2021 is... Mobi Latam Mobi Latam is a clean energy startup that focuses on transportation systems based on electric mobility. Santana De Parnaíba The winner of Santana De Parnaíba Startup of the Year award for 2021 is... MercadoLivre.com is an e-commerce technology company providing solutions to individuals and companies buying, selling, and more online. MercandoLivre.com More Information If you’re a winner, please answer ! If you’re a reader, participant, just a curious individual, consider answering one of our interviews . this interview here Read the FAQs about or . Startups here here To see all of the winners you may visit , pick any city and view the name that appears at the top. this site If you’re interested in partnering with us, we invite you to look through some of these or reach out to our business department. unique sponsorship options Perhaps creating your very own tickles your fancy or maybe you have a whole new idea for how we can partner! writing contest Thank you to all of our sponsors! We truly appreciate your contribution to the tech world and thus the future! The Startups Awards are sponsored by: , , , , , , , , , , Blockster Brex FreeCodeCamp LaunchDarkly Coil Udacity Udemy Vercel Asia Pacific Transgender Network Bybit Everscale Network This post was written/compiled by Shreya Kundu and Katarina Andrejević! If you have any questions or concerns or would like to report a discrepancy please email ellen@hackernoon.com.

Meet the Winners of Startups in South America

CEH v13 certification equips cybersecurity professionals with essential AI skills to prepare them for a new era of cybercrime TAMPA, FLORIDA September 23, 2024: EC-Council, a world leader in cybersecurity certification, education, training, today announced the launch of Certified Ethical Hacker CEH v13 with AI- powered capabilities, a first of its kind certification in cybersecurity. Long recognized as the #1 ethical hacking certification in the world, the new and improved groundbreaking program integrates artificial intelligence (AI) learning into ethical hacking practices, giving professionals advanced skills and tools to battle global cybercriminals. CEH v13 provides in-depth training by integrating AI into all five phases of ethical hacking, from reconnaissance and scanning to gaining access, maintaining access, and covering tracks to help bridge the AI Chasm. Through the course, learners will be equipped to leverage AI to enhance their hacking techniques and hack AI systems to automate their ethical hacking tasks while driving up to 40% efficiency in cyber defense and doubling their productivity. “AI has already begun to radically transform the cybersecurity landscape,” said Jay Bavisi, Group President of EC-Council. “The launch of CEH v13 represents a major leap forward in cybersecurity education, with AI taking center stage in the curriculum. As AI tools for both attackers and defenders continue to proliferate, building AI skills is already becoming a key factor for employers looking to hire and retain cybersecurity talent.” Inspired by the findings of EC-Council’s 2024 CEH Threat Report, the CEH v13 program is designed to leverage a unique four-phase learning framework based on the principles of Learn, Certify, Engage, and Compete, offering current and prospective cybersecurity professionals comprehensive knowledge of AI in cybersecurity through courses and hands-on learning experiences. The program includes 221 labs, attack vectors, hacking tools, a lab environment that simulates real-time environment, and a Capture the Flag global hacking competition that will run all year long. The year-long competition allows individuals to continually sharpen their skills as needed while keeping them prepared to operate and defend in an increasingly complex cyber environment. As part of the CEHv13 program, trainees will participate in 40 hours of rigorous training though instructional practice and education, followed by a yearlong, month-to-month competition across various technologies to help learners obtain world class operational skills in ethical hacking. Course participants will experience more than 550 attack techniques and 4000+ commercial-grade security tools. Key highlights of CEH v13 include World's first ethical hacking program harnessing the power of AI Master AI-driven skills and Learn Hacking AI systems: Learn to leverage AI for enhanced hacking techniques and how to identify and exploit vulnerabilities in AI systems. Core domains of Cybersecurity: CEH strengthens your core domains of cybersecurity with a practical learning approach, which enables CEH to map across 45+ job roles over many industries. Focus on the OWASP Top 10 AI Attacks: Gain expertise in mitigating threats like prompt injection, insecure output handling, training data poisoning, and more. 221 hands-on labs and 4000+ commercial-grade hacking tools: Develop practical skills in a real-time simulated environment. Knowledge-Based and Practical Exam: Validate your skills with both a 4-hour, knowledge-based exam and a 6-hour, fully practical exam as part of the CEH Master program. For more information about the Certified Ethical Hacker CEH v13, visit https://www.eccouncil.org/cehv13ai/. ABOUT EC-COUNCIL EC-Council invented the Certified Ethical Hacker. Founded in 2001 in response to 9/11, EC-Council's mission is to provide the training and certifications apprentice and experienced cybersecurity professionals need to keep corporations, government agencies, and others who employ them safe from attack. Best known for its Certified Ethical Hacker program, EC-Council today offers 200 different pieces of training, certificates, and degrees in everything from Computer Forensic Investigation and Security Analysis to Threat Intelligence and Information Security. An ISO/IEC 17024 Accredited Organization recognized under the U.S. Defense Department Directive 8140/8570 and many other authoritative cybersecurity bodies worldwide, EC-Council has certified over 350,000 professionals across the globe. EC-Council is the gold standard in cybersecurity education and certification, trusted by seven of the Fortune 10, half of the Fortune 100, and the intelligence communities of 150 nations. A truly global organization with a driving belief in bringing diversity, equity, and inclusion to the modern cybersecurity workforce, EC-Council maintains 11 offices in the US, the UK, India, Malaysia, Singapore, and Indonesia. The company can be reached online at https://www.eccouncil.org/
Press Contact: press@eccouncil.org CEH v13 certification equips cybersecurity professionals with essential AI skills to prepare them for a new era of cybercrime CEH v13 certification equips cybersecurity professionals with essential AI skills to prepare them for a new era of cybercrime TAMPA, FLORIDA September 23, 2024: EC-Council, a world leader in cybersecurity certification, education, training, today announced the launch of Certified Ethical Hacker CEH v13 with AI- powered capabilities, a first of its kind certification in cybersecurity. Long recognized as the #1 ethical hacking certification in the world, the new and improved groundbreaking program integrates artificial intelligence (AI) learning into ethical hacking practices, giving professionals advanced skills and tools to battle global cybercriminals. TAMPA, FLORIDA September 23, 2024: CEH v13 provides in-depth training by integrating AI into all five phases of ethical hacking, from reconnaissance and scanning to gaining access, maintaining access, and covering tracks to help bridge the AI Chasm. Through the course, learners will be equipped to leverage AI to enhance their hacking techniques and hack AI systems to automate their ethical hacking tasks while driving up to 40% efficiency in cyber defense and doubling their productivity. “AI has already begun to radically transform the cybersecurity landscape,” said Jay Bavisi, Group President of EC-Council. “The launch of CEH v13 represents a major leap forward in cybersecurity education, with AI taking center stage in the curriculum. As AI tools for both attackers and defenders continue to proliferate, building AI skills is already becoming a key factor for employers looking to hire and retain cybersecurity talent.” “AI has already begun to radically transform the cybersecurity landscape,” said Jay Bavisi, Group President of EC-Council. “The launch of CEH v13 represents a major leap forward in cybersecurity education, with AI taking center stage in the curriculum. As AI tools for both attackers and defenders continue to proliferate, building AI skills is already becoming a key factor for employers looking to hire and retain cybersecurity talent.” Inspired by the findings of EC-Council’s 2024 CEH Threat Report , the CEH v13 program is designed to leverage a unique four-phase learning framework based on the principles of Learn, Certify, Engage, and Compete, offering current and prospective cybersecurity professionals comprehensive knowledge of AI in cybersecurity through courses and hands-on learning experiences. The program includes 221 labs, attack vectors, hacking tools, a lab environment that simulates real-time environment, and a Capture the Flag global hacking competition that will run all year long. The year-long competition allows individuals to continually sharpen their skills as needed while keeping them prepared to operate and defend in an increasingly complex cyber environment. 2024 CEH Threat Report As part of the CEHv13 program, trainees will participate in 40 hours of rigorous training though instructional practice and education, followed by a yearlong, month-to-month competition across various technologies to help learners obtain world class operational skills in ethical hacking. Course participants will experience more than 550 attack techniques and 4000+ commercial-grade security tools. Key highlights of CEH v13 include World's first ethical hacking program harnessing the power of AI Master AI-driven skills and Learn Hacking AI systems: Learn to leverage AI for enhanced hacking techniques and how to identify and exploit vulnerabilities in AI systems. Core domains of Cybersecurity: CEH strengthens your core domains of cybersecurity with a practical learning approach, which enables CEH to map across 45+ job roles over many industries. Focus on the OWASP Top 10 AI Attacks: Gain expertise in mitigating threats like prompt injection, insecure output handling, training data poisoning, and more. 221 hands-on labs and 4000+ commercial-grade hacking tools: Develop practical skills in a real-time simulated environment. Knowledge-Based and Practical Exam: Validate your skills with both a 4-hour, knowledge-based exam and a 6-hour, fully practical exam as part of the CEH Master program. For more information about the Certified Ethical Hacker CEH v13, visit https://www.eccouncil.org/cehv13ai/ . https://www.eccouncil.org/cehv13ai/ ABOUT EC-COUNCIL EC-Council invented the Certified Ethical Hacker. Founded in 2001 in response to 9/11, EC-Council's mission is to provide the training and certifications apprentice and experienced cybersecurity professionals need to keep corporations, government agencies, and others who employ them safe from attack. Best known for its Certified Ethical Hacker program, EC-Council today offers 200 different pieces of training, certificates, and degrees in everything from Computer Forensic Investigation and Security Analysis to Threat Intelligence and Information Security. An ISO/IEC 17024 Accredited Organization recognized under the U.S. Defense Department Directive 8140/8570 and many other authoritative cybersecurity bodies worldwide, EC-Council has certified over 350,000 professionals across the globe. EC-Council is the gold standard in cybersecurity education and certification, trusted by seven of the Fortune 10, half of the Fortune 100, and the intelligence communities of 150 nations. A truly global organization with a driving belief in bringing diversity, equity, and inclusion to the modern cybersecurity workforce, EC-Council maintains 11 offices in the US, the UK, India, Malaysia, Singapore, and Indonesia. The company can be reached online at https://www.eccouncil.org/ Press Contact: press@eccouncil.org https://www.eccouncil.org/ press@eccouncil.org

