Yinglun Xu
YOU?
Author Swipe
View article: Learning a Pessimistic Reward Model in RLHF
Learning a Pessimistic Reward Model in RLHF Open
This work proposes `PET', a novel pessimistic reward fine-tuning method, to learn a pessimistic reward model robust against reward hacking in offline reinforcement learning from human feedback (RLHF). Traditional reward modeling techniques…
View article: Robust Thompson Sampling Algorithms Against Reward Poisoning Attacks
Robust Thompson Sampling Algorithms Against Reward Poisoning Attacks Open
Thompson sampling is one of the most popular learning algorithms for online sequential decision-making problems and has rich real-world applications. However, current Thompson sampling algorithms are limited by the assumption that the rewa…
View article: Binary Reward Labeling: Bridging Offline Preference and Reward-Based Reinforcement Learning
Binary Reward Labeling: Bridging Offline Preference and Reward-Based Reinforcement Learning Open
Offline reinforcement learning has become one of the most practical RL settings. However, most existing works on offline RL focus on the standard setting with scalar reward feedback. It remains unknown how to universally transfer the exist…
View article: Universal Black-Box Reward Poisoning Attack against Offline Reinforcement Learning
Universal Black-Box Reward Poisoning Attack against Offline Reinforcement Learning Open
We study the problem of universal black-boxed reward poisoning attacks against general offline reinforcement learning with deep neural networks. We consider a black-box threat model where the attacker is entirely oblivious to the learning …
View article: Two-Step Offline Preference-Based Reinforcement Learning with Constrained Actions
Two-Step Offline Preference-Based Reinforcement Learning with Constrained Actions Open
Preference-based reinforcement learning (PBRL) in the offline setting has succeeded greatly in industrial applications such as chatbots. A two-step learning framework where one applies a reinforcement learning step after a reward modeling …
View article: On the Robustness of Epoch-Greedy in Multi-Agent Contextual Bandit Mechanisms
On the Robustness of Epoch-Greedy in Multi-Agent Contextual Bandit Mechanisms Open
Efficient learning in multi-armed bandit mechanisms such as pay-per-click (PPC) auctions typically involves three challenges: 1) inducing truthful bidding behavior (incentives), 2) using personalization in the users (context), and 3) circu…
View article: Black-Box Targeted Reward Poisoning Attack Against Online Deep Reinforcement Learning
Black-Box Targeted Reward Poisoning Attack Against Online Deep Reinforcement Learning Open
We propose the first black-box targeted attack against online deep reinforcement learning through reward poisoning during training time. Our attack is applicable to general environments with unknown dynamics learned by unknown algorithms a…
View article: Efficient Reward Poisoning Attacks on Online Deep Reinforcement Learning
Efficient Reward Poisoning Attacks on Online Deep Reinforcement Learning Open
We study reward poisoning attacks on online deep reinforcement learning (DRL), where the attacker is oblivious to the learning algorithm used by the agent and the dynamics of the environment. We demonstrate the intrinsic vulnerability of s…
View article: Single-molecule optofluidic microsensor with interface whispering gallery modes
Single-molecule optofluidic microsensor with interface whispering gallery modes Open
Significance Optical microresonators have emerged as promising platforms for label-free detection of molecules. However, approaching optimum sensitivity is hindered due to the weak tail of evanescent fields. Here, we report the implementat…