Meta tags:
description=Long (Tony) Lian s Personal Website;
keywords=Long Lian, Tony Lian, UC Berkeley;
author=Long Lian;

Headings (most frequently used words):

models, for, diffusion, image, and, language, learning, long, parallel, video, llm, attention, unsupervised, adaptive, efficient, reasoning, with, self, grounded, understanding, scalable, diverse, benchmarks, constantly, improving, from, by, visual, tony, lian, publications, equal, contribution, side, projects, threadweaver, threading, in, describe, anything, detailed, localized, captioning, crossmae, rethinking, patch, dependence, masked, autoencoders, correcting, controlled, enhancing, prompt, of, text, to, large, attend, before, via, autoregressive, gazing, v1, unifying, generation, verification, reasoners, visgym, customizable, environments, multimodal, agents, visually, prompted, are, surprisingly, fragile, pillar, new, frontier, radiology, foundation, need, tulip, towards, unified, pretraining, atlas, multi, scale, improves, context, modeling, universal, segmentation, quantizing, bootstrapping, objectness, videos, relaxed, common, fate, grouping, selective, labeling, more, effective, semi, supervised, debiased, naturally, imbalanced, pseudo, labels, invariance, reinforcement, tailed, recognition, routing, distribution, aware, experts, stable, xl, demo, webui, animegan, js, rainbow,

Text of the page (most frequently used words):
and (38), long (29), code (25), lian (24), for (23), paper (23), bibtex (22), learning (19), with (18), image (17), #project (16), conference (15), wang (15), trevor (15), darrell (15), models (14), the (12), video (12), diffusion (10), xudong (10), computer (10), adam (10), yala (10), vision (9), generation (9), international (8), reasoning (8), parallel (8), demo (7), that (7), visual (7), boyi (7), text (7), language (7), research (7), from (6), based (6), liu (6), stella (6), recognition (6), tasks (6), framework (5), poster (5), data (5), training (5), attention (5), cvpr (5), ieee (5), cvf (5), pattern (5), 2024 (5), efficient (5), understanding (5), alane (5), suhr (5), david (5), model (5), 2026 (5), improving (5), reinforcement (4), iclr (4), unsupervised (4), adaptive (4), any (4), xiuyu (4), multi (4), scale (4), 2025 (4), chan (4), large (4), scalable (4), self (4), machine (4), llm (4), universal (3), diverse (3), improves (3), accuracy (3), inference (3), representations (3), method (3), label (3), without (3), videos (3), high (3), kurt (3), keutzer (3), han (3), atlas (3), new (3), modeling (3), significantly (3), benchmarks (3), lin (3), multimodal (3), verification (3), before (3), enabling (3), baifeng (3), shi (3), complex (3), llms (3), threadweaver (3), prof (3), free (2), use (2), this (2), website (2), are (2), implementation (2), rainbow (2), animegan (2), webui (2), stable (2), proposed (2), tailed (2), routing (2), experts (2), speed (2), 2021 (2), distribution (2), aware (2), policy (2), test (2), shift (2), invariance (2), task (2), towards (2), novel (2), effective (2), debiased (2), naturally (2), imbalanced (2), pseudo (2), labels (2), 2022 (2), zhirong (2), semi (2), supervised (2), small (2), compute (2), annotation (2), efficiency (2), stage (2), 2023 (2), quality (2), time (2), iccv (2), zhang (2), unified (2), segmentation (2), supervision (2), roei (2), herzig (2), using (2), achieves (2), resolution (2), kumar (2), krishna (2), agrawal (2), longchao (2), natalia (2), harguindeguy (2), maggie (2), chung (2), tulip (2), contrastive (2), reconstruction (2), stronger (2), workshop (2), real (2), jiaxin (2), sewon (2), min (2), constantly (2), pillar (2), radiology (2), foundation (2), across (2), marker (2), size (2), lisa (2), dunlap (2), renhao (2), visgym (2), interaction (2), environments (2), agents (2), letian (2), ken (2), goldberg (2), joseph (2), gonzalez (2), reasoners (2), scaling (2), through (2), icml (2), reducing (2), yin (2), enhancing (2), prompt (2), tmlr (2), transactions (2), grounded (2), prompts (2), correcting (2), patch (2), describe (2), anything (2), detailed (2), descriptions (2), latency (2), yuandong (2), tian (2), victoria (2), where (2), agent (2), end (2), berkeley (2), tony (2), feel, references, designs, fonts, uses, bootstrap, jekyll, jon, barron, haozhi, source, algorithm, parl, which, converts, photos, anime, style, online, gradio, allows, playing, around, locally, colab, sdxl, side, projects, leveraging, bias, variance, tradeoff, classification, distributed, ride, both, spotlight, zhongqi, miao, ziwei, rather, than, invariant, train, vai, disperse, interference, irrelevant, factors, robust, distractions, propose, counterfactual, margins, deal, undesired, effect, focus, selecting, right, query, information, our, demonstrates, spent, careful, labeled, selection, brings, big, eccv, european, selective, labeling, more, segmenting, objects, human, bootstrapping, objectness, relaxed, common, fate, grouping, running, bit, weights, first, yijiang, huanrui, yang, zhen, dong, daniel, kang, shanghang, quantizing, adept, performing, various, instance, semantic, panoptic, dantong, niu, xinyang, neural, network, cross, state, art, better, alexander, bick, context, generative, augmentation, richer, regularization, fine, grained, mmfm4, zineng, tang, seun, eisape, pretraining, echo, constructs, world, social, media, evidence, surfacing, emerging, differentiation, community, informed, metrics, grace, luo, heekyung, lee, nishant, malpani, aleksander, holynski, need, trained, volumetric, mri, paired, rate, extraction, hundreds, radiologic, findings, michael, nercessian, yufu, peter, mikhael, gigin, lecia, sequist, florian, fintelmann, yutong, bai, frontier, vpbench, shows, prompting, details, including, color, dataset, jpeg, compression, can, substantially, vlm, benchmark, performance, rankings, haiwen, feng, jiahao, shu, angjoo, kanazawa, visually, prompted, surprisingly, fragile, suite, horizon, evaluating, symbolic, puzzles, navigation, manipulation, best, award, intelligence, zirui, junyi, ion, stoica, customizable, unifies, pairwise, tournament, ranking, joint, harman, singh, kusha, sareen, monishwaran, maheswaran, sijun, tan, xiaoxia, junxiong, alpay, ariyak, qingyang, samir, khaki, rishabh, tiwari, yucheng, ben, athiwaratkun, unifying, autogaze, removes, redundant, patches, vit, mllm, processing, tokens, 100x, stephanie, hanrong, eigen, aaron, reite, jan, kautz, song, pavlo, molchanov, hongxu, attend, via, autoregressive, gazing, blog, capabilities, grounding, featured, certification, make, plans, actual, achieving, realistic, align, input, iteratively, corrects, inaccuracies, integrating, alignment, editing, tsung, controlled, independent, partial, facilitates, representation, alexei, efros, crossmae, rethinking, dependence, masked, autoencoders, dam, generates, user, specified, regions, images, marked, points, boxes, scribbles, masks, introduce, dlc, bench, evaluate, such, region, yifan, ding, yunhao, sifei, hanzi, mao, marco, pavone, ming, cui, localized, captioning, demonstrate, dimension, giving, spawn, join, functions, control, when, reason, sequentially, lower, improved, scalability, colm, jiayi, pan, charlie, victor, snell, yifei, zhou, par, cutting, edge, sequential, while, utilizes, two, trajectory, generator, trie, design, parallelization, grpo, oral, sida, felix, juefei, tsu, jui, threading, publications, equal, contribution, member, technical, staff, work, post, mllms, primarily, focuses, developing, systems, core, contributor, system, release, obtained, phd, eecs, advised, was, scientist, intern, meta, genai, working, interned, nvidia, hold, science, conducted, under, during, undergraduate, studies, deep, imagination, team, bair, thinking, machines, lab, personal,

Text of the page (random words):
long tony lian s personal website long tony lian i am a member of technical staff at thinking machines lab where i work on post training llms mllms my research primarily focuses on developing scalable multi agent systems through end to end rl i am a core contributor for the multi agent system in the interaction models release i obtained my phd in eecs at uc berkeley and bair advised by prof trevor darrell and prof adam yala i was a research scientist intern at meta genai with victoria lin and yuandong tian working on rl for reasoning llms in 2025 i interned with the deep imagination research team at nvidia research in 2024 i hold a b a in computer science from uc berkeley where i conducted research under the supervision of prof stella yu during my undergraduate studies publications equal contribution threadweaver adaptive threading for efficient parallel reasoning in language models long lian sida wang felix juefei xu tsu jui fu xiuyu li adam yala trevor darrell alane suhr yuandong tian xi victoria lin international conference on machine learning icml 2026 oral threadweaver is a framework for adaptive parallel reasoning that achieves accuracy on par with cutting edge sequential reasoning models while significantly reducing inference latency threadweaver utilizes a two stage parallel trajectory generator trie based training inference co design and parallelization aware reinforcement learning p grpo paper project code bibtex tl dr learning adaptive parallel reasoning with language models jiayi pan xiuyu li long lian charlie