Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stages

Plasticity, the ability of a neural network to evolve with new data, is crucial for high-performance and sample-efficient visual reinforcement learning (VRL). Although methods like resetting and regularization can potentially mitigate plasticity loss, the influences of various components within the VRL framework on the agent's plasticity are still poorly understood. In this work, we conduct a systematic empirical exploration focusing on three primary underexplored facets and derive the following insightful conclusions: (1) data augmentation is essential in maintaining plasticity; (2) the critic's plasticity loss serves as the principal bottleneck impeding efficient training; and (3) without timely intervention to recover critic's plasticity in the early stages, its loss becomes catastrophic. These insights suggest a novel strategy to address the high replay ratio (RR) dilemma, where exacerbated plasticity loss hinders the potential improvements of sample efficiency brought by increased reuse frequency. Rather than setting a static RR for the entire training process, we propose Adaptive RR, which dynamically adjusts the RR based on the critic's plasticity level. Extensive evaluations indicate that Adaptive RR not only avoids catastrophic plasticity loss in the early stages but also benefits from more frequent reuse in later phases, resulting in superior sample efficiency.

Towards Robust Offline-to-Online Reinforcement Learning via Uncertainty and Smoothness

To obtain a near-optimal policy with fewer interactions in Reinforcement Learning (RL), a promising approach involves the combination of offline RL, which enhances sample efficiency by leveraging offline datasets, and online RL, which explores informative transitions by interacting with the environment. Offline-to-Online (O2O) RL provides a paradigm for improving an offline trained agent within limited online interactions. However, due to the significant distribution shift between online experiences and offline data, most offline RL algorithms suffer from performance drops and fail to achieve stable policy improvement in O2O adaptation. To address this problem, we propose the Robust Offline-to-Online (RO2O) algorithm, designed to enhance offline policies through uncertainty and smoothness, and to mitigate the performance drop in online adaptation. Specifically, RO2O incorporates Q-ensemble for uncertainty penalty and adversarial samples for policy and value smoothness, which enable RO2O to maintain a consistent learning procedure in online adaptation without requiring special changes to the learning objective. Theoretical analyses in linear MDPs demonstrate that the uncertainty and smoothness lead to a tighter optimality bound in O2O against distribution shift. Experimental results illustrate the superiority of RO2O in facilitating stable offline-to-online learning and achieving significant improvement with limited online interactions.

Node-Aligned Graph-to-Graph Generation for Retrosynthesis Prediction

Single-step retrosynthesis is a crucial task in organic chemistry and drug design, requiring the identification of required reactants to synthesize a specific compound. with the advent of computer-aided synthesis planning, there is growing interest in using machine-learning techniques to facilitate the process. Existing template-free machine learning-based models typically utilize transformer structures and represent molecules as ID sequences. However, these methods often face challenges in fully leveraging the extensive topological information of the molecule and aligning atoms between the production and reactants, leading to results that are not as competitive as those of semi-template models. Our proposed method, Node-Aligned Graph-to-Graph (NAG2G), also serves as a transformer-based template-free model but utilizes 2D molecular graphs and 3D conformation information. Furthermore, our approach simplifies the incorporation of production-reactant atom mapping alignment by leveraging node alignment to determine a specific order for node generation and generating molecular graphs in an auto-regressive manner node-by-node. This method ensures that the node generation order coincides with the node order in the input graph, overcoming the difficulty of determining a specific node generation order in an auto-regressive manner. Our extensive benchmarking results demonstrate that the proposed NAG2G can outperform the previous state-of-the-art baselines in various metrics.

Learning Point-wise Abstaining Penalty for Point Cloud Anomaly Detection

LiDAR-based semantic scene understanding is an important module in the modern autonomous driving perception stack. However, identifying Out-Of-Distribution (OOD) points in a LiDAR point cloud is challenging as point clouds lack semantically rich features when compared with RGB images. We revisit this problem from the perspective of selective classification, which introduces a selective function into the standard closed-set classification setup. Our solution is built upon the basic idea of abstaining from choosing any known categories but learns a point-wise abstaining penalty with a marginbased loss. Synthesizing outliers to approximate unlimited OOD samples is also critical to this idea, so we propose a strong synthesis pipeline that generates outliers originated from various factors: unrealistic object categories, sampling patterns and sizes. We demonstrate that learning different abstaining penalties, apart from point-wise penalty, for different types of (synthesized) outliers can further improve the performance. We benchmark our method on SemanticKITTI and nuScenes and achieve state-of-the-art results. Risk-coverage analysis further reveals intrinsic properties of different methods. Codes and models will be publicly available.

Least Squares Maximum and Weighted Generalization-Memorization Machines

In this paper, we propose a new way of remembering by introducing a memory influence mechanism for the least squares support vector machine (LSSVM). Without changing the equation constraints of the original LSSVM, this mechanism, allows an accurate partitioning of the training set without overfitting. The maximum memory impact model (MIMM) and the weighted impact memory model (WIMM) are then proposed. It is demonstrated that these models can be degraded to the LSSVM. Furthermore, we propose some different memory impact functions for the MIMM and WIMM. The experimental results show that that our MIMM and WIMM have better generalization performance compared to the LSSVM and significant advantage in time cost compared to other memory models.

