Activity
5K followers
-
Anant Raj shared thisI am looking for a highly motivated Postdoctoral Researcher to join my group at the Department of Computer Science & Automation, Indian Institute of Science (IISc). About Us: We are a young and growing lab pursuing deep, foundational questions in machine learning theory and optimization. The group is headed by me, Anant Raj. My homepage and Google Scholar profiles are available at the end of this post for those interested. Research: The research will focus on the theory of machine learning, with topics including: • Optimization theory (SGD, GD, overparameterized networks) • Sampling from arbitrary distributions, interacting particle systems, and large-deviation/mean-field methods • Robustness, generalization, and dynamics of modern ML systems 📍 Location: IISc Bangalore 📅 Start date: Flexible 🕒 Duration: 1–2 years (renewable after 1 year) If you have a strong background in mathematics, optimization, probability, or theoretical ML, and are excited about deep, foundational problems in modern machine learning, I’d be happy to hear from you. 📧 To apply: Send a CV, research statement, and two representative publications to anantraj@iisc.ac.in. Applications are reviewed on a rolling basis. Please feel free to share with anyone who might be interested! Required Qualifications: • Ph.D. in CS, Mathematics, Statistics, EE, or related fields. • Strong theoretical foundation in optimization, applied probability, or machine learning theory. • Evidence of strong research capability (publications at top machine learning conferences (ICML, Neurips, COLT, AISTATS, ALT, etc), publications at top journals in the field, preprints, thesis). My Homepage: https://lnkd.in/gE8asR-k Google Scholar: https://lnkd.in/gau-UqK2
-
Anant Raj shared thisI had the privilege of interacting with him only once, during my undergraduate days at IIT Kanpur, when he visited to deliver a talk on NLP. At that time, I was as uncertain about my future path as any undergraduate could be. After speaking with him, I was deeply struck by how clearly he understood the research problems he wished to pursue. More than a decade later, I can truly appreciate the wisdom in what he shared that day. Our nation remains deeply grateful to him for transforming the NLP landscape in India. Rest in peace, Sir!!Anant Raj shared thisWith deep sorrow and profound grief, the Department of Computer Science and Engineering at IIT Bombay conveys the tragic news that Prof. Pushpak Bhattacharyya passed away on October 5th, 2025. Prof. Bhattacharyya joined the department in 1988 and dedicated over 37 years to academia, leaving behind a remarkable legacy as a scholar, leader, and mentor. A foremost authority in Artificial Intelligence and Natural Language Processing, particularly for Indian languages, he authored several books, produced hundreds of research publications, and guided generations of students with distinction. In addition to his academic contributions, Prof. Bhattacharyya played key leadership roles in shaping technology and education in India. He served as Director of IIT Patna (2015–2020) and as the first Indian President of the Association for Computational Linguistics (2016). Most recently, he chaired the Reserve Bank of India’s Committee on Ethical AI in Finance, the Committee for Indian Language Standards at MeitY, and the Bureau of Indian Standards committee on AI standardization, representing India in international forums. Prof. Bhattacharyya will be remembered not only for his intellectual stature and accomplishments but also for his generosity as a teacher and mentor, and his lifelong dedication to building institutions and advancing technology for society. He will be deeply missed by the IIT Bombay community, and colleagues across the world. We extend our heartfelt condolences to his family and loved ones, and pray for peace for his soul. To share your eulogy, please visit: https://lnkd.in/dbt6h3uJ
-
Anant Raj shared thisHiring Announcement: Research Assistants or Pre-Doctoral Fellows at IISc! We are excited to announce two openings for Research Assistants or Pre-Doctoral Fellows, jointly hosted by the Department of Computer Science and Automation and the Department of Electrical Communication Engineering at the Indian Institute of Science (IISc). These positions are well-suited for motivated individuals interested in advancing foundational research in machine learning, statistics, and applied probability. Supervisors: Successful candidates will be jointly supervised by Prof. Shubhada Agrawal (Department of Electrical Communication Engineering) and Prof. Anant Raj (Department of Computer Science and Automation) at the Indian Institute of Science (IISc). The successful candidate will benefit from: -Collaboration with leading researchers. -Opportunities for publication in top-tier machine learning venues. -Strong letters of recommendation for future academic pursuits. Qualifications: -A robust academic background in applied mathematics, statistics, optimization, or computer science. -Prior research experience in theoretical machine learning is preferred but not mandatory. Duration: 1-2 years (renewable) If you are passionate about advancing the field of machine learning, I encourage you to apply by filling out this form: https://lnkd.in/g-ahi-tW Update : We will stop accepting applications starting Monday, August 11, 2025. If you haven’t applied yet, we encourage you to do so before the deadline.