EC-Council Introduces AI-Powered Ethical Hacking against Cybercrime

EC-Council በሳይበር ወንጀል ላይ AI-Powered Ethical Hacking አስተዋወቀ

EC-Consejo ukax Hacking Ético AI-Powered ukax Ciberdelincuencia ukar uñt’ayi

El Consejo de la CE presenta un hackeo ético basado en inteligencia artificial contra el cibercrimen

Le Conseil de l'UE présente le piratage éthique basé sur l'IA contre la cybercriminalité

EC評議会、サイバー犯罪対策としてAIを活用した倫理的ハッキングを導入

EC-Зөвлөл цахим гэмт хэргийн эсрэг хиймэл оюун ухаанаар ажилладаг ёс зүйн хакердалтыг нэвтрүүллээ.

Совет ЕС представляет этичную хакерскую атаку на основе искусственного интеллекта против киберпреступности

EC-කවුන්සිලය සයිබර් අපරාධවලට එරෙහිව AI-බලවත් සදාචාරාත්මක හැකිං හඳුන්වා දෙයි

EC-Council Inaanzisha Udukuzi wa Kimaadili unaoendeshwa na AI dhidi ya Uhalifu wa Mtandao

EC-کونسل نے سائبر کرائم کے خلاف AI سے چلنے والی اخلاقی ہیکنگ متعارف کرائی

EC-Council giới thiệu công nghệ hack đạo đức hỗ trợ bởi AI chống lại tội phạm mạng

I-EC-Council Yethula I-AI-Powered Ethical Hacking ngokumelene nobugebengu be-Cybercrime

