Metod Jazbec: Reinforcement Learning for Sampling in Diffusion LLMs

V torek, 24. 3. 2026, ob 16:30 bo v predavalnici 3.06 v okviru seminarja VeSFiM potekalo predavanje Metoda Jazbeca (University of Amsterdam) z naslovom Reinforcement Learning for Sampling in Diffusion LLMs.

Povzetek: I'll talk about diffusion LLMs (dLLMs), how they are different from traditional autoregressive LLMs, and what is the current state-of-affairs for sampling in dLLMs. Then I'll present our recent project where we propose to "learn" sampling strategies using reinforcement learning: https://arxiv.org/abs/2512.09106

Predavanje bo potekalo v živo.

Vljudno vabljeni!