-
Anant Raj shared thisI am delighted to announce that I have been awarded the Early Career Research Grant (PM-ECRG) by the Anusandhan National Research Foundation (ANRF). As part of this grant, I will soon be looking to hire two Junior Research Fellows (JRFs) for an initial period of one year, extendable up to three years. Candidates with undergraduate degrees from top colleges (IITs, Top NITs, ISI, IMSc, IISER Pune etc) and good academic records are encouraged to reach out via email. Additionally, I anticipate hiring 0–2 PhD students this fall. The IISc PhD admission application is now open—interested candidates are encouraged to apply. I am also seeking a postdoctoral researcher to work in the area of theoretical machine learning. A strong research record, including publications in top-tier ML conferences (ICML, NeurIPS, AISTATS, COLT, ICLR, ALT), is a key requirement for this position. Stay tuned for more updates!
-
Anant Raj posted this🌟 Excited to announce that our paper, "Variational Principles for Mirror Descent and Mirror Langevin Dynamics," co-authored with Belinda Tzen, Maxim Raginsky, and Francis Bach, has received the 2024 IEEE CSS Roberto Tempo Best CDC Paper Award! This prestigious recognition, awarded by the IEEE Control Systems Society, celebrates the paper’s originality, potential impact on control theory and related fields, and the clarity of its presentation. It’s an honor to receive this award, named after Roberto Tempo, a pioneer in the field. I’m deeply grateful to my incredible collaborators and the broader research community for their support and inspiration. The paper was presented at the 62nd IEEE Conference on Decision and Control (CDC 2023). Here is the arXiv link for the paper: https://lnkd.in/g62mzdyM
-
Anant Raj shared thisI am honored to receive the "Google India Research Award, 2024" to work on overparametrization in Machine Learning. My sincere thanks to Google Research for their invaluable support and to my collaborators for their continued partnership and insights. Together, I am excited to push the boundaries of understanding in this critical area of machine learning. #GoogleResearchAward #IISc #CSA
-
Anant Raj shared thisHiring Announcement: Research Assistant or Pre-Doctoral Fellow at IISc! I am pleased to announce an opening for a Research Assistant or Pre-Doctoral Fellow in the Department of Computer Science and Automation at the Indian Institute of Science (IISc). This position is ideal for individuals eager to contribute to cutting-edge theoretical research in machine learning and optimization, specifically focusing on: -The effectiveness of Stochastic Gradient Descent (SGD) and Gradient Descent (GD) in training overparametrized neural networks. -Exploring theoretical frameworks for generative models. The successful candidate will benefit from: -Collaboration with leading researchers. -Opportunities for publication in top-tier machine learning venues. -Strong letters of recommendation for future academic pursuits. Qualifications: -A robust academic background in applied mathematics, statistics, optimization, or computer science. -Prior research experience in theoretical machine learning is preferred but not mandatory. Position Details: Duration: 1-2 years (renewable) Salary: By government norms (~50K INR/Month). If you are passionate about advancing the field of machine learning, I encourage you to apply by filling out this form: https://lnkd.in/gc9NUNZA #IISc #MachineLearning #ResearchAssistant #JoinUs
-
Anant Raj posted thisCareer Update: I am delighted to announce that I will be joining the Indian Institute of Science (IISc), Bangalore as an Assistant Professor in the Department of Computer Science and Automation. After spending nine years in Europe and the USA, I am excited to return to India and contribute to the Indian academic and research ecosystem. It is an honor to join the company of distinguished researchers at IISc. I look forward to contributing to the Indian research ecosystem in AI and ML theory and joining the community of researchers dedicated to building a strong foundation for the future of AI in India. This marks another life-changing moment in my career. In the near future, I will be seeking PhD students and research assistants with strong backgrounds in statistics, computer science, mathematics, or physics. Please stay tuned for further updates.
Anant Raj commented on a post
1y
Great Opportunity! CFBR.
