On the Value of Myopic Behavior in Policy Reuse

Leveraging learned strategies in unfamiliar scenarios is fundamental to human intelligence. In reinforcement learning, rationally reusing the policies acquired from other tasks or human experts is critical for tackling problems that are difficult to learn from scratch. In this work, we present a framework called Selective Myopic bEhavior Control~(SMEC), which results from the insight that the short-term behaviors of prior policies are sharable across tasks. By evaluating the behaviors of prior policies via a hybrid value function architecture, SMEC adaptively aggregates the sharable short-term behaviors of prior policies and the long-term behaviors of the task policy, leading to coordinated decisions. Empirical results on a collection of manipulation and locomotion tasks demonstrate that SMEC outperforms existing methods, and validate the ability of SMEC to leverage related prior policies.

Learning Better with Less: Effective Augmentation for Sample-Efficient Visual Reinforcement Learning

Data augmentation (DA) is a crucial technique for enhancing the sample efficiency of visual reinforcement learning (RL) algorithms. Notably, employing simple observation transformations alone can yield outstanding performance without extra auxiliary representation tasks or pre-trained encoders. However, it remains unclear which attributes of DA account for its effectiveness in achieving sample-efficient visual RL. To investigate this issue and further explore the potential of DA, this work conducts comprehensive experiments to assess the impact of DA's attributes on its efficacy and provides the following insights and improvements: (1) For individual DA operations, we reveal that both ample spatial diversity and slight hardness are indispensable. Building on this finding, we introduce Random PadResize (Rand PR), a new DA operation that offers abundant spatial diversity with minimal hardness. (2) For multi-type DA fusion schemes, the increased DA hardness and unstable data distribution result in the current fusion schemes being unable to achieve higher sample efficiency than their corresponding individual operations. Taking the non-stationary nature of RL into account, we propose a RL-tailored multi-type DA fusion scheme called Cycling Augmentation (CycAug), which performs periodic cycles of different DA operations to increase type diversity while maintaining data distribution consistency. Extensive evaluations on the DeepMind Control suite and CARLA driving simulator demonstrate that our methods achieve superior sample efficiency compared with the prior state-of-the-art methods.

Reasoning with Language Model is Planning with World Model

Large language models (LLMs) have shown remarkable reasoning capabilities, especially when prompted to generate intermediate reasoning steps (e.g., Chain-of-Thought, CoT). However, LLMs can still struggle with problems that are easy for humans, such as generating action plans for executing tasks in a given environment, or performing complex math, logical, and commonsense reasoning. The deficiency stems from the key fact that LLMs lack an internal $\textit{world model}$ to predict the world $\textit{state}$ (e.g., environment status, intermediate variable values) and simulate long-term outcomes of actions. This prevents LLMs from performing deliberate planning akin to human brains, which involves exploring alternative reasoning paths, anticipating future states and rewards, and iteratively refining existing reasoning steps. To overcome the limitations, we propose a new LLM reasoning framework, $\underline{R}\textit{easoning vi}\underline{a} \underline{P}\textit{lanning}$ $\textbf{(RAP)}$. RAP repurposes the LLM as both a world model and a reasoning agent, and incorporates a principled planning algorithm (based on Monto Carlo Tree Search) for strategic exploration in the vast reasoning space. During reasoning, the LLM (as agent) incrementally builds a reasoning tree under the guidance of the LLM (as world model) and task-specific rewards, and obtains a high-reward reasoning path efficiently with a proper balance between exploration $\textit{vs.}$ exploitation. We apply RAP to a variety of challenging reasoning problems including plan generation, math reasoning, and logical inference. Empirical results on these tasks demonstrate the superiority of RAP over various strong baselines, including CoT and least-to-most prompting with self-consistency. RAP on LLAMA-33B surpasses CoT on GPT-4 with 33% relative improvement in a plan generation setting.

Efficient Bilateral Cross-Modality Cluster Matching for Unsupervised Visible-Infrared Person ReID

Unsupervised visible-infrared person re-identification (USL-VI-ReID) aims to match pedestrian images of the same identity from different modalities without annotations. Existing works mainly focus on alleviating the modality gap by aligning instance-level features of the unlabeled samples. However, the relationships between cross-modality clusters are not well explored. To this end, we propose a novel bilateral cluster matching-based learning framework to reduce the modality gap by matching cross-modality clusters. Specifically, we design a Many-to-many Bilateral Cross-Modality Cluster Matching (MBCCM) algorithm through optimizing the maximum matching problem in a bipartite graph. Then, the matched pairwise clusters utilize shared visible and infrared pseudo-labels during the model training. Under such a supervisory signal, a Modality-Specific and Modality-Agnostic (MSMA) contrastive learning framework is proposed to align features jointly at a cluster-level. Meanwhile, the cross-modality Consistency Constraint (CC) is proposed to explicitly reduce the large modality discrepancy. Extensive experiments on the public SYSU-MM01 and RegDB datasets demonstrate the effectiveness of the proposed method, surpassing state-of-the-art approaches by a large margin of 8.76% mAP on average.

