Meta tags:

Headings (most frequently used words):

research, scientist, gdm, sehwag, vikash, gmail, com, 2023, 2022, 2021, 2020, 2025, 2024,

Text of the page (most frequently used words):
and (45), the (38), vikash (28), #sehwag (28), learning (28), pdf (25), adversarial (21), models (21), robustness (20), machine (18), data (18), code (18), generative (17), from (16), for (15), diffusion (15), prateek (15), mittal (15), robust (14), 2023 (13), training (12), benchmark (10), 2021 (10), using (10), model (10), synthetic (9), that (9), with (9), 2024 (8), vision (8), princeton (8), github (8), 2022 (8), com (8), open (8), world (8), video (8), 2020 (7), robustbench (7), security (7), research (7), wang (7), stable (7), 2025 (7), networks (7), https (7), slides (7), performance (7), images (7), image (7), university (6), privacy (6), seminar (6), workshop (6), art (6), attacks (6), 000 (6), icml (6), neurips (6), time (6), more (5), cvpr (5), private (5), risks (5), train (5), under (5), detection (5), against (5), arjun (5), nitin (5), bhagoji (5), mung (5), chiang (5), state (5), webpage (5), developed (5), chen (5), lingjuan (5), lyu (5), gemini (5), than (4), zhenting (4), distribution (4), approach (4), how (4), inspire (4), group (4), demonstrate (4), framework (4), chong (4), xiang (4), standardized (4), our (4), test (4), saeed (4), mahloujifar (4), multiple (4), work (4), also (4), cost (4), scaling (4), watermark (4), reasoning (4), end (4), computer (3), one (3), series (3), trustworthy (3), edoardo (3), debenedetti (3), song (3), teaching (3), 2019 (3), deep (3), scientist (3), safety (3), google (3), representation (3), similarity (3), pruning (3), leaderboard (3), source (3), defense (3), artificial (3), sampling (3), develop (3), highly (3), samples (3), scale (3), showed (3), techniques (3), threat (3), show (3), real (3), dataset (3), differentially (3), propose (3), across (3), latent (3), black (3), box (3), compact (3), foundation (3), post (3), generation (3), development (3), june (2), core (2), talks (2), ieee (2), conference (2), other (2), services (2), christian (2), cianfarani (2), tinashe (2), handina (2), xianghao (2), kong (2), minzhou (2), pan (2), assistant (2), fall (2), intel (2), reu (2), academic (2), mar (2), generating (2), novel (2), low (2), density (2), regions (2), meta (2), aug (2), jan (2), enhancing (2), nov (2), non (2), improving (2), proxy (2), distributions (2), hydra (2), self (2), supervised (2), outlier (2), ssd (2), selected (2), proposed (2), liwei (2), chawin (2), sitawarin (2), daniel (2), cullina (2), fundamental (2), evaluation (2), fast (2), convergent (2), federated (2), improves (2), each (2), areas (2), communications (2), when (2), parameters (2), neural (2), adversarially (2), provable (2), patches (2), usenix (2), symposium (2), small (2), track (2), progress (2), has (2), been (2), francesco (2), croce (2), maksym (2), andriushchenko (2), nicolas (2), flammarion (2), detect (2), only (2), iclr (2), lower (2), bounds (2), cross (2), entropy (2), loss (2), datasets (2), generalization (2), blog (2), sihui (2), dai (2), improve (2), tail (2), adaptive (2), high (2), fidelity (2), augmentation (2), vits (2), transformers (2), attack (2), multirobustbench (2), benchmarking (2), adaptation (2), tong (2), xiangyu (2), jiachen (2), this (2), first (2), web (2), text (2), florian (2), eric (2), random (2), processes (2), better (2), while (2), simultaneously (2), xinyu (2), tang (2), ashwinee (2), panda (2), jailbreaking (2), artifacts (2), jailbreakbench (2), hyperparameter (2), optimization (2), linear (2), achieving (2), tasks (2), trace (2), generated (2), invisible (2), finding (2), quality (2), budget (2), 14x (2), combining (2), pixel (2), yet (2), versatile (2), compute (2), inference (2), are (2), arxiv (2), complexity (2), contributed (2), team (2), deepmind (2), publications (2), higher (2), content (2), genai (2), previous (2), reviewed, papers, major, conferences, journals, volunteered, junior, mentor, olcf, nvidia, gpu, hackathon, maintainers, organized, spml, program, committe, member, secure, organizer, iccv, arrow, mentored, ten, students, over, years, rajvardhan, oak, matteo, russo, wen, jie, ren, ele, 535, pattern, recognition, taught, mini, course, defenses, winterssion, ece, 