Authors:
(1) Corby Rosset, Microsoft Research and Correspondence to corbyrosset@microsoft.com;
(2) Ching-An Cheng, Microsoft Research;
(3) Arindam Mitra, Microsoft Research;
(4) Michael Santacroce, Microsoft Research;
(5) Ahmed Awadallah, Microsoft Research and Correspondence to hassanam@microsoft.com;
(6) Tengyang Xie, Microsoft Research and Correspondence to tengyangxie@microsoft.com. Table of Links Abstract and 1 Introduction 2 Preliminaries 2.1 RLHF Based on Reward Models 2.2 RLHF with General Preferences 3 Direct Nash Optimization and 3.1 Derivation of Algorithm 1 3.2 Theoretical Analysis 4 Practical Algorithm – Iterative Contrastive Self-Improvement 5 Experiments and 5.1 Experimental Setup 5.2 Results and Analysis 6 Related Work 7 Conclusion and References Appendix A Extension to Regularized Preferences B Detailed Proofs C Additional Experimental Details 5 Experiments Algorithm 2 is chosen for its efficiency and simplicity from an implementation standpoint (in this section, we will use DNO to denote Algorithm 2 or DNO-Prct for simplicity). Once the input dataset {xi ∈ X } is chosen, each iteration of DNO proceeds in three phrases: sampling outputs from the current policy, annotating outputs for preference pair generation, and then training the next policy with the new training pairs. Iteration 0 is defined to start by sampling from the initial SFT model to produce training data for iteration 1. 5.1 Experimental Setup Every experiment except one in this study solely uses UltraFeedback. The exception is one “scaled up” experiment with about 10x more data sourced from a mixture of datasets aggregated including Anthropic HH (Bai et al., 2022a), UltraChat (Ding et al., 2023), MetaMathQA (Yu et al., 2023), EvolInstruct (Xu et al., 2023a), UltraFeedback (Cui et al., 2023) and Orca-2 (Mitra et al., 2023). Note that we only use the input prompts for these datasets and collect a GPT-4-Turbo responses for all 600k of these input prompts. Sampling from the Policy: At the end of training, we sample 5 outputs from the resulting student policy using top p sampling with p = 0.95 and temperature 0.7. Several works have shown the benefit of sampling and comparing multiple diverse outputs from the policy (Yuan et al., 2023a; Mitra et al., 2024; Liu et al., 2024b; Dong et al., 2023; Wang et al., 2022). We implement a simple defect detection system which flags any sample that has a high amount of repeated n-grams as automatic negative. Preference Annotation: We use GPT-4-Turbo “as a judge” to label preferences among the 5 policy samples and 1 gold sample (which is also GPT-4-Turbo) as shown in Figure 3. This prompt contains a few minor modifications from the that used in (Yuan et al., 2024). It implements an additive scoring framework on a 6-point scale where a score of 6 represents the highest quality answer according to certain dimensions like “correctness”, “expert knowledge”, “conciseness” etc. By following this rubric, GPT-4 acting as an annotator represents a best-effort general preference model because it compares multiple candidate responses side-by-side in the context window, and stratifies them along meaningful dimensions of quality. Training Pair Construction: Adhering to Line 6 in Algorithm 2 implies that not all pairs are suitable for training. Firstly, we must enforce the positives to be high quality in an absolute sense, and secondly, the negatives are directionally worse by a large margin. On the 6 point annotation scale, only samples that score a 5 or 6 are allowed to be positives. From the positives that meet this criteria, if any, we then construct all pairs such that the negative is at least 2 points lower. If the positive happens to be from the student, we relax this constraint to 1 point margin since the GPT-4-Turbo teacher outputs rarely receive a score less than 5 (as shown by the average teacher score in Table 2). Additionally, we are motivated to preserve the preference behavior from previous iterations so that new policies do not inadvertently regress to past bad behavior. To enforce this, we incorporate an exponentially decaying proportion of prior iterations’ training pairs into the current iteration, i.e. we sample at most 30% of training pairs from iteration t − 1, 15% from t − 2, and so on. We do not re-inference outputs for those inputs from the most recent policy. Recall that previous iterations’ inputs are non-overlapping with the splits for other iterations. Training: To prevent overfitting, we train our batched on-policy methods for at most one epoch on newly constructed pairs. Our effective batch size is fixed to 64 for all experiments. Our learning rate, beta, and alpha are found with brief hyperparameter searches. For most experiments, the learning rate is 5E-5, beta is either 0.1 or 0.05, and alpha is 0.005. We found that at higher iterations, the learning rate needs to be lowered. In SFT (supervised fine-tuning) experiments, our learning rate is 5E-6 and we mask out loss for the inputs. We use the open-source TRL library’s implementation to run our experiments. Evaluation: Our primary goal is to train a policy that is comparable to the most powerful state-of-the-art langauge models. Hence, AlpacaEval 2.0 (Dubois et al., 2023) is an appropriate benchmark because it computes win-rate against GPT-4-Turbo in a head-to-head fashion on a dataset of 805 input prompts that is shown to correlate with human preferences (0.93 spearman correlation with Chatbot Arena). While it is known that auto-eval methods also correlate with spurious features such as length, a new version of AlpacaEval 2.0 corrects for this with a length-controlled win-rate that has an even higher spearman correlation (0.98) with Chatbot Arena [4]. We also evaluate on MT-Bench (Zheng et al., 2023) which allows the llm-as-a-judge to first explain its reasoning before providing a scalar score on 1-10 for the candidate response to a bank of 80 questions. One crucial difference between AlpacaEval 2.0 and MT Bench is that the former asks GPT-4-Turbo to predict which of two side-by-side responses humans would prefer, weighted by the logits to represent its uncertainty, whereas MT-Bench asks the model to first generate a justification and then output a score on 1-10, but it neither defines the ratings (e.g. how a 7 is different than a 5) nor accounts for uncertainty in the logits of the score. We also evaluate on the OpenLLM leaderboard (Beeching et al., 2023), which measures reasoning ability on downstream NLP tasks like coding and question answering by evaluating the accuracy of the multiple choice answer option with the highest logit. Since our training data is primarily instruction-following and not trained to output just the sole answer option, this benchmark is not the primary target of this study; nonetheless, DNO on instruction tuning tasks ought to show no regression on reasoning tasks. This paper is available on arxiv under CC BY 4.0 DEED license. [4] https://github.com/tatsu-lab/alpaca_eval Authors: (1) Corby Rosset, Microsoft Research and Correspondence to corbyrosset@microsoft.com; (2) Ching-An Cheng, Microsoft Research; (3) Arindam Mitra, Microsoft Research; (4) Michael Santacroce, Microsoft Research; (5) Ahmed Awadallah, Microsoft Research and Correspondence to hassanam@microsoft.com; (6) Tengyang Xie, Microsoft Research and Correspondence to tengyangxie@microsoft.com. Authors: Authors: (1) Corby Rosset, Microsoft Research and Correspondence to corbyrosset@microsoft.com; (2) Ching-An Cheng, Microsoft Research; (3) Arindam Mitra, Microsoft Research; (4) Michael Santacroce, Microsoft Research; (5) Ahmed Awadallah, Microsoft Research and Correspondence to hassanam@microsoft.com; (6) Tengyang Xie, Microsoft Research and Correspondence to tengyangxie@microsoft.com. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries 2 Preliminaries 2.1 RLHF Based on Reward Models 2.1 RLHF Based on Reward Models 2.2 RLHF with General Preferences 2.2 RLHF with General Preferences 3 Direct Nash Optimization and 3.1 Derivation of Algorithm 1 3 Direct Nash Optimization and 3.1 Derivation of Algorithm 1 3.2 Theoretical Analysis 3.2 Theoretical Analysis 4 Practical Algorithm – Iterative Contrastive Self-Improvement 4 Practical Algorithm – Iterative Contrastive Self-Improvement 5 Experiments and 5.1 Experimental Setup 5 Experiments and 5.1 Experimental Setup 5.2 Results and Analysis 5.2 Results and Analysis 6 Related Work 6 Related Work 7 Conclusion and References 7 Conclusion and References Appendix Appendix A Extension to Regularized Preferences A Extension to Regularized Preferences B Detailed Proofs B Detailed Proofs C Additional Experimental Details C Additional Experimental Details 5 Experiments Algorithm 2 is chosen for its efficiency and simplicity from an implementation standpoint (in this section, we will use DNO to denote Algorithm 2 or DNO-Prct for simplicity). Once the input dataset {xi ∈ X } is chosen, each iteration of DNO proceeds in three phrases: sampling outputs from the current policy, annotating outputs for preference pair generation, and then training the next policy with the new training pairs. Iteration 0 is defined to start by sampling from the initial SFT model to produce training data for iteration 1. 5.1 Experimental Setup Every experiment except one in this study solely uses UltraFeedback. The exception is one “scaled up” experiment with about 10x more data sourced from a mixture of datasets aggregated including Anthropic HH (Bai et al., 2022a), UltraChat (Ding et al., 2023), MetaMathQA (Yu et al., 2023), EvolInstruct (Xu et al., 2023a), UltraFeedback (Cui et al., 2023) and Orca-2 (Mitra et al., 2023). Note that we only use the input prompts for these datasets and collect a GPT-4-Turbo responses for all 600k of these input prompts. Sampling from the Policy: At the end of training, we sample 5 outputs from the resulting student policy using top p sampling with p = 0.95 and temperature 0.7. Several works have shown the benefit of sampling and comparing multiple diverse outputs from the policy (Yuan et al., 2023a; Mitra et al., 2024; Liu et al., 2024b; Dong et al., 2023; Wang et al., 2022). We implement a simple defect detection system which flags any sample that has a high amount of repeated n-grams as automatic negative. Sampling from the Policy: Preference Annotation: We use GPT-4-Turbo “as a judge” to label preferences among the 5 policy samples and 1 gold sample (which is also GPT-4-Turbo) as shown in Figure 3. This prompt contains a few minor modifications from the that used in (Yuan et al., 2024). It implements an additive scoring framework on a 6-point scale where a score of 6 represents the highest quality answer according to certain dimensions like “correctness”, “expert knowledge”, “conciseness” etc. By following this rubric, GPT-4 acting as an annotator represents a best-effort general preference model because it compares multiple candidate responses side-by-side in the context window, and stratifies them along meaningful dimensions of quality. Preference Annotation: Training Pair Construction: Adhering to Line 6 in Algorithm 2 implies that not all pairs are suitable for training. Firstly, we must enforce the positives to be high quality in an absolute sense, and secondly, the negatives are directionally worse by a large margin. On the 6 point annotation scale, only samples that score a 5 or 6 are allowed to be positives. From the positives that meet this criteria, if any, we then construct all pairs such that the negative is at least 2 points lower. If the positive happens to be from the student, we relax this constraint to 1 point margin since the GPT-4-Turbo teacher outputs rarely receive a score less than 5 (as shown by the average teacher score in Table 2). Training Pair Construction: Additionally, we are motivated to preserve the preference behavior from previous iterations so that new policies do not inadvertently regress to past bad behavior. To enforce this, we incorporate an exponentially decaying proportion of prior iterations’ training pairs into the current iteration, i.e. we sample at most 30% of training pairs from iteration t − 1, 15% from t − 2, and so on. We do not re-inference outputs for those inputs from the most recent policy. Recall that previous iterations’ inputs are non-overlapping with the splits for other iterations. Training : To prevent overfitting, we train our batched on-policy methods for at most one epoch on newly constructed pairs. Our effective batch size is fixed to 64 for all experiments. Our learning rate, beta, and alpha are found with brief hyperparameter searches. For most experiments, the learning rate is 5E-5, beta is either 0.1 or 0.05, and alpha is 0.005. We found that at higher iterations, the learning rate needs to be lowered. In SFT (supervised fine-tuning) experiments, our learning rate is 5E-6 and we mask out loss for the inputs. We use the open-source TRL library’s implementation to run our experiments. Training Evaluation : Our primary goal is to train a policy that is comparable to the most powerful state-of-the-art langauge models. Hence, AlpacaEval 2.0 (Dubois et al., 2023) is an appropriate benchmark because it computes win-rate against GPT-4-Turbo in a head-to-head fashion on a dataset of 805 input prompts that is shown to correlate with human preferences (0.93 spearman correlation with Chatbot Arena). While it is known that auto-eval methods also correlate with spurious features such as length, a new version of AlpacaEval 2.0 corrects for this with a length-controlled win-rate Evaluation that has an even higher spearman correlation (0.98) with Chatbot Arena [4]. We also evaluate on MT-Bench (Zheng et al., 2023) which allows the llm-as-a-judge to first explain its reasoning before providing a scalar score on 1-10 for the candidate response to a bank of 80 questions. One crucial difference between AlpacaEval 2.0 and MT Bench is that the former asks GPT-4-Turbo to predict which of two side-by-side responses humans would prefer, weighted by the logits to represent its uncertainty, whereas MT-Bench asks the model to first generate a justification and then output a score on 1-10, but it neither defines the ratings (e.g. how a 7 is different than a 5) nor accounts for uncertainty in the logits of the score. We also evaluate on the OpenLLM leaderboard (Beeching et al., 2023), which measures reasoning ability on downstream NLP tasks like coding and question answering by evaluating the accuracy of the multiple choice answer option with the highest logit. Since our training data is primarily instruction-following and not trained to output just the sole answer option, this benchmark is not the primary target of this study; nonetheless, DNO on instruction tuning tasks ought to show no regression on reasoning tasks. 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 [4] https://github.com/tatsu-lab/alpaca_eval

