Model-based methods in reinforcement learning offer a promising approach to enhance data efficiency by facilitating policy exploration within a dynamics model. However, accurately predicting sequential steps in the dynamics model remains a challenge due to the bootstrapping prediction, which attributes the next state to the prediction of the current state. This leads to accumulated errors during model roll-out. In this paper, we propose the Any-step Dynamics Model (ADM) to mitigate the compounding error by reducing bootstrapping prediction to direct prediction. ADM allows for the use of variable-length plans as inputs for predicting future states without frequent bootstrapping. We design two algorithms, ADMPO-ON and ADMPO-OFF, which apply ADM in online and offline model-based frameworks, respectively. In the online setting, ADMPO-ON demonstrates improved sample efficiency compared to previous state-of-the-art methods. In the offline setting, ADMPO-OFF not only demonstrates superior performance compared to recent state-of-the-art offline approaches but also offers better quantification of model uncertainty using only a single ADM.

One-shot imitation learning (OSIL) is to learn an imitator agent that can execute multiple tasks with only a single demonstration. In real-world scenario, the environment is dynamic, e.g., unexpected changes can occur after demonstration. Thus, achieving generalization of the imitator agent is crucial as agents would inevitably face situations unseen in the provided demonstrations. While traditional OSIL methods excel in relatively stationary settings, their adaptability to such unforeseen changes, which asking for a higher level of generalization ability for the imitator agents, is limited and rarely discussed. In this work, we present a new algorithm called Deep Demonstration Tracing (DDT). In DDT, we propose a demonstration transformer architecture to encourage agents to adaptively trace suitable states in demonstrations. Besides, it integrates OSIL into a meta-reinforcement-learning training paradigm, providing regularization for policies in unexpected situations. We evaluate DDT on a new navigation task suite and robotics tasks, demonstrating its superior performance over existing OSIL methods across all evaluated tasks in dynamic environments with unforeseen changes. The project page is in https://osil-ddt.github.io.

The task of webly-supervised fine-grained recognition is to boost recognition accuracy of classifying subordinate categories (e.g., different bird species) by utilizing freely available but noisy web data. As the label noises significantly hurt the network training, it is desirable to distinguish and eliminate noisy images. In this paper, we propose two strategies, i.e., open-set noise removal and closed-set noise correction, to both remove such two kinds of web noises w.r.t. fine-grained recognition. Specifically, for open-set noise removal, we utilize a pre-trained deep model to perform deep descriptor transformation to estimate the positive correlation between these web images, and detect the open-set noises based on the correlation values. Regarding closed-set noise correction, we develop a top-k recall optimization loss for firstly assigning a label set towards each web image to reduce the impact of hard label assignment for closed-set noises. Then, we further propose to correct the sample with its label set as the true single label from a partial label learning perspective. Experiments on several webly-supervised fine-grained benchmark datasets show that our method obviously outperforms other existing state-of-the-art methods.