574, lecture, basics, mentoring, made, practical, qualcomm, oct, embedding, make, reliable, epfl, primer, july, hard, instances, form, riken, aip, trustml, young, ciss, role, uchicago, uberkeley, may, distilled, msr, prospects, pitfalls, modern, perspective, aaai, feb, chatgpt, dalle, columbia, spark, buzzrobot, community, dlct, collective, invited, analysis, robust_representation_similarity, 137, 290, minimalistic, implementation, vsehwag, minimal, 730, 1512, microdiffusion, sonyresearch, micro_diffusion, repositories, vulnerability, examples, acm, intelligence, aisec, analyzing, conflict, between, objectives, uncertainty, critical, algorithm, called, folb, which, convergence, speed, smart, devices, round, journal, sac, hung, nguyen, seyyedali, hosseinalipour, christopher, brinton, vincent, poor, achieved, clean, accuracy, aggressively, shiqi, suman, jana, general, acheive, adversrial, patchguard, masks, receptive, fields, insightful, visited, 40k, users, matthias, hein, unlabeled, succesful, outliers, out, ssl, unified, provide, persence, common, poster, presence, adversaries, provides, termendous, boost, meets, can, metrics, such, cka, interesting, characteristics, compared, ben, zhao, haitao, zheng, understanding, through, lens, similarities, strong, generate, caner, hazirbas, albert, gordo, firat, ozgenel, cristian, canton, ferrer, contrary, conventional, wisdom, heavy, lighter, along, bag, tricks, achieves, satml, light, recipe, going, beyond, single, multi, vectors, pin, technique, aims, increases, exposure, feiran, jia, uncovering, was, works, significant, memorization, imagen, findings, further, motivated, deduplication, news, nicholas, carlini, jamie, hayes, milad, nasr, matthew, jagielski, tramèr, borja, balle, daphne, ippolito, wallace, extracting, pre, enables, during, finetuning, avoiding, leakage, associated, pretraining, spotlight, classification, priors, centralized, repository, date, benchmarks, patrick, chao, alexander, robey, edgar, dobriban, george, pappas, tramer, hamed, hassani, wong, large, language, consider, strategy, prvoides, hyperparameters, thus, reducing, nlp, new, rule, signature, autoencoders, back, dimitris, metaxas, shiqing, without, detector, wmd, method, annotation, free, setting, eccv, xin, dong, xue, lin, needles, haystack, reduction, publicly, available, 37m, not, requiring, any, proprietry, billion, jingtao, michael, spranger, stretching, dollar, scratch, micro, achieve, cues, semantics, like, number, fingers, subtle, level, release, spybench, comprehensive, top, 50k, wild, deepfakes, siyuan, cheng, zhang, spy, semantic, features, created, unifies, architecture, delivering, fraction, existing, weiming, zhuang, peter, stone, argus, dual, relationship, but, degrades, traces, visible, adversary, weichen, does, really, help, latest, featured, modalities, task, curation, improvements, pushing, frontier, advanced, multimodality, long, context, next, agentic, capabilities, current, efforts, focused, prior, spanned, five, key, conducted, depth, exploration, including, circumventing, its, sample, limits, demonstrating, effect, compression, transitioning, closed, domain, systems, widely, adopted, followed, account, most, recently, jailbreaks, llms, have, written, detailed, discussion, nuanced, distinction, approaches, towards, evals, demonstrated, preserving, methods, benchmarked, automated, unsafe, safer, concerned, vast, amount, online, identify, even, absence, watermarks, tracing, them, provenance, age, led, cut, less, comparable, early, multitask, democratizing, summary, received, phd, undergraduate, experience, includes, sony, microsoft, iit, kharagpur, overall, effectiveness, gmail, gdm,

Text of the page (random words):
vikash sehwag academic webpage vikash sehwag publications code blog research scientist gdm sehwag vikash gmail com i am a research scientist in gemini core post training team at google deepmind i work on improving reasoning rl and overall post training effectiveness i received my phd from princeton university and undergraduate from iit kharagpur my previous experience includes meta ai sony ai and microsoft research summary of previous research work my current research efforts are focused on enhancing reasoning and rl in gemini prior to this my research spanned five key areas democratizing end to end genai development i led end to end development of text to image generative models that cut training cost by 14x to less than 2 000 while achieving image generation quality comparable to early stable diffusion models 1 i also contributed to the development of a highly compact yet