AI That Trains Itself? Here's How it Works

This is Part 6 of a 11-part series based on the research paper “Reinforcement Learning In Agent-based Market Simulation: Unveiling Realistic Stylized Facts And Behavior”. Use the table of links below to navigate to the next part. Table of Links Part 1: Abstract & Introduction Part 2: Important Concepts Part 3: System Description Part 4: Agents & Simulation Details Part 5: Experiment Design Part 6: Continual Learning Part 7: Experiment Results Part 8: Market and Agent Responsiveness to External Events Part 9: Conclusion & References Part 10: Additional Simulation Results Part 11: Simulation Configuration 4.2 Continual Learning We introduce three groups of agents in the simulation. • Group A - Continual Training Group. The agents are pre-trained for 10 hours (36,000 steps), and training continues throughout the time of the simulation (for another 10 hours or 36,000 steps). • Group B - Testing Group. The agents in this group are pre-trained for 10 hours and are used in the simulation without continuing training. • Group C - Untrained Group. The third group serves as a control to understand the performance improvement obtained from training. The agents in this group load the random initialized parameters and run simulations without training. For each random seed, we generate the parameters of the neural networks for the Group C agents directly. Each agent in Group C is trained for 10 hours and their parameters become the parameters used for each agent in Group B. The same parameters are used to initialize the agents in Group A. We are describing the process in detail as this is similar to a matched pairs testing design to minimize randomness for comparison purposes. This is important because we only repeat this process for 10 random seeds, that is 10 simulations. Each simulation takes 20 hours when running all of them in parallel and there are a lot of computational resources required for this study. This process is illustrated in Figure 2. To compare the results produced by the RL agents we introduce an additional simulation model using 100 ZeroIntelligence (ZI) agents. This system is using the agent design in Farmer et al.’s work [9] and [22]. We analyze and compare stylized facts obtained using the RL agents system, the ZI agents system, and real data. Additionally, we investigate the evolution of Market-Maker (MM) agents’ inventory and PnL components across different groups. To assess responsiveness, we introduce a sequence of flash sale events, and we examine the price impact during the flash sale period. We also examine the MM’s change in behavior due to the flash sale, and we evaluate the adaptability of continual learning agents by comparing policies before and after training with the flash sale prices. Authors:
(1) Zhiyuan Yao, Stevens Institute of Technology, Hoboken, New Jersey, USA (zyao9@stevens.edu);
(2) Zheng Li, Stevens Institute of Technology, Hoboken, New Jersey, USA (zli149@stevens.edu);
(3) Matthew Thomas, Stevens Institute of Technology, Hoboken, New Jersey, USA (mthomas3@stevens.edu);
(4) Ionut Florescu, Stevens Institute of Technology, Hoboken, New Jersey, USA (ifloresc@stevens.edu). This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This is Part 6 of a 11-part series based on the research paper “Reinforcement Learning In Agent-based Market Simulation: Unveiling Realistic Stylized Facts And Behavior”. Use the table of links below to navigate to the next part. This is Part 6 of a 11-part series based on the research paper “ Reinforcement Learning In Agent-based Market Simulation: Unveiling Realistic Stylized Facts And Behavior ”. Use the table of links below to navigate to the next part. Part 6 Reinforcement Learning In Agent-based Market Simulation: Unveiling Realistic Stylized Facts And Behavior Table of Links Part 1: Abstract & Introduction Part 1: Abstract & Introduction Part 2: Important Concepts Part 2: Important Concepts Part 3: System Description Part 3: System Description Part 4: Agents & Simulation Details Part 4: Agents & Simulation Details Part 5: Experiment Design Part 5: Experiment Design Part 6: Continual Learning Part 6: Continual Learning Part 7: Experiment Results Part 7: Experiment Results Part 8: Market and Agent Responsiveness to External Events Part 8: Market and Agent Responsiveness to External Events Part 9: Conclusion & References Part 9: Conclusion & References Part 10: Additional Simulation Results Part 10: Additional Simulation Results Part 11: Simulation Configuration Part 11: Simulation Configuration 4.2 Continual Learning We introduce three groups of agents in the simulation. • Group A - Continual Training Group. The agents are pre-trained for 10 hours (36,000 steps), and training continues throughout the time of the simulation (for another 10 hours or 36,000 steps). • Group B - Testing Group. The agents in this group are pre-trained for 10 hours and are used in the simulation without continuing training. without • Group C - Untrained Group. The third group serves as a control to understand the performance improvement obtained from training. The agents in this group load the random initialized parameters and run simulations without training. For each random seed, we generate the parameters of the neural networks for the Group C agents directly. Each agent in Group C is trained for 10 hours and their parameters become the parameters used for each agent in Group B. The same parameters are used to initialize the agents in Group A. We are describing the process in detail as this is similar to a matched pairs testing design to minimize randomness for comparison purposes. This is important because we only repeat this process for 10 random seeds, that is 10 simulations. Each simulation takes 20 hours when running all of them in parallel and there are a lot of computational resources required for this study. This process is illustrated in Figure 2. To compare the results produced by the RL agents we introduce an additional simulation model using 100 ZeroIntelligence (ZI) agents. This system is using the agent design in Farmer et al.’s work [9] and [22]. We analyze and compare stylized facts obtained using the RL agents system, the ZI agents system, and real data. Additionally, we investigate the evolution of Market-Maker (MM) agents’ inventory and PnL components across different groups. To assess responsiveness, we introduce a sequence of flash sale events, and we examine the price impact during the flash sale period. We also examine the MM’s change in behavior due to the flash sale, and we evaluate the adaptability of continual learning agents by comparing policies before and after training with the flash sale prices. Authors: (1) Zhiyuan Yao, Stevens Institute of Technology, Hoboken, New Jersey, USA (zyao9@stevens.edu); (2) Zheng Li, Stevens Institute of Technology, Hoboken, New Jersey, USA (zli149@stevens.edu); (3) Matthew Thomas, Stevens Institute of Technology, Hoboken, New Jersey, USA (mthomas3@stevens.edu); (4) Ionut Florescu, Stevens Institute of Technology, Hoboken, New Jersey, USA (ifloresc@stevens.edu). Authors: Authors: (1) Zhiyuan Yao, Stevens Institute of Technology, Hoboken, New Jersey, USA (zyao9@stevens.edu); (2) Zheng Li, Stevens Institute of Technology, Hoboken, New Jersey, USA (zli149@stevens.edu); (3) Matthew Thomas, Stevens Institute of Technology, Hoboken, New Jersey, USA (mthomas3@stevens.edu); (4) Ionut Florescu, Stevens Institute of Technology, Hoboken, New Jersey, USA (ifloresc@stevens.edu). This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Continual Learning in Reinforcement Learning Agents with Pre-trained, Testing, and Untrained Groups