-
Anant Raj liked thisAnant Raj liked this♟️ No double gold this time — but India still returns with three major podium finishes in Chess Olympiad. 🇮🇳 After the extraordinary high of two team golds last year, this year India still manages to win medals in all categories: 🏆 Gaprindashvili Trophy ( the best combined score in the open and women's events ) — India (consecutive third win) 🥈 Open Team — Silver 🥉 Women’s Team — Bronze Congratulations to Uzbekistan for the open team gold and to China for the women's team gold! #Chess #ChessOlympiad #ViswanathanAnand #Sports #India
-
Anant Raj liked thisAnant Raj liked thisThis weekend, I finally had time to dive into OpenAI's new construction for finite-time blow-up in the forced Navier–Stokes equations, which was announced last week. The geometry immediately looked familiar: an axisymmetric swirl core in anisotropic similarity variables. It’s the exact porous-cylinder swirl flow Nail Gumerov and I solved and computed via Chebyshev collocation back in 1998 (GD1998). While OpenAI's paper is a proof of existence (computing nothing and relying on schematics), I wanted to answer two questions: 1. Could I compute something similar using our old methods? 2. Could this propose a real laboratory experiment to realize the blow-up configuration? Over Saturday and Sunday, I put the GD1998 method to the test on this new problem using modernizations: Newton iteration with an exact complex-step Jacobian, an explicitly pinned pressure gauge, and pseudo-arclength continuation to pass previously impassable folds. My collaborator? Claude Fable 5.1. The pace was unlike anything I’ve experienced: a recast, a verified solver, a sweep of 6,000 converged points, the blow-up core as a Cauchy problem, and a complete draft paper—all in two days. (It’s a striking glimpse into what doing science at this speed and scale looks like. The upcoming preprint will carry Claude as co-author--raising fascinating questions about AI and scientific attribution that we will certainly discuss another day.) What we found: ◆ The generalization of our old boundary value problem reveals a fold: steady swirl between porous walls is lost beyond a critical swirl scaling with the square of the inflow. ◆ A non-symmetric axial through-flow core meets the construction's moment identities to 0.2%. ◆ Real fluids (water/air) would likely cavitate or shock before the anomalous exponent is measurable, suggesting the 1998 porous-wall chamber as a viable experimental apparatus. The second layer—oscillatory pulses supplying missing momentum—is currently running across two clusters. arXiv preprint coming soon! 🚀 Department of Computer Science, UMD Artificial Intelligence Interdisciplinary Institute at Maryland (AIM), UMIACS Post dedicated to the memory of Nail Gumerov (1960–2022). Image on the left: OpenAI, from the announcement of "Finite time blowup for Navier–Stokes" (September 2026). (Links to the OpenAI article and to the GD1998 paper are in the comments.)
-
Anant Raj liked thisAnant Raj liked thisElected Fellow of NASI: Honoured to Join the Legacy 🚀 I have been elected a Fellow of the National Academy of Sciences, India (NASI). What makes this especially meaningful to me is the history behind those four letters. Founded in 1930 at Prayagraj by the legendary astrophysicist Prof. Meghnad Saha, NASI is India’s oldest science academy. For nearly a century, the Academy has carried forward a beautiful mandate—“Science & Society.” Along with promoting scientific excellence, NASI has worked extensively to cultivate scientific temper and take science beyond academic institutions—to students, women, rural and tribal communities, and society at large. To become a Fellow of an institution carrying this legacy feels very special. Behind this moment are many years of research, wonderful students and collaborators, mentors and colleagues, failed attempts, small breakthroughs, and the joy of continuing to ask new questions. I am deeply grateful to everyone who has been part of this journey. There is another beautiful coincidence. I will be inducted into the Fellowship in the holy city of Varanasi—my first visit to this ancient city. An inner journey towards liberation has long been underway within me. I wonder where my first encounter with Varanasi will take that journey. Varanasi—the land of Shiva and Kashi Vishwanath; a city associated for centuries with life, death, knowledge, faith and liberation. And for the Bengali in me, it is also the unforgettable Varanasi of Satyajit Ray—where Felu Mittir and Maganlal Meghraj engage in their memorable battle of intellect and nerve. I know that Varanasi today also painfully bears the burden of human pollution. Yet Kashi remains one of the greatest destinations of a spiritual journey. I look forward to being inducted into India’s oldest science academy in a city so deeply associated with knowledge, questioning and transcendence. Perhaps science and spirituality are very different journeys. Yet both begin with the humility of knowing how much remains unknown. Feeling happy, humbled, and grateful. Yours Prof. Arpita Patra, FNASc, #NASI #womaninscience #varanasi ASHISH CHOUDHURY
-
Anant Raj liked thisAnant Raj liked this10,000 AI agents versus Grigori Perelman. Perelman attacked the Poincaré Conjecture through profound conceptual innovation. He did not succeed by examining thousands of possible proofs until one happened to work. He understood the geometric structure of the problem deeply enough to develop the ideas needed to complete Hamilton’s Ricci-flow program. The OpenAI approach was almost the opposite. Roughly 10,000 agents were deployed in parallel to explore a huge number of possible mathematical directions, discard failures, exchange promising ideas, and continue searching until a successful construction emerged. If one reasoning trajectory is unreliable, run thousands. If it is unclear which direction is promising, investigate many of them simultaneously. If most attempts fail, the system can simply afford an enormous number of failures. In this sense, the AI approach resembles an extremely powerful search machine more than the concentrated mathematical insight that we associate with the greatest human mathematicians. Perhaps the next real breakthrough in AI mathematics will therefore not be deploying 100,000 agents instead of 10,000. It will be building an AI that understands a problem deeply enough to realize that 9,999 of them were unnecessary.