multitask foundation model for vision 2 data provenance in age of genai concerned by the vast amount of generative content online we developed techniques to identify synthetic samples 3 even in the absence of artificial watermarks and tracing them to source generative models 4 safer generative ai we demonstrated privacy risks in real world diffusion models and developed privacy preserving sampling and training methods 5 6 7 8 we also developed techniques and benchmarked automated generation of adversarial and unsafe content from generative models 9 10 benchmarking and evals we developed the widely adopted robustbench benchmark 11 followed by multirobustbench to account for multiple attacks 12 and most recently jailbreakbench 13 to benchmark progress on jailbreaks against llms we have also written a detailed discussion on nuanced similarity and distinction in security and safety approaches towards trustworthy ai 14 robust machine learning we conducted an in depth exploration of adversarial robust learning including circumventing its higher sample complexity using synthetic data 15 finding fundamental limits on robustness 16 demonstrating higher robustness with transformers 17 robustness across threat models 18 19 20 the effect of model scaling and compression 21 22 and adversarial risks in transitioning from closed domain to open world systems 23 24 publications 2025 gemini 2 5 pushing the frontier with advanced reasoning multimodality long context and next generation agentic capabilities gemini team google deepmind arxiv 2025 pdf our latest series of gemini models featured state of the art performance across modalities and task complexity my work contributed to post training data curation and rl improvements in the model does more inference time compute really help robustness tong wu chong xiang jiachen t wang weichen yu chawin sitawarin vikash sehwag prateek mittal arxiv 2025 pdf we show that inference time scaling has a dual relationship with robustness it improves robustness in a black box threat model but degrades robustness when reasoning traces are visible to the adversary argus a compact and versatile foundation model for vision weiming zhuang chen chen vikash sehwag peter stone lingjuan lyu cvpr 2025 pdf we created a compact yet versatile vision foundation model that unifies multiple vision tasks under one architecture delivering state of the art performance with a fraction of the data compute and parameters of existing models co spy combining semantic and pixel features to detect synthetic images by ai siyuan cheng lingjuan lyu zhenting wang xiangyu zhang vikash sehwag cvpr 2025 pdf code we achieve up to 34 better ai generated image detection by combining cues from image semantics like the number of fingers with subtle pixel level artifacts and release co spybench a comprehensive dataset of synthetic images from 22 top generative models and 50k in the wild deepfakes stretching each dollar diffusion training from scratch on a micro budget vikash sehwag xianghao kong jingtao li michael spranger lingjuan lyu cvpr 2025 pdf code we train a stable diffusion quality model with only 2 000 budget 14x cost reduction and publicly available 37m images not requiring any proprietry or billion image dataset 2024 finding needles in a haystack a black box approach to invisible watermark detection minzhou pan zhenting wang xin dong vikash sehwag lingjuan lyu xue lin eccv 2024 pdf we propose watermark detector wmd the first invisible watermark detection method under a black box and annotation free setting how to trace latent generative model generated images without artificial watermark zhenting wang vikash sehwag chen chen lingjuan lyu dimitris n metaxas shiqing ma icml 2024 pdf code using signature from the latent autoencoders we propose an approach to trace synthetic images back to the source latent generative model a new linear scaling rule for private adaptive hyperparameter optimization ashwinee panda xinyu tang vikash sehwag saeed mahloujifar prateek mittal icml 2024 pdf we consider the cost of hyperparameter optimization in differentially private learning and propose a strategy that prvoides linear scaling of hyperparameters thus reducing the privacy cost and simultaneously achieving state of the art performance across 22 benchmark tasks in cv and nlp jailbreakbench an open robustness benchmark for jailbreaking large language models patrick chao edoardo debenedetti alexander robey maksym