Continual Task Learning through Adaptive Policy Self-Composition

Training a generalizable agent to continually learn a sequence of tasks from offline trajectories is a natural requirement for long-lived agents, yet remains a significant challenge for current offline reinforcement learning (RL) algorithms. Specifically, an agent must be able to rapidly adapt to new tasks using newly collected trajectories (plasticity), while retaining knowledge from previously learned tasks (stability). However, systematic analyses of this setting are scarce, and it remains unclear whether conventional continual learning (CL) methods are effective in continual offline RL (CORL) scenarios. In this study, we develop the Offline Continual World benchmark and demonstrate that traditional CL methods struggle with catastrophic forgetting, primarily due to the unique distribution shifts inherent to CORL scenarios. To address this challenge, we introduce CompoFormer, a structure-based continual transformer model that adaptively composes previous policies via a meta-policy network. Upon encountering a new task, CompoFormer leverages semantic correlations to selectively integrate relevant prior policies alongside newly trained parameters, thereby enhancing knowledge sharing and accelerating the learning process. Our experiments reveal that CompoFormer outperforms conventional CL methods, particularly in longer task sequences, showcasing a promising balance between plasticity and stability.

Task-Aware Harmony Multi-Task Decision Transformer for Offline Reinforcement Learning

The purpose of offline multi-task reinforcement learning (MTRL) is to develop a unified policy applicable to diverse tasks without the need for online environmental interaction. Recent advancements approach this through sequence modeling, leveraging the Transformer architecture's scalability and the benefits of parameter sharing to exploit task similarities. However, variations in task content and complexity pose significant challenges in policy formulation, necessitating judicious parameter sharing and management of conflicting gradients for optimal policy performance. Furthermore, identifying the optimal parameter subspace for each task often necessitates prior knowledge of the task identifier during inference, limiting applicability in real-world scenarios with variable task content and unknown current tasks. In this work, we introduce the Harmony Multi-Task Decision Transformer (HarmoDT), a novel solution designed to identify an optimal harmony subspace of parameters for each task. We formulate this as a bi-level optimization problem within a meta-learning framework, where the upper level learns masks to define the harmony subspace, while the inner level focuses on updating parameters to improve the overall performance of the unified policy. To eliminate the need for task identifiers, we further design a group-wise variant (G-HarmoDT) that clusters tasks into coherent groups based on gradient information, and utilizes a gating network to determine task identifiers during inference. Empirical evaluations across various benchmarks highlight the superiority of our approach, demonstrating its effectiveness in the multi-task context with specific improvements of 8% gain in task-provided settings, 5% in task-agnostic settings, and 10% in unseen settings.

Reconstruct the Pruned Model without Any Retraining

Structured pruning is a promising hardware-friendly compression technique for large language models (LLMs), which is expected to be retraining-free to avoid the enormous retraining cost. This retraining-free paradigm involves (1) pruning criteria to define the architecture and (2) distortion reconstruction to restore performance. However, existing methods often emphasize pruning criteria while using reconstruction techniques that are specific to certain modules or criteria, resulting in limited generalizability. To address this, we introduce the Linear Interpolation-based Adaptive Reconstruction (LIAR) framework, which is both efficient and effective. LIAR does not require back-propagation or retraining and is compatible with various pruning criteria and modules. By applying linear interpolation to the preserved weights, LIAR minimizes reconstruction error and effectively reconstructs the pruned output. Our evaluations on benchmarks such as GLUE, SQuAD, WikiText, and common sense reasoning show that LIAR enables a BERT model to maintain 98% accuracy even after removing 50% of its parameters and achieves top performance for LLaMA in just a few minutes.

Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts

This paper presents a Domain-Inspired Sharpness-Aware Minimization (DISAM) algorithm for optimization under domain shifts. It is motivated by the inconsistent convergence degree of SAM across different domains, which induces optimization bias towards certain domains and thus impairs the overall convergence. To address this issue, we consider the domain-level convergence consistency in the sharpness estimation to prevent the overwhelming (deficient) perturbations for less (well) optimized domains. Specifically, DISAM introduces the constraint of minimizing variance in the domain loss, which allows the elastic gradient calibration in perturbation generation: when one domain is optimized above the averaging level \textit{w.r.t.} loss, the gradient perturbation towards that domain will be weakened automatically, and vice versa. Under this mechanism, we theoretically show that DISAM can achieve faster overall convergence and improved generalization in principle when inconsistent convergence emerges. Extensive experiments on various domain generalization benchmarks show the superiority of DISAM over a range of state-of-the-art methods. Furthermore, we show the superior efficiency of DISAM in parameter-efficient fine-tuning combined with the pretraining models. The source code is released at https://github.com/MediaBrain-SJTU/DISAM.

Federated Learning with Bilateral Curation for Partially Class-Disjoint Data

Partially class-disjoint data (PCDD), a common yet under-explored data formation where each client contributes a part of classes (instead of all classes) of samples, severely challenges the performance of federated algorithms. Without full classes, the local objective will contradict the global objective, yielding the angle collapse problem for locally missing classes and the space waste problem for locally existing classes. As far as we know, none of the existing methods can intrinsically mitigate PCDD challenges to achieve holistic improvement in the bilateral views (both global view and local view) of federated learning. To address this dilemma, we are inspired by the strong generalization of simplex Equiangular Tight Frame~(ETF) on the imbalanced data, and propose a novel approach called FedGELA where the classifier is globally fixed as a simplex ETF while locally adapted to the personal distributions. Globally, FedGELA provides fair and equal discrimination for all classes and avoids inaccurate updates of the classifier, while locally it utilizes the space of locally missing classes for locally existing classes. We conduct extensive experiments on a range of datasets to demonstrate that our FedGELA achieves promising performance~(averaged improvement of 3.9% to FedAvg and 1.5% to best baselines) and provide both local and global convergence guarantees. Source code is available at:https://github.com/MediaBrain-SJTU/FedGELA.git.