-
Anant Raj liked thisAnant Raj liked this#AISTATS 2027 Call for Paper is out! This year's edition is taking place in Montreal on May 3-6th, 2027 🇨🇦 Quentin Berthet and I are chairing We welcome your best submissions and workshop proposals! https://lnkd.in/gQ4gESZE New this year: AI review, and more! Abstract deadline: September 29, 2026 Paper deadline: October 6, 2026 Workshop proposal: October 20, 2026
-
Anant Raj liked thisAnant Raj liked thisSome happy news. My book, A Primer on Nonsmooth Convex Optimization, is finally out! This has been published by World Scientific as part of the IISc Lecture Notes Series. The book grew out of a course that I teach at IISc. I owe a big thanks to my former students, who encouraged me to turn the notes into a book that could reach students beyond the IISc classroom. It seemed like a great idea, but the actual job turned out to be far more difficult than I had anticipated. In the end, it took ten intense months of work, which makes it all the more gratifying to finally hold a copy in my hands. It was also a wonderful learning experience for me, giving me the opportunity to rethink the exposition and find simpler, more intuitive ways to explain some of the more difficult concepts. One challenge in teaching nonsmooth convex optimization is that the subject draws on several interconnected areas, including convex analysis, subdifferential calculus, Fenchel duality, monotone operator theory, and proximal algorithms. Doing justice to all of them in a three-month course or a short book is difficult. The present book follows a carefully chosen path through the subject, developing the core ideas from first principles and highlighting the connections among the different topics. The aim is to give readers a coherent view of the subject within a single semester, while keeping the treatment as rigorous and complete as possible. The book is written for graduate and advanced undergraduate students, but I hope it will also be useful to researchers working with optimization in signal processing, machine learning, statistics, control, and operations research. I have used numerous examples and counterexamples to motivate the main concepts and results. I have also interspersed carefully chosen and graded exercises throughout the text and collected solutions to many of them at the end. Special thanks to my students, Akash Mondal, Debraj Banerjee, and Trishit Mukherjee, for meticulously proofreading the drafts, to Debarghya Nandi for preparing the neat figures, and to Ishaq Hamza for volunteering to review the final draft. I am also grateful to S Lakshmivarahan, and my colleagues PS Sastry, Yadati Narahari, Chiranjib Bhattacharyya, Debraj Ghosh, Chandra Sekhar Seelamantula, and Chandra Murthy for their encouragement and advice. https://lnkd.in/g7VJ_Pzj #ConvexOptimization #NonsmoothOptimization #WorldScientific #IIScPress #IISc
-
Anant Raj liked thisAnant Raj liked this𝗦𝘁𝗮𝘁𝗶𝗻𝗴 𝗮𝗻 𝗮𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺 𝗺𝗮𝘆 𝘁𝗮𝗸𝗲 𝗮 𝗳𝗲𝘄 𝗺𝗶𝗻𝘂𝘁𝗲𝘀. 𝗣𝗿𝗼𝘃𝗶𝗻𝗴 𝘁𝗵𝗮𝘁 𝗶𝘁 𝗶𝘀 𝗴𝗼𝗼𝗱 𝗰𝗮𝗻 𝘁𝗮𝗸𝗲 𝗮 𝗳𝗲𝘄 𝗹𝗲𝗰𝘁𝘂𝗿𝗲𝘀. 📦 This contrast in "design" and analysis" is one of my favourite things about algorithms. In Bin Packing, one of the fundamental problems in CS, we are given items of different sizes and bins of fixed capacity. The goal is to pack all items using as few bins as possible. A very natural greedy algorithm is 𝗙𝗶𝗿𝘀𝘁 𝗙𝗶𝘁: — 𝗧𝗮𝗸𝗲 𝗲𝗮𝗰𝗵 𝗮𝗿𝗿𝗶𝘃𝗶𝗻𝗴 𝗶𝘁𝗲𝗺 𝗮𝗻𝗱 𝗽𝘂𝘁 𝗶𝘁 𝗶𝗻 𝘁𝗵𝗲 𝗳𝗶𝗿𝘀𝘁 𝗯𝗶𝗻 𝘄𝗵𝗲𝗿𝗲 𝗶𝘁 𝗳𝗶𝘁𝘀. Simple. But why does this extremely simple algorithm use at most about 1.7 times the optimal number of bins? The first analysis was given by a group of stalwarts in CS: Jeff Ullman (Turing Awardee), Ron Graham (coauthor of Concrete Mathematics), Michael Garey and David Johnson (who wrote the famous book on NP-completeness). The analysis is quite