andriushchenko francesco croce vikash sehwag edgar dobriban nicolas flammarion george j pappas florian tramer hamed hassani eric wong neurips 2024 datasets and benchmarks track pdf webpage code a centralized benchmark for 1 repository of jailbreaking attacks and artifacts 2 standardized evaluation framework and 3 up to date leaderboard 2023 differentially private image classification by learning priors from random processes xinyu tang ashwinee panda vikash sehwag prateek mittal neurips 2023 spotlight pdf code we show that pre training on data from random processes enables better performance during differentially private finetuning while simultaneously avoiding privacy leakage associated with real pretraining images extracting training data from diffusion models nicholas carlini jamie hayes milad nasr matthew jagielski vikash sehwag florian tramèr borja balle daphne ippolito eric wallace usenix security symposium 2023 pdf video news 1 2 3 4 this was one of the first works to demonstrate significant memorization of real world images in web scale text to image generative models stable diffusion imagen our findings further motivated web scale data deduplication in training dataset of generative models uncovering adversarial risks of test time adaptation tong wu feiran jia xiangyu qi jiachen t wang vikash sehwag saeed mahloujifar prateek mittal icml 2023 pdf webpage code we show that test time adaptation a technique that aims to improve performance at test time also increases exposure to novel security risks multirobustbench benchmarking robustness against multiple attacks sihui dai saeed mahloujifar chong xiang vikash sehwag pin yu chen prateek mittal icml 2023 pdf webpage code going beyond single attack robustness robustbench we develop a standardized benchmark for multi attack threat vectors a light recipe to train robust vision transformers edoardo debenedetti vikash sehwag prateek mittal satml 2023 pdf video slides code contrary to the conventional wisdom of using heavy data augmentation in vits we showed that a lighter data augmentation along with other bag of tricks achieves state of the art performance with vits adversarial training 2022 generating high fidelity data from low density regions using diffusion models vikash sehwag caner hazirbas albert gordo firat ozgenel cristian canton ferrer cvpr 2022 pdf our work showed strong generalization of diffusion models in the tail of the data distribution and developed adaptive sampling techniques to generate high fidelity samples from the tail of the data distribution understanding robust learning through the lens of representation similarities christian cianfarani arjun nitin bhagoji vikash sehwag ben y zhao prateek mittal haitao zheng neurips 2022 pdf video slides code using representation similarity metrics such as cka we demonstrate multiple interesting characteristics of adversarially robust networks compared to non robust networks robust learning meets generative models can proxy distributions improve adversarial robustness vikash sehwag saeed mahloujifar tinashe handina sihui dai chong xiang mung chiang prateek mittal iclr 2022 pdf video slides code blog we showed that synthetic data from diffusion model provides a termendous boost in the generalization performance of adversarial training 2021 lower bounds on cross entropy loss in the presence of test time adversaries arjun nitin bhagoji daniel cullina vikash sehwag prateek mittal icml 2021 pdf video slides poster code we provide lower bounds on cross entropy loss in the persence of adversarial attacks for common small scale computer vision datasets ssd a unified framework for self supervised outlier detection vikash sehwag mung chiang prateek mittal iclr 2021 neurips ssl workshop 2020 pdf video slides code using only unlabeled data we develop a highly succesful framework to detect outliers out of distribution samples robustbench a standardized adversarial robustness benchmark francesco croce maksym andriushchenko vikash sehwag nicolas flammarion mung chiang prateek mittal matthias hein neurips 2021 leaderboard pdf code we develop a standardized benchmark to track progress on adversarial robustness in deep learning our benchmark has been highly insightful and been visited by more than 40k users patchguard provable defense against adversarial patches using masks on small receptive fields chong xiang arjun nitin bhagoji vikash sehwag prateek mittal usenix security symposium 2021 pdf video