victor snell yifei zhou adam yala trevor darrell kurt keutzer alane suhr conference on language modeling colm 2026 we demonstrate a new dimension of scaling parallel reasoning by giving llms spawn and join functions to control when to reason sequentially or in parallel enabling lower latency and improved scalability in complex reasoning tasks paper code bibtex tl dr describe anything detailed localized image and video captioning long lian yifan ding yunhao ge sifei liu hanzi mao boyi li marco pavone ming yu liu trevor darrell adam yala yin cui international conference on computer vision iccv 2025 describe anything model dam generates detailed descriptions for user specified regions in images and videos marked by points boxes scribbles or masks we introduce dlc bench to evaluate such region based descriptions paper project demo code bibtex tl dr crossmae rethinking patch dependence for masked autoencoders letian fu long lian renhao wang baifeng shi xudong wang adam yala trevor darrell alexei a efros ken goldberg transactions on machine learning research tmlr 2025 independent partial patch reconstruction facilitates efficient representation learning paper project code bibtex tl dr self correcting llm controlled diffusion models tsung han wu long lian joseph e gonzalez boyi li trevor darrell ieee cvf conference on computer vision and pattern recognition cvpr 2024 a text to image generation framework that iteratively corrects inaccuracies by integrating a self correcting llm with diffusion models significantly improving image alignment with complex prompts and enabling image editing tasks paper project code bibtex tl dr llm grounded video diffusion models long lian baifeng shi adam yala trevor darrell boyi li international conference on learning representations iclr 2024 improving text to video generation by using a large language model to make plans before the actual video generation achieving realistic video generation that align with complex input prompts paper project code bibtex tl dr llm grounded diffusion enhancing prompt understanding of text to image diffusion models with large language models long lian boyi li adam yala trevor darrell transactions on machine learning research tmlr 2024 featured certification enhancing the prompt understanding capabilities of text to image diffusion models with large language models for grounding paper blog project demo code poster bibtex tl dr attend before attention efficient and scalable video understanding via autoregressive gazing baifeng shi stephanie fu long lian hanrong ye david eigen aaron reite boyi li jan kautz song han david m chan pavlo molchanov trevor darrell hongxu yin autogaze removes redundant video patches before vit or mllm processing reducing visual tokens by 4x 100x and enabling efficient long high resolution video understanding paper project demo code bibtex tl dr v1 unifying generation and self verification for parallel reasoners harman singh xiuyu li kusha sareen monishwaran maheswaran sijun tan xiaoxia wu junxiong wang alpay ariyak qingyang wu samir khaki rishabh tiwari long lian yucheng lu boyi li alane suhr ben athiwaratkun kurt keutzer international conference on machine learning icml 2026 v1 unifies generation and pairwise self verification for parallel reasoners improving test time scaling through tournament based ranking and reinforcement learning for joint generation verification paper project bibtex tl dr visgym diverse customizable scalable environments for multimodal agents zirui wang junyi zhang jiaxin ge long lian letian fu lisa dunlap ken goldberg xudong wang ion stoica david m chan sewon min joseph e gonzalez iclr workshop on multimodal intelligence 2026 best paper award visgym is a suite of 17 long horizon visual interaction environments for evaluating and training multimodal agents across symbolic puzzles real image tasks navigation and manipulation paper project code bibtex tl dr visually prompted benchmarks are surprisingly fragile haiwen feng long lian lisa dunlap jiahao shu xudong wang renhao wang trevor darrell alane suhr angjoo kanazawa vpbench shows that small visual prompting details including marker color marker size dataset size and jpeg compression can substantially shift vlm benchmark performance and rankings paper project code bibtex tl dr pillar 0 a new frontier for radiology foundation models kumar krishna agrawal longchao liu long lian michael nercessian natalia harguindeguy yufu wu peter mikhael gigin lin lecia v sequist florian fintelmann trevor darrell yutong bai maggie chung adam yala pillar 0 is a radiology foundation model trained on large scale volumetric ct and mri data paired with rate for