2 papers
cs.LG2026
Drowning in Routine: Signal Dilution in Multi-Turn Agent Training
Yann Pernot, Vi Retault
Multi-turn agents interleave consequential decisions with routine execution: some actions change the downstream return distribution, while others are necessary but reward-equivalen…
cs.LG2025
Random Initialization of Gated Sparse Adapters
Vi Retault, Yohaï-Eliel Berreby
When fine-tuning language models on new tasks, catastrophic forgetting -- performance degradation on previously-learned tasks -- is a ubiquitous problem. While Parameter-Efficient…