Poster · Paris

Control Reinforcement Learning: Token-Level Mechanistic Analysis via Learned SAE Feature Steering

Name: Control Reinforcement Learning: Token-Level Mechanistic Analysis via Learned SAE Feature Steering
Start: 2026-12-10T17:30:00+01:00
End: 2026-12-10T19:30:00+01:00
Location: Paris Poster Hall, Paris

When: Thu 10 Dec 2026, 17:30–19:30 CET
Where: Paris Poster Hall
Session: Paris Poster Session 4
Authors: Seonglae Cho, Zekun Wu, Adriano Koshiyama
Organisations: Holistic AI, University College London
QRS classification: Reinforcement learning, Language models

Open in schedule Official page