Locally Estimated Global Perturbations are Better than Local Perturbations for Federated Sharpness-aware Minimization

In federated learning (FL), the multi-step update and data heterogeneity among clients often lead to a loss landscape with sharper minima, degenerating the performance of the resulted global model. Prevalent federated approaches incorporate sharpness-aware minimization (SAM) into local training to mitigate this problem. However, the local loss landscapes may not accurately reflect the flatness of global loss landscape in heterogeneous environments; as a result, minimizing local sharpness and calculating perturbations on client data might not align the efficacy of SAM in FL with centralized training. To overcome this challenge, we propose FedLESAM, a novel algorithm that locally estimates the direction of global perturbation on client side as the difference between global models received in the previous active and current rounds. Besides the improved quality, FedLESAM also speed up federated SAM-based approaches since it only performs once backpropagation in each iteration. Theoretically, we prove a slightly tighter bound than its original FedSAM by ensuring consistent perturbation. Empirically, we conduct comprehensive experiments on four federated benchmark datasets under three partition strategies to demonstrate the superior performance and efficiency of FedLESAM.

Federated Learning under Partially Class-Disjoint Data via Manifold Reshaping

Statistical heterogeneity severely limits the performance of federated learning (FL), motivating several explorations e.g., FedProx, MOON and FedDyn, to alleviate this problem. Despite effectiveness, their considered scenario generally requires samples from almost all classes during the local training of each client, although some covariate shifts may exist among clients. In fact, the natural case of partially class-disjoint data (PCDD), where each client contributes a few classes (instead of all classes) of samples, is practical yet underexplored. Specifically, the unique collapse and invasion characteristics of PCDD can induce the biased optimization direction in local training, which prevents the efficiency of federated learning. To address this dilemma, we propose a manifold reshaping approach called FedMR to calibrate the feature space of local training. Our FedMR adds two interplaying losses to the vanilla federated learning: one is intra-class loss to decorrelate feature dimensions for anti-collapse; and the other one is inter-class loss to guarantee the proper margin among categories in the feature expansion. We conduct extensive experiments on a range of datasets to demonstrate that our FedMR achieves much higher accuracy and better communication efficiency. Source code is available at: https://github.com/MediaBrain-SJTU/FedMR.git.

HarmoDT: Harmony Multi-Task Decision Transformer for Offline Reinforcement Learning

The purpose of offline multi-task reinforcement learning (MTRL) is to develop a unified policy applicable to diverse tasks without the need for online environmental interaction. Recent advancements approach this through sequence modeling, leveraging the Transformer architecture's scalability and the benefits of parameter sharing to exploit task similarities. However, variations in task content and complexity pose significant challenges in policy formulation, necessitating judicious parameter sharing and management of conflicting gradients for optimal policy performance. In this work, we introduce the Harmony Multi-Task Decision Transformer (HarmoDT), a novel solution designed to identify an optimal harmony subspace of parameters for each task. We approach this as a bi-level optimization problem, employing a meta-learning framework that leverages gradient-based techniques. The upper level of this framework is dedicated to learning a task-specific mask that delineates the harmony subspace, while the inner level focuses on updating parameters to enhance the overall performance of the unified policy. Empirical evaluations on a series of benchmarks demonstrate the superiority of HarmoDT, verifying the effectiveness of our approach.

Q-value Regularized Transformer for Offline Reinforcement Learning

Recent advancements in offline reinforcement learning (RL) have underscored the capabilities of Conditional Sequence Modeling (CSM), a paradigm that learns the action distribution based on history trajectory and target returns for each state. However, these methods often struggle with stitching together optimal trajectories from sub-optimal ones due to the inconsistency between the sampled returns within individual trajectories and the optimal returns across multiple trajectories. Fortunately, Dynamic Programming (DP) methods offer a solution by leveraging a value function to approximate optimal future returns for each state, while these techniques are prone to unstable learning behaviors, particularly in long-horizon and sparse-reward scenarios. Building upon these insights, we propose the Q-value regularized Transformer (QT), which combines the trajectory modeling ability of the Transformer with the predictability of optimal future returns from DP methods. QT learns an action-value function and integrates a term maximizing action-values into the training loss of CSM, which aims to seek optimal actions that align closely with the behavior policy. Empirical evaluations on D4RL benchmark datasets demonstrate the superiority of QT over traditional DP and CSM methods, highlighting the potential of QT to enhance the state-of-the-art in offline RL.

FedSkip: Combatting Statistical Heterogeneity with Federated Skip Aggregation

The statistical heterogeneity of the non-independent and identically distributed (non-IID) data in local clients significantly limits the performance of federated learning. Previous attempts like FedProx, SCAFFOLD, MOON, FedNova and FedDyn resort to an optimization perspective, which requires an auxiliary term or re-weights local updates to calibrate the learning bias or the objective inconsistency. However, in addition to previous explorations for improvement in federated averaging, our analysis shows that another critical bottleneck is the poorer optima of client models in more heterogeneous conditions. We thus introduce a data-driven approach called FedSkip to improve the client optima by periodically skipping federated averaging and scattering local models to the cross devices. We provide theoretical analysis of the possible benefit from FedSkip and conduct extensive experiments on a range of datasets to demonstrate that FedSkip achieves much higher accuracy, better aggregation efficiency and competing communication efficiency. Source code is available at: https://github.com/MediaBrain-SJTU/FedSkip.