code a general defense framework to acheive provable robustness against adversrial patches 2020 hydra pruning adversarially robust neural networks vikash sehwag shiqi wang prateek mittal suman jana neurips 2020 webpage pdf video slides code we achieved state of the art clean and robust accuracy when aggressively pruning the parameters of deep neural networks fast convergent federated learning hung t nguyen vikash sehwag seyyedali hosseinalipour christopher g brinton mung chiang h vincent poor ieee journal on selected areas in communications j sac series on machine learning for communications and networks 2020 pdf we proposed a fast convergent federated learning algorithm called folb which improves convergence speed by an smart sampling of devices in each round a critical evaluation of open world machine learning liwei song vikash sehwag arjun nitin bhagoji prateek mittal icml workshop on uncertainty robustness 2020 pdf code we demonstrate a fundamental conflict between the learning objectives of open world machine learning and adversarial robustness analyzing the robustness of open world machine learning vikash sehwag arjun nitin bhagoji liwei song chawin sitawarin daniel cullina mung chiang prateek mittal acm workshop on artificial intelligence and security aisec 2019 pdf slides code we demonstrate the vulnerability of open world machine learning models to adversarial examples and proposed a defense against the open world adversarial attacks selected open source repositories https github com sonyresearch micro_diffusion 1512 microdiffusion training stable diffusion in 2 000 https github com robustbench robustbench 730 robustbench leaderboard https github com vsehwag minimal diffusion 290 minimalistic implementation of diffusion models https github com inspire group ssd 137 self supervised outlier detection https github com inspire group hydra 92 pruning adversarial robust networks https github com inspire group proxy distributions 29 improving adversarial robustness using synthetic data https github com inspire group robust_representation_similarity 7 representation similarity analysis for robust and non robust networks invited talks how to train stable diffusion under 2 000 june 2025 dlct seminar ml collective how to train stable diffusion under 2 000 nov 2024 buzzrobot ai community how to train stable diffusion under 2 000 aug 2024 spark seminar google on safety risks of generative ai from chatgpt to dalle 3 nov 2023 columbia university prospects and pitfalls of modern generative models an ai safety perspective feb 2023 aaai enhancing machine learning using synthetic data distilled from generative models jan 2023 msr role of synthetic data in trustworthy machine learning may 2022 uchicago uberkeley a generative approach to robust machine learning mar 2022 ciss conference a generative approach to robust machine learning jan 2022 riken aip trustml young scientist seminar generating novel hard instances form low density regions using generative models aug 2021 meta ai a primer on adversarial machine learning july 2021 princeton intel reu seminar embedding data distribution to make machine learning more reliable mar 2021 epfl private deep learning made practical oct 2019 qualcomm academic services teaching and mentoring lecture on basics of adversarial machine learning princeton intel reu seminar 2021 teaching assistant for ece 574 security privacy fall 2021 princeton university taught a mini course on adversarial attacks defenses winterssion 2020 princeton university teaching assistant for ele 535 machine learning and pattern recognition fall 2019 princeton university mentored ten students in ai research over the years edoardo debenedetti rajvardhan oak christian cianfarani tinashe handina matteo russo xianghao kong song wen minzhou pan zhenting wang jie ren other services workshop organizer iccv 2023 arrow workshop cvpr 2023 workshop of adversarial machine learning on computer vision art of robustness 2023 program committe member for ieee conference on secure and trustworthy machine learning 2023 organized more than 20 talks on security privacy in machine learning spml seminar series 2022 one of the core maintainers of adversarial robustness benchmark robustbench github io volunteered as junior mentor at princeton olcf nvidia gpu hackathon june 2020 princeton university reviewed more than 50 papers for major computer vision and machine learning conferences and journals 2024 vikash sehwag