5 papers
Provable Transformer Context Compression for In-Context Reinforcement Learning
Yajie Zhao, Yingjun Mo, Zhiyong Wang +3
Transformers can perform in-context reinforcement learning without parameter updates. However, existing theoretical constructions often require context sizes that grow with the num…
From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning
Yingjun Mo, Yajie Zhao, Fang Kong
Transformers can implement reinforcement-learning algorithms in context, yet how off-policy control emerges within Transformers remains poorly understood. We study Q-learning, a fo…
Tamed Euler-Maruyama method for SDEs with non-globally Lipschitz drift and multiplicative noise
Xiang Li, Yingjun Mo, Haoran Yang
Consider the following stochastic differential equation driven by multiplicative noise on with a superlinearly growing drift coefficient, \begin{align*} \mathrm{d} X…
Approximation to Deep Q-Network by Stochastic Delay Differential Equations
Jianya Lu, Yingjun Mo
Despite the significant breakthroughs that the Deep Q-Network (DQN) has brought to reinforcement learning, its theoretical analysis remains limited. In this paper, we construct a s…
Existence and non-existence of the CLT for a family of SDEs driven by stable process
Yingjun Mo, Yu Wang
Stochastic differential equations (SDEs) without global Lipschitz drift often demonstrate unusual phenomena. In this paper, we consider the following SDE on : \begin{a…