REVIEW 2 cited by
Imitating Language via Scalable Inverse Reinforcement Learning
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
The majority of language model training builds on imitation learning. It covers pretraining, supervised fine-tuning, and affects the starting conditions for reinforcement learning from human feedback (RLHF). The simplicity and scalability of maximum likelihood estimation (MLE) for next token prediction led to its role as predominant paradigm. However, the broader field of imitation learning can more effectively utilize the sequential structure underlying autoregressive generation. We focus on investigating the inverse reinforcement learning (IRL) perspective to imitation, extracting rewards and directly optimizing sequences instead of individual token likelihoods and evaluate its benefits for fine-tuning large language models. We provide a new angle, reformulating inverse soft-Q-learning as a temporal difference regularized extension of MLE. This creates a principled connection between MLE and IRL and allows trading off added complexity with increased performance and diversity of generations in the supervised fine-tuning (SFT) setting. We find clear advantages for IRL-based imitation, in particular for retaining diversity while maximizing task performance, rendering IRL a strong alternative on fixed SFT datasets even without online data generation. Our analysis of IRL-extracted reward functions further indicates benefits for more robust reward functions via tighter integration of supervised and preference-based LLM post-training.
Forward citations
Cited by 2 Pith papers
-
Value from Observations: Towards Large-Scale Imitation Learning via Self-Improvement
VfO trains a state-value function on action-free expert demonstrations mixed with lower-quality background data, then uses advantage-weighted regression on the background data to improve the agent, approaching oracle ...
-
Inverse Reinforcement Learning with Switching Rewards and History Dependency for Characterizing Animal Behaviors
SWIRL adds state-dependent mode switching and reward functions that depend on recent state history to inverse RL, improving test likelihood on simulated and labyrinth mouse behavior, but not on spontaneous behavior syllables.
Discussion (0). Continue with ORCID to comment.