scalable label extraction across hundreds of radiologic findings paper project code bibtex tl dr constantly improving image models need constantly improving benchmarks jiaxin ge grace luo heekyung lee nishant malpani long lian xudong wang aleksander holynski trevor darrell sewon min david m chan international conference on learning representations iclr 2026 echo constructs image generation benchmarks from real world social media evidence of model use surfacing emerging tasks stronger model differentiation and community informed quality metrics paper project code bibtex tl dr tulip towards unified language image pretraining zineng tang long lian seun eisape xudong wang roei herzig adam yala alane suhr trevor darrell david m chan international conference on computer vision mmfm4 workshop 2025 tulip improves contrastive image text learning with generative data augmentation richer image image and text text contrastive learning and reconstruction regularization for stronger fine grained visual understanding paper project code bibtex tl dr atlas multi scale attention improves long context image modeling kumar krishna agrawal long lian longchao liu natalia harguindeguy boyi li alexander bick maggie chung trevor darrell adam yala atlas is a new neural network using multi scale attention for efficient cross scale image modeling atlas achieves state of the art accuracy with significantly better speed and compute efficiency on high resolution image tasks paper code bibtex tl dr unsupervised universal image segmentation dantong niu xudong wang xinyang han long lian roei herzig trevor darrell ieee cvf conference on computer vision and pattern recognition cvpr 2024 a novel unified framework adept at performing various instance semantic and panoptic segmentation tasks without any supervision paper code bibtex tl dr q diffusion quantizing diffusion models xiuyu li yijiang liu long lian huanrui yang zhen dong daniel kang shanghang zhang kurt keutzer international conference on computer vision iccv 2023 running stable diffusion in 4 bit weights with high generation quality for the first time paper project code bibtex tl dr bootstrapping objectness from videos by relaxed common fate and visual grouping long lian zhirong wu stella x yu ieee cvf conference on computer vision and pattern recognition cvpr 2023 segmenting objects in videos without any human annotation in any training stage paper project video demo video code poster bibtex tl dr unsupervised selective labeling for more effective semi supervised learning xudong wang long lian stella x yu european conference on computer vision eccv 2022 we focus on selecting the right data to query for label for semi supervised learning without any label or task information our method demonstrates that a small compute spent on careful labeled data selection brings big annotation efficiency paper video code poster bibtex tl dr debiased learning from naturally imbalanced pseudo labels xudong wang zhirong wu long lian stella x yu ieee cvf conference on computer vision and pattern recognition cvpr 2022 we propose a novel and effective debiased learning method based on counterfactual reasoning and adaptive margins to deal with the undesired effect from naturally imbalanced pseudo labels paper code bibtex tl dr unsupervised visual attention and invariance for reinforcement learning xudong wang long lian stella x yu ieee cvf conference on computer vision and pattern recognition cvpr 2021 rather than training a universal rl policy invariant to train test distribution shift we proposed unsupervised visual attention and invariance method vai to disperse interference task irrelevant factors towards a rl policy robust to distractions paper code poster bibtex tl dr long tailed recognition by routing diverse distribution aware experts xudong wang long lian zhongqi miao ziwei liu stella x yu international conference on learning representations iclr 2021 spotlight leveraging bias variance tradeoff we proposed a universal classification framework for long tailed distributed data with routing diverse experts ride that improves both accuracy and inference speed paper video code poster bibtex tl dr side projects stable diffusion xl demo webui a gradio based webui that allows playing around with sdxl locally and on colab for free code demo animegan js an implementation of animegan which converts photos to anime style online with tf js code demo rainbow an implementation of rainbow algorithm with parl reinforcement learning framework code feel free to use the source code of this website references designs from haozhi qi and jon barron the fonts are based on this project uses jekyll and bootstrap