Meta tags:
author=Metod Jazbec;
description=Academic website of Metod Jazbec ;

Headings (most frequently used words):

metod, jazbec, selected, publications,

Text of the page (most frequently used words):
and (25), the (24), that (12), diffusion (10), models (9), sampling (8), metod (7), #jazbec (7), samples (7), with (6), performance (6), for (6), quality (6), generative (6), how (5), when (5), this (5), our (5), language (5), learning (4), their (4), such (4), inference (4), risk (4), control (4), can (4), while (4), code (4), html (4), abs (4), eric (4), nalisnick (4), have (4), propose (4), uncertainty (4), based (4), simple (4), dllms (4), efficient (4), work (4), heuristics (4), 2026 (3), however (3), early (3), exit (3), eenns (3), produce (3), yet (3), fast (3), address (3), post (3), computational (3), bayesian (3), during (3), autoregressive (3), confidence (3), unmasking (3), diversity (3), scaling (2), machine (2), significantly (2), cost (2), being (2), solution (2), they (2), issue (2), without (2), safe (2), eenn (2), hoc (2), exiting (2), empirically (2), tasks (2), dan (2), zhang (2), christian (2), naesseth (2), modeling (2), state (2), art (2), high (2), individual (2), remains (2), framework (2), large (2), sample (2), experiments (2), demonstrate (2), proposed (2), existing (2), model (2), stephan (2), mandt (2), best (2), paper (2), iclr (2), match (2), downstream (2), which (2), tokens (2), has (2), both (2), token (2), block (2), using (2), dllm (2), show (2), these (2), policies (2), outperforming (2), theo (2), olausson (2), icml (2), between (2), tempered (2), remasking (2), pass (2), time (2), diverse (2), from (2), publications (2), phd (2), data (2), copyright, powered, theme, folio, jekyll, don, stranger, reach, out, improves, gains, come, slow, resource, intensive, neural, networks, offer, promising, accelerate, allowing, intermediate, layers, prediction, fundamental, determine, severely, degrading, other, words, investigate, adapt, frameworks, offers, distribution, free, tunes, mechanism, exits, only, occur, output, sufficient, validate, insights, range, vision, demonstrating, substantial, savings, all, preserving, user, specified, goals, 2024, alexander, timans, tin, hadži, veljković, kaspar, sakmann, neurips, recently, driven, significant, breakthroughs, average, still, low, detecting, human, inspection, challenging, task, estimating, synthetic, outline, make, practical, modern, introduce, new, semantic, likelihood, evaluated, latent, space, feature, extractor, challenges, posed, dimensional, spaces, through, effectively, identifies, poor, outperforms, methods, notably, applied, any, pretrained, flow, matching, via, laplace, approximation, effective, techniques, minimize, its, overhead, 2025, eliot, wong, toi, guoxuan, xia, question, uai, now, counterparts, many, holding, promise, more, one, critical, design, aspect, procedure, selects, unmask, each, step, indeed, recent, found, heuristic, strategies, thresholding, improve, throughput, compared, random, downsides, require, manual, tuning, observe, degrades, larger, sizes, instead, train, procedures, reinforcement, specifically, formalize, masked, markov, decision, process, serves, environment, lightweight, policy, single, layer, transformer, maps, confidences, decisions, trained, combined, semi, generation, them, full, setting, louis, béthune, pierre, ablin, michael, kirchhof, joão, monteiro, victor, turrisi, jason, ramapuram, marco, cuturi, delta, oral, much, been, done, designing, accurate, efforts, largely, focused, tradeoff, speed, additionally, ensure, across, less, well, understood, increased, softened, versions, familiar, retaining, benefits, offering, implementations, motivate, approach, introducing, idealized, formal, fork, studying, impact, expected, entropy, forks, close, exploration, gap, hence, controlling, nfe, further, study, increase, translates, training, test, compute, overall, findings, possible, wang, armando, solar, lezama, tale, two, temperatures, colm, selected, spent, research, intern, microsoft, cambridge, apple, paris, before, completed, masters, science, eth, zurich, worked, industry, scientist, quantitative, engineer, student, university, amsterdam, working, robust, amlab, about, current, toggle, navigation,