Ensuring Reproducibility in AI Research: Code and Pre-trained Weights Open-Sourced

Authors:
(1) Albert Q. Jiang;
(2) Alexandre Sablayrolles;
(3) Antoine Roux;
(4) Arthur Mensch;
(5) Blanche Savary;
(6) Chris Bamford;
(7) Devendra Singh Chaplot;
(8) Diego de las Casas;
(9) Emma Bou Hanna;
(10) Florian Bressand;
(11) Gianna Lengyel;
(12) Guillaume Bour;
(13) Guillaume Lample;
(14) Lélio Renard Lavaud;
(15) Lucile Saulnier;
(16) Marie-Anne Lachaux;
(17) Pierre Stock;
(18) Sandeep Subramanian;
(19) Sophia Yang;
(20) Szymon Antoniak;
(21) Teven Le Scao;
(22) Théophile Gervet;
(23) Thibaut Lavril;
(24) Thomas Wang;
(25) Timothée Lacroix;
(26) William El Sayed. Table of Links Abstract and 1. Introduction 2 Architectural details and 2.1 Sparse Mixture of Experts 3 Results 3.1 Multilingual benchmarks, 3.2 Long range performance, and 3.3 Bias Benchmarks 4 Instruction Fine-tuning 5 Routing analysis 6 Conclusion, Acknowledgements, and References Abstract We introduce Mixtral 8x7B, a Sparse Mixture of Experts (SMoE) language model. Mixtral has the same architecture as Mistral 7B, with the difference that each layer is composed of 8 feedforward blocks (i.e. experts). For every token, at each layer, a router network selects two experts to process the current state and combine their outputs. Even though each token only sees two experts, the selected experts can be different at each timestep. As a result, each token has access to 47B parameters, but only uses 13B active parameters during inference. Mixtral was trained with a context size of 32k tokens and it outperforms or matches Llama 2 70B and GPT-3.5 across all evaluated benchmarks. In particular, Mixtral vastly outperforms Llama 2 70B on mathematics, code generation, and multilingual benchmarks. We also provide a model finetuned to follow instructions, Mixtral 8x7B – Instruct, that surpasses GPT-3.5 Turbo, Claude-2.1, Gemini Pro, and Llama 2 70B – chat model on human benchmarks. Both the base and instruct models are released under the Apache 2.0 license. Code: https://github.com/mistralai/mistral-src Webpage: https://mistral.ai/news/mixtral-of-experts/ 1 Introduction In this paper, we present Mixtral 8x7B, a sparse mixture of experts model (SMoE) with open weights, licensed under Apache 2.0. Mixtral outperforms Llama 2 70B and GPT-3.5 on most benchmarks. As it only uses a subset of its parameters for every token, Mixtral allows faster inference speed at low batch-sizes, and higher throughput at large batch-sizes. Mixtral is a sparse mixture-of-experts network. It is a decoder-only model where the feedforward block picks from a set of 8 distinct groups of parameters. At every layer, for every token, a router network chooses two of these groups (the “experts”) to process the token and combine their output additively. This technique increases the number of parameters of a model while controlling cost and latency, as the model only uses a fraction of the total set of parameters per token. Mixtral is pretrained with multilingual data using a context size of 32k tokens. It either matches or exceeds the performance of Llama 2 70B and GPT-3.5, over several benchmarks. In particular, Mixtral demonstrates superior capabilities in mathematics, code generation, and tasks that require multilingual understanding, significantly outperforming Llama 2 70B in these domains. Experiments show that Mixtral is able to successfully retrieve information from its context window of 32k tokens, regardless of the sequence length and the location of the information in the sequence. We also present Mixtral 8x7B – Instruct, a chat model fine-tuned to follow instructions using supervised fine-tuning and Direct Preference Optimization [25]. Its performance notably surpasses that of GPT-3.5 Turbo, Claude-2.1, Gemini Pro, and Llama 2 70B – chat model on human evaluation benchmarks. Mixtral – Instruct also demonstrates reduced biases, and a more balanced sentiment profile in benchmarks such as BBQ, and BOLD. We release both Mixtral 8x7B and Mixtral 8x7B – Instruct under the Apache 2.0 license1 , free for academic and commercial usage, ensuring broad accessibility and potential for diverse applications. To enable the community to run Mixtral with a fully open-source stack, we submitted changes to the vLLM project, which integrates Megablocks CUDA kernels for efficient inference. Skypilot also allows the deployment of vLLM endpoints on any instance in the cloud. This paper is available on arxiv under CC 4.0 license. Authors: (1) Albert Q. Jiang; (2) Alexandre Sablayrolles; (3) Antoine Roux; (4) Arthur Mensch; (5) Blanche Savary; (6) Chris Bamford; (7) Devendra Singh Chaplot; (8) Diego de las Casas; (9) Emma Bou Hanna; (10) Florian Bressand; (11) Gianna Lengyel; (12) Guillaume Bour; (13) Guillaume Lample; (14) Lélio Renard Lavaud; (15) Lucile Saulnier; (16) Marie-Anne Lachaux; (17) Pierre Stock; (18) Sandeep Subramanian; (19) Sophia Yang; (20) Szymon Antoniak; (21) Teven Le Scao; (22) Théophile Gervet; (23) Thibaut Lavril; (24) Thomas Wang; (25) Timothée Lacroix; (26) William El Sayed. Authors: Authors: (1) Albert Q. Jiang; (2) Alexandre Sablayrolles; (3) Antoine Roux; (4) Arthur Mensch; (5) Blanche Savary; (6) Chris Bamford; (7) Devendra Singh Chaplot; (8) Diego de las Casas; (9) Emma Bou Hanna; (10) Florian Bressand; (11) Gianna Lengyel; (12) Guillaume Bour; (13) Guillaume Lample; (14) Lélio Renard Lavaud; (15) Lucile Saulnier; (16) Marie-Anne Lachaux; (17) Pierre Stock; (18) Sandeep Subramanian; (19) Sophia Yang; (20) Szymon Antoniak; (21) Teven Le Scao; (22) Théophile Gervet; (23) Thibaut Lavril; (24) Thomas Wang; (25) Timothée Lacroix; (26) William El Sayed. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Architectural details and 2.1 Sparse Mixture of Experts 2 Architectural details and 2.1 Sparse Mixture of Experts 3 Results 3 Results 3.1 Multilingual benchmarks, 3.2 Long range performance, and 3.3 Bias Benchmarks 3.1 Multilingual benchmarks, 3.2 Long range performance, and 3.3 Bias Benchmarks 4 Instruction Fine-tuning 4 Instruction Fine-tuning 5 Routing analysis 5 Routing analysis 6 Conclusion, Acknowledgements, and References 6 Conclusion, Acknowledgements, and References Abstract We introduce Mixtral 8x7B, a Sparse Mixture of Experts (SMoE) language model. Mixtral has the same architecture as Mistral 7B, with the difference that each layer is composed of 8 feedforward blocks (i.e. experts). For every token, at each layer, a router network selects two experts to process the current state and combine their outputs. Even though each token only sees two experts, the selected experts can be different at each timestep. As a result, each token has access to 47B parameters, but only uses 13B active parameters during inference. Mixtral was trained with a context size of 32k tokens and it outperforms or matches Llama 2 70B and GPT-3.5 across all evaluated benchmarks. In particular, Mixtral vastly outperforms Llama 2 70B on mathematics, code generation, and multilingual benchmarks. We also provide a model finetuned to follow instructions, Mixtral 8x7B – Instruct, that surpasses GPT-3.5 Turbo, Claude-2.1, Gemini Pro, and Llama 2 70B – chat model on human benchmarks. Both the base and instruct models are released under the Apache 2.0 license. Code : https://github.com/mistralai/mistral-src Code Webpage : https://mistral.ai/news/mixtral-of-experts/ Webpage 1 Introduction In this paper, we present Mixtral 8x7B, a sparse mixture of experts model (SMoE) with open weights, licensed under Apache 2.0. Mixtral outperforms Llama 2 70B and GPT-3.5 on most benchmarks. As it only uses a subset of its parameters for every token, Mixtral allows faster inference speed at low batch-sizes, and higher throughput at large batch-sizes. Mixtral is a sparse mixture-of-experts network. It is a decoder-only model where the feedforward block picks from a set of 8 distinct groups of parameters. At every layer, for every token, a router network chooses two of these groups (the “experts”) to process the token and combine their output additively. This technique increases the number of parameters of a model while controlling cost and latency, as the model only uses a fraction of the total set of parameters per token. Mixtral is pretrained with multilingual data using a context size of 32k tokens. It either matches or exceeds the performance of Llama 2 70B and GPT-3.5, over several benchmarks. In particular, Mixtral demonstrates superior capabilities in mathematics, code generation, and tasks that require multilingual understanding, significantly outperforming Llama 2 70B in these domains. Experiments show that Mixtral is able to successfully retrieve information from its context window of 32k tokens, regardless of the sequence length and the location of the information in the sequence. We also present Mixtral 8x7B – Instruct, a chat model fine-tuned to follow instructions using supervised fine-tuning and Direct Preference Optimization [25]. Its performance notably surpasses that of GPT-3.5 Turbo, Claude-2.1, Gemini Pro, and Llama 2 70B – chat model on human evaluation benchmarks. Mixtral – Instruct also demonstrates reduced biases, and a more balanced sentiment profile in benchmarks such as BBQ, and BOLD. We release both Mixtral 8x7B and Mixtral 8x7B – Instruct under the Apache 2.0 license1 , free for academic and commercial usage, ensuring broad accessibility and potential for diverse applications. To enable the community to run Mixtral with a fully open-source stack, we submitted changes to the vLLM project, which integrates Megablocks CUDA kernels for efficient inference. Skypilot also allows the deployment of vLLM endpoints on any instance in the cloud. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Mixtral—a Multilingual Language Model Trained with a Context Size of 32k Tokens 

