6 papers
Safety Hacking in Constrained Best-of- Inference-time Scaling
Akifumi Wachi, Takumi Tanabe, Youhei Akimoto
Inference-time pipelines often sample multiple outputs, filter them with a learned safety model, and return the proxy-feasible output with the highest learned reward. We show that…
Sample-Efficient Hypergradient Estimation for Decentralized Bi-Level Reinforcement Learning
Mikoto Kudo, Takumi Tanabe, Akifumi Wachi +1
Many strategic decision-making problems, such as environment design for warehouse robots, can be naturally formulated as bi-level reinforcement learning (RL), where a leader agent…
Cost-Minimized Label-Flipping Poisoning Attack to LLM Alignment
Shigeki Kusaka, Keita Saito, Mikoto Kudo +3
Large language models (LLMs) are increasingly deployed in real-world systems, making it critical to understand their vulnerabilities. While data poisoning attacks during RLHF/DPO a…
A Provable Approach for End-to-End Safe Reinforcement Learning
Akifumi Wachi, Kohei Miyaguchi, Takumi Tanabe +2
A longstanding goal in safe reinforcement learning (RL) is a method to ensure the safety of a policy throughout the entire process, from learning to operation. However, existing sa…
Vulnerability Mitigation for Safety-Aligned Language Models via Debiasing
Thien Q. Tran, Akifumi Wachi, Rei Sato +2
Safety alignment is an essential research topic for real-world AI applications. Despite the multifaceted nature of safety and trustworthiness in AI, current safety alignment method…
Stepwise Alignment for Constrained Language Model Policy Optimization
Akifumi Wachi, Thien Q. Tran, Rei Sato +2
Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs). This paper formulates human value alignment…