interesting and much trickier than the algorithm itself. There are many videos explaining Bin Packing and mention how First Fit works, but I found very few that explain why it works. In these two Algo-rindam videos (links in the comments), I explain the analysis (in fact, a simpler and natural analysis not present in the original paper) using a type of charging argument called "weight functions". Weight function is a powerful technique for analyzing approximation and online algorithms. Even more interestingly, the same proof can be understood through dual fitting and the configuration linear program (LP) for Bin Packing. The algorithm itself never solves an LP. The LP appears only in the analysis of why the algorithm works. I find that beautiful. I have always liked algorithms that are simple to state or design, but whose performance guarantees require surprisingly nontrivial analysis. QuickSort is a classic example. Karger’s randomized min-cut algorithm is another. For me, First Fit belongs to the same family. 𝗪𝗵𝗮𝘁 𝗶𝘀 𝘆𝗼𝘂𝗿 𝗳𝗮𝘃𝗼𝘂𝗿𝗶𝘁𝗲 𝗲𝘅𝗮𝗺𝗽𝗹𝗲 𝗼𝗳 𝗮 “𝘀𝗶𝗺𝗽𝗹𝗲 𝗮𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺, 𝗱𝗶𝗳𝗳𝗶𝗰𝘂𝗹𝘁 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀”? #Algorithms #BinPacking #ApproximationAlgorithms #LinearProgramming #TheoryCS
-
Anant Raj liked thisAnant Raj liked thisFirst PhD defence from my lab at Indian Institute of Science (IISc). Even though I had graduated 4 students at Indian Institute of Technology, Delhi, this one is special as it's the first one from the Indian Institute of Science (IISc). Congrats to Dr. Piyush Tiwary. Piyush joined me for a PhD right after his BTech from Indian Institute of Technology, Patna. Got PMRF fellowship. Worked on several aspects of EBM based GenAI. Interned at Adobe and Google DeepMind. Published at A* conferences. All within 5 years. He is a senior researcher at Dolby Laboratories. Interestingly, he hasn't crossed 25 years yet (reminds me of myself when I had a PhD before 25). Many Congratulations Piyush Tiwary and wish you well in your career. Seeing one's students reaching heights is one of the best things about being an academic. #phd #thesis #research
-
Anant Raj liked thisAnant Raj liked thisAt the start of this century, on the request of International Mathematical Union, Steve Smale listed 18 grand mathematical challenges for the 21st century. One of them was the Jacobian Conjecture. Today, it has been disproved by a counterexample by Claude over a tweet. This tweet alone is worth an award winning PhD. Ramanujan used to say mathematics came to him in dreams revealed by a goddess. It might often be over chat sessions from now on. P = NP proof on TikTok next week?
Experience & Education
Honors & Awards
-
Marie-Curie Global Fellowship
-
-
Max Planck Society Doctoral Fellowship
Max Planck Institute of Intelligent System, Tuebingen
-
TagMe - Machine Learning Programming Contest
Microsoft Bing Research India and Indian Institute of Science
Secured 1st position in this contest
-
Embedded Innovation of the year award
Intel India
Awarded a cash prize of Rs. 0.2 million for our project "eTab (The Emotional tablet)" at Intel India Embedded Chllenge 2012
-
Jury's Popular Choice Award
Intel India
Awarded a cash prize of Rs. 50K for our project "eTab (The Emotional tablet)" at Intel India Embedded Chllenge 2012
-
Embedded Design Challenge
IIT Kanpur
Won Embedded Design Challenge in Techkriti’ 12(Inter Collegiate Technical Festival) out of 100 teams countrywide
Languages
-
Hindi
Native or bilingual proficiency
-
English
Professional working proficiency
Organizations
-
Georgia Tech, School of Computational science and Engineering
Student Research Intern
-Worked on Large scale Kernel Methods for pattern analysis task.
Other similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More