Doina Precup

Robust Reward Modeling via Causal Rubrics

Balancing Plasticity and Stability with Fast and Slow Successor Features

On The Geometry and Topology of Representations: the Manifolds of Modular Addition

SCAR: Shapley Credit Assignment for More Efficient RLHF

RL Fine-Tuning Heals OOD Forgetting in SFT

Permanent and Transient Representations for Continual Reinforcement Learning

REPO: Detoxifying LLMs via Representation Erasure-based Preference Optimization

Training Language Models to Self-Correct via Reinforcement Learning

MaestroMotif: Skill Design from Artificial Intelligence Feedback

Uncovering a Universal Abstract Algorithm for Modular Addition in Neural Networks

Capturing Individual Human Preferences with Reward Features

Plasticity as the Mirror of Empowerment

Language Agents Mirror Human Causal Reasoning Biases. How Can We Help Them Think Like Scientists?

Exploring Sparse Adapters for Scalable Merging of Parameter Efficient Experts

Langevin Soft Actor-Critic: Efficient Exploration through Uncertainty-Driven Critic Learning

Selective Unlearning via Representation Erasure Using Domain Adversarial Training

Long-Term Fairness in Reinforcement Learning with Bisimulation Metrics

Identifying and Addressing Delusions for Target-Directed Decision Making

Interpreting neural networks depends on the level of abstraction: Revisiting modular addition

Rejecting Hallucinated State Targets during Planning

EnzymeFlow: Generating Reaction-specific Enzyme Catalytic Pockets through Flow Matching and Co-Evolutionary Dynamics

Learning subgoal representations from state graphs in goal-conditioned hierarchical reinforcement learning