IOB: Integrating Optimization Transfer and Behavior Transfer for Multi-Policy Reuse

Humans have the ability to reuse previously learned policies to solve new tasks quickly, and reinforcement learning (RL) agents can do the same by transferring knowledge from source policies to a related target task. Transfer RL methods can reshape the policy optimization objective (optimization transfer) or influence the behavior policy (behavior transfer) using source policies. However, selecting the appropriate source policy with limited samples to guide target policy learning has been a challenge. Previous methods introduce additional components, such as hierarchical policies or estimations of source policies' value functions, which can lead to non-stationary policy optimization or heavy sampling costs, diminishing transfer effectiveness. To address this challenge, we propose a novel transfer RL method that selects the source policy without training extra components. Our method utilizes the Q function in the actor-critic framework to guide policy selection, choosing the source policy with the largest one-step improvement over the current target policy. We integrate optimization transfer and behavior transfer (IOB) by regularizing the learned policy to mimic the guidance policy and combining them as the behavior policy. This integration significantly enhances transfer effectiveness, surpasses state-of-the-art transfer RL baselines in benchmark tasks, and improves final performance and knowledge transferability in continual learning scenarios. Additionally, we show that our optimization transfer technique is guaranteed to improve target policy learning.

Roll Up Your Sleeves: Working with a Collaborative and Engaging Task-Oriented Dialogue System

We introduce TacoBot, a user-centered task-oriented digital assistant designed to guide users through complex real-world tasks with multiple steps. Covering a wide range of cooking and how-to tasks, we aim to deliver a collaborative and engaging dialogue experience. Equipped with language understanding, dialogue management, and response generation components supported by a robust search engine, TacoBot ensures efficient task assistance. To enhance the dialogue experience, we explore a series of data augmentation strategies using LLMs to train advanced neural models continuously. TacoBot builds upon our successful participation in the inaugural Alexa Prize TaskBot Challenge, where our team secured third place among ten competing teams. We offer TacoBot as an open-source framework that serves as a practical example for deploying task-oriented dialogue systems.

Sequential Attention Source Identification Based on Feature Representation

Snapshot observation based source localization has been widely studied due to its accessibility and low cost. However, the interaction of users in existing methods does not be addressed in time-varying infection scenarios. So these methods have a decreased accuracy in heterogeneous interaction scenarios. To solve this critical issue, this paper proposes a sequence-to-sequence based localization framework called Temporal-sequence based Graph Attention Source Identification (TGASI) based on an inductive learning idea. More specifically, the encoder focuses on generating multiple features by estimating the influence probability between two users, and the decoder distinguishes the importance of prediction sources in different timestamps by a designed temporal attention mechanism. It's worth mentioning that the inductive learning idea ensures that TGASI can detect the sources in new scenarios without knowing other prior knowledge, which proves the scalability of TGASI. Comprehensive experiments with the SOTA methods demonstrate the higher detection performance and scalability in different scenarios of TGASI.

MAT: Mixed-Strategy Game of Adversarial Training in Fine-tuning

Fine-tuning large-scale pre-trained language models has been demonstrated effective for various natural language processing (NLP) tasks. Previous studies have established that incorporating adversarial training during the fine-tuning stage can significantly enhance model generalization and robustness. However, from the perspective of game theory, such utilizations of adversarial training correspond to pure-strategy games, which are inherently limited in terms of the scope of their strategies, thereby still having room for improvement. In order to push the performance boundaries, we propose a novel Mixed-strategy Adversarial Training algorithm (MAT). Methodologically, we derive the Nash equilibrium of a mixed-strategy game for adversarial training using Entropy Mirror Descent to establish MAT by sampling method. To verify the effectiveness of MAT, we conducted extensive benchmark experiments on large-scale pre-trained models, such as BERT and RoBERTa. MAT significantly outperforms the state-of-the-art methods on both the GLUE and ANLI benchmarks in terms of generalization and robustness.

Cross-Domain Policy Adaptation via Value-Guided Data Filtering

Generalizing policies across different domains with dynamics mismatch poses a significant challenge in reinforcement learning. For example, a robot learns the policy in a simulator, but when it is deployed in the real world, the dynamics of the environment may be different. Given the source and target domain with dynamics mismatch, we consider the online dynamics adaptation problem, in which case the agent can access sufficient source domain data while online interactions with the target domain are limited. Existing research has attempted to solve the problem from the dynamics discrepancy perspective. In this work, we reveal the limitations of these methods and explore the problem from the value difference perspective via a novel insight on the value consistency across domains. Specifically, we present the Value-Guided Data Filtering (VGDF) algorithm, which selectively shares transitions from the source domain based on the proximity of paired value targets across the two domains. Empirical results on various environments with kinematic and morphology shifts demonstrate that our method achieves superior performance compared to prior approaches.