Mixtral – en flersproget sprogmodel trænet med en kontekststørrelse på 32k tokens 

Mixtral: un modelo de lenguaje multilingüe entrenado con un tamaño de contexto de 32 000 tokens

Mixtral - یک مدل زبان چندزبانه آموزش دیده با اندازه زمینه 32 کیلو توکن 

Mixtral: un modelo lingüístico multilingüe adestrado cun tamaño de contexto de 32k tokens 

Mixtral - 32,000 トークンのコンテキスト サイズでトレーニングされた多言語モデル

Mixtral — მრავალენოვანი ენობრივი მოდელი, რომელიც მომზადებულია კონტექსტის ზომით 32 კ ჟეტონებით 

Mixtral - 32k 토큰의 컨텍스트 크기로 훈련된 다국어 언어 모델

Mixtral—Modely amin'ny fiteny maro nofanina miaraka amin'ny haben'ny toe-javatra misy marika 32k 

Mixtral—32к жетон контекстийн хэмжээтэй бэлтгэгдсэн олон хэлний загвар 

Mixtral — um modelo de linguagem multilíngue treinado com um tamanho de contexto de 32k tokens

Mixtral — viacjazyčný jazykový model trénovaný s veľkosťou kontextu 32 000 tokenov 

Mixtral—Mô hình ngôn ngữ đa ngôn ngữ được đào tạo với kích thước ngữ cảnh là 32k mã thông báo

Pawsitive Results: How to Build the Ultimate Puppy Trainer AI-chatbot With Coze

Pawsitive Results: Hoe om die Ultimate Puppy Trainer AI-chatbot met Coze te bou

Pawsitive Nəticələr: Coze ilə Ən Yaxşı Puppy Təlimçisi AI-chatbotunu necə qurmaq olar

Pawsitive Results: Как да изградите Ultimate Puppy Trainer AI-chatbot с Coze

Resultados positivos: Cómo crear el mejor chatbot de inteligencia artificial para adiestrar cachorros con Coze

نتایج Pawsitive: چگونه با Coze ربات هوش مصنوعی بهترین مربی توله سگ بسازیم

ポジティブな結果: Coze で究極の子犬トレーナー AI チャットボットを構築する方法

Pawsitive Results: Ndenge nini kotonga AI-chatbot ya Formateur ya mwana ya mbwa ya suka na Coze

Pawsitive rezultāti: kā izveidot izcilu kucēnu treneri AI tērzēšanas robotu ar Coze

Resultados positivos: como criar o melhor chatbot de IA para treinador de filhotes com Coze

Rezultatet Pawsitive: Si të ndërtoni AI-chatbotin e fundit të trajnerit të qenve me Coze

Павситивни резултати: Како са Цозе-ом изградити врхунског тренера штенаца са вештачком интелигенцијом

Vuyelo bya Pawsitive: Ndlela yo Aka AI-chatbot ya Mudyondzisi wa Timbyana wo Hetelela Hi Coze

Google Trains Its Employees To Shield Emails And Other Documents From Review 

A few months ago when I first started learning about Cryptocurrencies, It felt like I had been thrown into a whole new world with people using complex terms I didn't understand anything about. One concept in particular I struggled to understand was that of APIs. It took hours of tutorials and personal research to finally understand what APIs are and what they do. So, if you're a trainee web developer, crypto enthusiast, or a newbie to crypto trading, and you currently have a similar problem that I had, then this is the one tutorial you are going to need. Using simple language and easy-to-understand analogies, at the end of this tutorial, you'll save yourself hours of research time, and you will be able to answer the basic questions: What are Cryptocurrency APIs and how do they work?
Why are Crypto Currency APIs important?
What are the top APIs out there? Now let's dive in! What Are Cryptocurrency APIs, and How Do They Work? To understand this, we must first understand what APIs are. API is an acronym that stands for Application Programming Interface. Sounds complex? let's break it down. An application is software that helps people obtain information about specific things or carry out particular tasks like how the Expedia app on your phone helps you book flights on the go or the Amazon weather app helps you check weather forecasts. An interface in this context, is a program that helps 2 applications or websites communicate with each other, kinda like a middleman. So an API is basically a program or a “link” that helps two apps (or websites) “talk” to each other about a particular thing. So, when an app (let's call it app A) needs information about a particular thing or needs access to a service rendered by another app B, an API of app B is used. This API for app B serves as a link, a connection via which the specific information app A requires from app B is obtained. Wait … What?! Okay, okay. Let's break it down further using this analogy: Imagine you're opening a bakery, and you decide to sell cupcakes, muffins, and sourdough bread. For some reason, however,  you don't have the capacity to make sourdough bread of your own. What do you do? You reach out to another bakery that sells sourdough bread. Make an agreement for them to sell you sourdough bread when you request it. So, every time you need bread, you reach out to an employee of that bakery, and he brings you the bread. Now, bread isn't the only thing this second bakery sells, but it's the only thing you need from them. That's basically how an API works. When an app needs information that it cannot provide on its own, it uses the API of another app to gain access to that information from the other app. For a more technical analogy. Let's use a flight booking app to illustrate how APIs work: Imagine you're using a travel app like Expedia to book a flight. When you search for flights, the app doesn't have all the flight information stored locally. Instead, it uses specific APIs to fetch data from airlines' systems or other flight databases. Here's basically what happens behind the scenes: The travel app sends a request to the airline's API (e.g., "Get flights from Nigeria to New York on August 26th").