Text of the page (random words):
metod jazbec toggle navigation about current publications cv metod jazbec i am a machine learning phd student at amlab university of amsterdam working with eric nalisnick on efficient and robust generative modeling during my phd i have spent time as a research intern at microsoft cambridge uk and apple paris before that i completed a masters in data science at eth zurich and worked in industry as a data scientist and quantitative engineer selected publications colm a tale of two temperatures simple efficient and diverse sampling from diffusion language models theo x olausson metod jazbec xi wang armando solar lezama christian a naesseth stephan mandt and eric nalisnick 2026 abs html code much work has been done on designing fast and accurate sampling for diffusion language models dllms however these efforts have largely focused on the tradeoff between speed and quality of individual samples how to additionally ensure diversity across samples remains less well understood in this work we show that diversity can be increased by using softened tempered versions of familiar confidence based remasking heuristics retaining their computational benefits and offering simple implementations we motivate this approach by introducing an idealized formal model of fork tokens and studying the impact of remasking on the expected entropy at the forks empirically the proposed tempered heuristics close the exploration gap pass k between existing confidence based and autoregressive sampling hence outperforming both when controlling for cost pass nfe we further study how the increase in diversity translates to downstream post training and test time compute scaling overall our findings demonstrate that simple efficient and diverse sampling from dllms is possible icml oral icml best paper delta iclr learning unmasking policies for diffusion language models metod jazbec theo x olausson louis béthune pierre ablin michael kirchhof joão monteiro victor turrisi jason ramapuram and marco cuturi 2026 abs html code diffusion large language models dllms now match the downstream performance of their autoregressive counterparts on many tasks while holding the promise of being more efficient during inference one critical design aspect of dllms is the sampling procedure that selects which tokens to unmask at each diffusion step indeed recent work has found that heuristic strategies such as confidence thresholding improve both sample quality and token throughput compared to random unmasking however such heuristics have downsides they require manual tuning and we observe that their performance degrades with larger block sizes in this work we instead propose to train sampling procedures using reinforcement learning specifically we formalize masked diffusion sampling as a markov decision process in which the dllm serves as the environment and propose a lightweight policy based on a single layer transformer that maps dllm token confidences to unmasking decisions our experiments show that these trained policies match the performance of state of the art heuristics when combined with semi autoregressive block generation while outperforming them in the full diffusion setting uai best paper question iclr generative uncertainty in diffusion models metod jazbec eliot wong toi guoxuan xia dan zhang eric nalisnick and stephan mandt 2025 abs html code diffusion models have recently driven significant breakthroughs in generative modeling while state of the art models produce high quality samples on average individual samples can still be low quality detecting such samples without human inspection remains a challenging task to address this we propose a bayesian framework for estimating generative uncertainty of synthetic samples we outline how to make bayesian inference practical for large modern generative models and introduce a new semantic likelihood evaluated in the latent space of a feature extractor to address the challenges posed by high dimensional sample spaces through our experiments we demonstrate that the proposed generative uncertainty effectively identifies poor quality samples and significantly outperforms existing uncertainty based methods notably our bayesian framework can be applied post hoc to any pretrained diffusion or flow matching model via the laplace approximation and we propose simple yet effective techniques to minimize its computational overhead during sampling neurips fast yet safe early exiting with risk control metod jazbec alexander timans tin hadži veljković kaspar sakmann dan zhang christian a naesseth and eric nalisnick 2024 abs html code scaling machine learning models significantly improves their performance however such gains come at the cost of inference being slow and resource intensive early exit neural networks eenns offer a promising solution they accelerate inference by allowing intermediate layers to exit and produce a prediction early yet a fundamental issue with eenns is how to determine when to exit without severely degrading performance in other words when is it safe for an eenn to go fast to address this issue we investigate how to adapt frameworks of risk control to eenns risk control offers a distribution free post hoc solution that tunes the eenn s exiting mechanism so that exits only occur when the output is of sufficient quality we empirically validate our insights on a range of vision and language tasks demonstrating that risk control can produce substantial computational savings all the while preserving user specified performance goals don t be a stranger reach out copyright 2026 metod jazbec powered by jekyll with al folio theme