The airline's API processes the request and returns the relevant data (e.g., flight numbers, departure times, prices).


The travel app receives the data and displays it in a user-friendly format. So you see: The travel app is the "client" or "consumer" of the API. The airline's system is the "provider" of the API. The API acts as an intermediary, allowing the travel app to access the airline's data without having to replicate it locally. So, if I’m an app developer and the app I'm working on serves to provide specific information about the flight schedules from Qatar Airways for instance, I will use an API of Qatar Airways to get me and my future users access to that information. Now that we understand what an API is, applying that knowledge directly, we can say that cryptocurrency APIs are basically programs that enable different apps or websites to get access to specific data about Cryptocurrencies (i.e., chart updates, transactions, etc.) from other websites that directly monitor and interact with the Crypto Market. So, if I want information about the current status of Bitcoin on the Crypto Market, I open a website like Dervi. This website cannot monitor the Crypto market directly, so it uses an API to get information about current Bitcoin status from the Trading View website. Why Are Crypto Currency APIs Important? From what we've discussed earlier, we can say some of the importance of these APIs to Cryptocurrency are: Real-time data access: APIs provide up-to-date information on market prices, trading volumes, and blockchain data, enabling informed decisions.


Research and analysis: APIs provide data for researchers, analysts, and investors to study market trends and make predictions.


Integration with other services: APIs enable integration with payment systems, e-commerce platforms, and other services, expanding cryptocurrency adoption. In essence, cryptocurrency APIs facilitate innovation, automation, and accessibility in the cryptocurrency and blockchain ecosystem. What Are the Top Cryptocurrency APIs? Now that we understand what APIs are and can see why they are so important, let's look at some of the top Crypto CryptoCurrency APIs: Coin Market Cap API Coin Market Cap is one of the best cryptocurrency data APIs. So as a developer, incorporating it into your app or website will be really beneficial. You see it offers real-time cryptocurrency data and market information, such as: Historical cryptocurrency data since 2013
Live price updates for cryptocurrency assets
Market data, volume, and exchange information
Trading data and analytics
Data on non-fungible tokens (NFTs) Enterprise, professional, and free API plans This is an API that is used all over the world by cryptocurrency traders, developers, and financial firms. CoinGeko API CoinGeko is a big name amongst web developers! It offers many exciting features and has an easy-to-use interface. It has its own charts which it uses to track several coins in the Crypto market. Some of the important features it offers are these: Real-time and historical crypto prices,
Market capitalization data,
Trading volume and exchange data,
Metadata,
NFT collection floor prices. Coin Gecko API offers a free plan! As well as several paid plans with additional features and support. Binance: Binance is considered one of the largest cryptocurrency exchange platforms globally. It allows users to buy and sell digital assets. Its free API allows web developers to interact with the Binance platform and create custom trading applications like: Executing trades and checking their account balances.
Viewing their trading history using the API.
Enables developers to create custom trading strategies and automate their trading activities. The world of cryptocurrency is constantly changing, evolving, and expanding. APIs make it easier for web developers to build cryptocurrency-related apps, wallets, exchanges, and analytics tools. Which makes monitoring of trends within the Crypto market much easier for traders. A few months ago when I first started learning about Cryptocurrencies, It felt like I had been thrown into a whole new world with people using complex terms I didn't understand anything about. One concept in particular I struggled to understand was that of APIs. It took hours of tutorials and personal research to finally understand what APIs are and what they do. So, if you're a trainee web developer, crypto enthusiast, or a newbie to crypto trading, and you currently have a similar problem that I had, then this is the one tutorial you are going to need. Using simple language and easy-to-understand analogies, at the end of this tutorial, you'll save yourself hours of research time, and you will be able to answer the basic questions: What are Cryptocurrency APIs and how do they work? Why are Crypto Currency APIs important? What are the top APIs out there? What are Cryptocurrency APIs and how do they work? Why are Crypto Currency APIs important? What are the top APIs out there? Now let's dive in! What Are Cryptocurrency APIs, and How Do They Work? What Are Cryptocurrency APIs, and How Do They Work? To understand this, we must first understand what APIs are. API is an acronym that stands for Application Programming Interface. Sounds complex? let's break it down. An application is software that helps people obtain information about specific things or carry out particular tasks like how the Expedia app on your phone helps you book flights on the go or the Amazon weather app helps you check weather forecasts. An interface in this context, is a program that helps 2 applications or websites communicate with each other, kinda like a middleman. So an API is basically a program or a “link” that helps two apps (or websites) “talk” to each other about a particular thing. So, when an app (let's call it app A) needs information about a particular thing or needs access to a service rendered by another app B, an API of app B is used. This API for app B serves as a link, a connection via which the specific information app A requires from app B is obtained. Wait … What?! Wait … What?! Okay, okay. Let's break it down further using this analogy: Imagine you're opening a bakery, and you decide to sell cupcakes, muffins, and sourdough bread. For some reason, however,  you don't have the capacity to make sourdough bread of your own. What do you do? What do you do? You reach out to another bakery that sells sourdough bread. Make an agreement for them to sell you sourdough bread when you request it. So, every time you need bread, you reach out to an employee of that bakery, and he brings you the bread. Now, bread isn't the only thing this second bakery sells, but it's the only thing you need from them. That's basically how an API works. When an app needs information that it cannot provide on its own, it uses the API of another app to gain access to that information from the other app. For a more technical analogy. Let's use a flight booking app to illustrate how APIs work: Imagine you're using a travel app like Expedia to book a flight. When you search for flights, the app doesn't have all the flight information stored locally. Instead, it uses specific APIs to fetch data from airlines' systems or other flight databases. Here's basically what happens behind the scenes: The travel app sends a request to the airline's API (e.g., "Get flights from Nigeria to New York on August 26th"). The airline's API processes the request and returns the relevant data (e.g., flight numbers, departure times, prices). The travel app receives the data and displays it in a user-friendly format. The travel app sends a request to the airline's API (e.g., "Get flights from Nigeria to New York on August 26th"). The travel app sends a request to the airline's API (e.g., "Get flights from Nigeria to New York on August 26th"). The airline's API processes the request and returns the relevant data (e.g., flight numbers, departure times, prices). The airline's API processes the request and returns the relevant data (e.g., flight numbers, departure times, prices). The travel app receives the data and displays it in a user-friendly format. The travel app receives the data and displays it in a user-friendly format. So you see: The travel app is the "client" or "consumer" of the API. The travel app is the "client" or "consumer" of the API. The airline's system is the "provider" of the API. The airline's system is the "provider" of the API. The API acts as an intermediary, allowing the travel app to access the airline's data without having to replicate it locally. The API acts as an intermediary, allowing the travel app to access the airline's data without having to replicate it locally. So, if I’m an app developer and the app I'm working on serves to provide specific information about the flight schedules from Qatar Airways for instance, I will use an API of Qatar Airways to get me and my future users access to that information. Now that we understand what an API is, applying that knowledge directly, we can say that cryptocurrency APIs are basically programs that enable different apps or websites to get access to specific data about Cryptocurrencies (i.e., chart updates, transactions, etc.) from other websites that directly monitor and interact with the Crypto Market. So, if I want information about the current status of Bitcoin on the Crypto Market, I open a website like Dervi. This website cannot monitor the Crypto market directly, so it uses an API to get information about current Bitcoin status from the Trading View website. Why Are Crypto Currency APIs Important? Why Are Crypto Currency APIs Important? From what we've discussed earlier, we can say some of the importance of these APIs to Cryptocurrency are: Real-time data access: APIs provide up-to-date information on market prices, trading volumes, and blockchain data, enabling informed decisions. Research and analysis: APIs provide data for researchers, analysts, and investors to study market trends and make predictions. Integration with other services: APIs enable integration with payment systems, e-commerce platforms, and other services, expanding cryptocurrency adoption. Real-time data access: APIs provide up-to-date information on market prices, trading volumes, and blockchain data, enabling informed decisions. Real-time data access: APIs provide up-to-date information on market prices, trading volumes, and blockchain data, enabling informed decisions. Research and analysis: APIs provide data for researchers, analysts, and investors to study market trends and make predictions. Research and analysis: APIs provide data for researchers, analysts, and investors to study market trends and make predictions. Integration with other services: APIs enable integration with payment systems, e-commerce platforms, and other services, expanding cryptocurrency adoption. Integration with other services: APIs enable integration with payment systems, e-commerce platforms, and other services, expanding cryptocurrency adoption. In essence, cryptocurrency APIs facilitate innovation, automation, and accessibility in the cryptocurrency and blockchain ecosystem. What Are the Top Cryptocurrency APIs? What Are the Top Cryptocurrency APIs? Now that we understand what APIs are and can see why they are so important, let's look at some of the top Crypto CryptoCurrency APIs: Coin Market Cap API Coin Market Cap API Coin Market Cap is one of the best cryptocurrency data APIs. So as a developer, incorporating it into your app or website will be really beneficial. You see it offers real-time cryptocurrency data and market information, such as: Historical cryptocurrency data since 2013 Live price updates for cryptocurrency assets Market data, volume, and exchange information Trading data and analytics Data on non-fungible tokens (NFTs) Enterprise, professional, and free API plans Historical cryptocurrency data since 2013 Live price updates for cryptocurrency assets Market data, volume, and exchange information Trading data and analytics Data on non-fungible tokens (NFTs) Enterprise, professional, and free API plans This is an API that is used all over the world by cryptocurrency traders, developers, and financial firms. CoinGeko API CoinGeko API CoinGeko is a big name amongst web developers! It offers many exciting features and has an easy-to-use interface. It has its own charts which it uses to track several coins in the Crypto market. Some of the important features it offers are these: Real-time and historical crypto prices, Market capitalization data, Trading volume and exchange data, Metadata, NFT collection floor prices. Real-time and historical crypto prices, Market capitalization data, Trading volume and exchange data, Metadata, NFT collection floor prices. Coin Gecko API offers a free plan! As well as several paid plans with additional features and support. Binance: Binance: Binance is considered one of the largest cryptocurrency exchange platforms globally. It allows users to buy and sell digital assets. Its free API allows web developers to interact with the Binance platform and create custom trading applications like: Executing trades and checking their account balances. Viewing their trading history using the API. Enables developers to create custom trading strategies and automate their trading activities. Executing trades and checking their account balances. Viewing their trading history using the API. Enables developers to create custom trading strategies and automate their trading activities. The world of cryptocurrency is constantly changing, evolving, and expanding. APIs make it easier for web developers to build cryptocurrency-related apps, wallets, exchanges, and analytics tools. Which makes monitoring of trends within the Crypto market much easier for traders.

A Beginner's Introduction to Cryptocurrency APIs: The Basics You Need as a Trainee Web Developer

Hello World is a weekly newsletter—delivered every Saturday morning—that goes deep into our original reporting and the questions we put to big thinkers in the field. Browse the archive here. Hello World is a weekly newsletter—delivered every Saturday morning—that goes deep into our original reporting and the questions we put to big thinkers in the field. Browse the archive here. Browse the archive here Hi there, It’s me, Tara García Mathewson, back in the Hello World rotation to get you thinking about civil rights in education. Earlier this week, the Center for Democracy & Technology (CDT) put out a report arguing that decades-old civil rights laws should be protecting students from discrimination based on their race—but they’re not. Reporting from The Markup was at the heart of one of its examples. but they’re not Did you see the story Todd Feathers wrote in April about an early warning system in Wisconsin designed to identify students at risk of dropping out? Well, the point of systems like these is to help schools figure out where to focus limited resources. They use past data about individual students, their schools, and districts to make predictions about current students. In Wisconsin, the system tends to be used by high schools considering how to serve incoming freshmen. But Todd cited some research out of the University of California, Berkeley in his story that looked at 10 years’ worth of data from Wisconsin’s system and found being labeled “high risk”—and therefore targeted for additional services—may not have made a difference. an early warning system in Wisconsin Kristin Woelfel, policy counsel at the Center for Democracy & Technology and lead author of this week’s report, said those findings prompted alarm bells. Government entities have to be careful about where they take race into account when treating people differently. Woelfel told me there’s a basic standard in civil rights law: Use of race must be “narrowly tailored to achieve a compelling government interest.” The Wisconsin system ostensibly used race to help prevent dropouts. That sounds like a compelling government interest! But not if it didn’t work. The UC Berkeley researchers looked at students just above and below the cutoff for being labeled high risk, and expected to see that students labeled high risk got relatively more services and support, and therefore graduated at relatively higher rates—but they didn’t find that. “The “high risk” label may not have helped students at all: the researchers couldn’t rule out a 0 percent increase in graduation rates as a result of the label. I discussed this with Cody Venzke, a senior policy counsel at the ACLU, which co-signed a letter the CDT sent to the U.S. Department of Education and the White House this week. The coalition urged the federal government to follow up its 2022 Blueprint for an AI Bill of Rights with more emphasis on civil rights. to the U.S. Department of Education and the White House Blueprint for an AI Bill of Rights Venzke said it’s always worth being suspicious when race is used to make decisions about how to treat individuals, given historical discrimination. In Wisconsin, The Markup’s reporting made clear educators were not properly trained to understand the dropout risk labels and intervene accordingly. The investigation features students who said they found the labels stigmatizing. And if that stigma isn’t offset by demonstrable benefits for students—in higher graduation rates, for example—Venzke sees the potential for a civil rights violation. But maybe it doesn’t have to be that way. Venzke points to schools’ obligation to protect students’ civil rights, ensure algorithmic systems are fair, and protect student privacy. “One of the ways you can achieve all three of those is by using aggregate data where aggregate data will fulfill the purpose,” Venzke said. As it turns out, the UC Berkeley researchers found it would. In their paper, currently under peer review, they recommend Wisconsin scrap its use of individual factors like a student’s race or even test scores in determining that student’s dropout risk. Juan Perdomo, lead author on the paper and now a postdoctoral fellow at Harvard, told me that in highly segregated Wisconsin, building an early warning system with data from entire schools or districts offers equally reliable predictions about which students are likely to drop out. currently under peer review “It also provides a much more actionable way of improving graduation outcomes,” Perdomo told me. And isn’t that the point? Early warning systems aren’t supposed to just be statistical exercises in identifying students at risk. They’re supposed to help better serve those students. If an early warning system identifies entire schools as high risk—because it doesn’t factor in individual characteristics—maybe it would drive statewide spending or districtwide training toward those schools and create a clearer path toward making a difference. For its part, the Wisconsin Department of Public Instruction is studying its early warning systems and considering changes. Communications Officer Chris Bucher told me the department has heard about concerns for some time now. So, we’ll see. In the meantime, we at The Markup will continue to report on algorithms that fuel discrimination and what that means for people’s lives. Thanks for reading! Tara García Mathewson Reporter The Markup Tara García Mathewson, Investigative Reporter Tara García Mathewson, Investigative Reporter Tara García Mathewson, Investigative Reporter Tara García Mathewson Graphics Graphics Gabriel Hongsdusit Gabriel Hongsdusit Gabriel Hongsdusit Gabriel Hongsdusit Engagement Engagement Maria Puertas Maria Puertas Maria Puertas Maria Puertas Copy Editing and Production Copy Editing and Production Sabrina Toppa Sabrina Toppa Sabrina Toppa Sabrina Toppa Editing Editing Monica Riese
Ko Bragg Monica Riese Monica Riese Monica Riese Ko Bragg Ko Bragg Ko Bragg Also published here Also published here here Photo by Artyom Kabajev on Unsplash Photo by Artyom Kabajev on Unsplash Artyom Kabajev Unsplash

Stories, News & Mentions about TRAINFES

Categories

Trending Topics

Stories, News & Mentions about TRAINFES

Trainfes’s Stories on HackerNoon

Trainfes's News & Mentions

Categories

Trending Topics