ToolkenGPT: Augmenting Frozen Language Models with Massive Tools via Tool Embeddings

Augmenting large language models (LLMs) with external tools has emerged as a promising approach to solving complex problems. However, traditional methods, which finetune LLMs with tool demonstration data, can be both costly and restricted to a predefined set of tools. Recent in-context learning paradigm alleviates these issues, but the limited context length only allows for a few shots of demonstrations, leading to suboptimal understandings of the tools. Moreover, when there are numerous tools to choose from, in-context learning could completely fail to work. In this paper, we propose an alternative approach, $\textbf{ToolkenGPT}$, which combines the benefits of both sides. Our approach represents each $\underline{tool}$ as a to$\underline{ken}$ ($\textit{toolken}$) and learns an embedding for it, enabling tool calls in the same way as generating a regular word token. Once a toolken is triggered, the LLM is prompted to complete arguments for the tool to execute. ToolkenGPT offers the flexibility to plug in an arbitrary number of tools by expanding the set of toolkens on the fly. In addition, it improves tool use by allowing extensive demonstration data for learning the toolken embeddings. In diverse domains, including numerical reasoning, knowledge-based question answering, and embodied plan generation, our approach effectively augments LLMs with tools and substantially outperforms various latest baselines. ToolkenGPT demonstrates the promising ability to use relevant tools from a large tool set in complex scenarios.

ReDirTrans: Latent-to-Latent Translation for Gaze and Head Redirection

Learning-based gaze estimation methods require large amounts of training data with accurate gaze annotations. Facing such demanding requirements of gaze data collection and annotation, several image synthesis methods were proposed, which successfully redirected gaze directions precisely given the assigned conditions. However, these methods focused on changing gaze directions of the images that only include eyes or restricted ranges of faces with low resolution (less than $128\times128$) to largely reduce interference from other attributes such as hairs, which limits application scenarios. To cope with this limitation, we proposed a portable network, called ReDirTrans, achieving latent-to-latent translation for redirecting gaze directions and head orientations in an interpretable manner. ReDirTrans projects input latent vectors into aimed-attribute embeddings only and redirects these embeddings with assigned pitch and yaw values. Then both the initial and edited embeddings are projected back (deprojected) to the initial latent space as residuals to modify the input latent vectors by subtraction and addition, representing old status removal and new status addition. The projection of aimed attributes only and subtraction-addition operations for status replacement essentially mitigate impacts on other attributes and the distribution of latent vectors. Thus, by combining ReDirTrans with a pretrained fixed e4e-StyleGAN pair, we created ReDirTrans-GAN, which enables accurately redirecting gaze in full-face images with $1024\times1024$ resolution while preserving other attributes such as identity, expression, and hairstyle. Furthermore, we presented improvements for the downstream learning-based gaze estimation task, using redirected samples as dataset augmentation.

Behavior Contrastive Learning for Unsupervised Skill Discovery

In reinforcement learning, unsupervised skill discovery aims to learn diverse skills without extrinsic rewards. Previous methods discover skills by maximizing the mutual information (MI) between states and skills. However, such an MI objective tends to learn simple and static skills and may hinder exploration. In this paper, we propose a novel unsupervised skill discovery method through contrastive learning among behaviors, which makes the agent produce similar behaviors for the same skill and diverse behaviors for different skills. Under mild assumptions, our objective maximizes the MI between different behaviors based on the same skill, which serves as an upper bound of the previous MI objective. Meanwhile, our method implicitly increases the state entropy to obtain better state coverage. We evaluate our method on challenging mazes and continuous control tasks. The results show that our method generates diverse and far-reaching skills, and also obtains competitive performance in downstream tasks compared to the state-of-the-art methods.

Single Node Injection Label Specificity Attack on Graph Neural Networks via Reinforcement Learning

Graph neural networks (GNNs) have achieved remarkable success in various real-world applications. However, recent studies highlight the vulnerability of GNNs to malicious perturbations. Previous adversaries primarily focus on graph modifications or node injections to existing graphs, yielding promising results but with notable limitations. Graph modification attack~(GMA) requires manipulation of the original graph, which is often impractical, while graph injection attack~(GIA) necessitates training a surrogate model in the black-box setting, leading to significant performance degradation due to divergence between the surrogate architecture and the actual victim model. Furthermore, most methods concentrate on a single attack goal and lack a generalizable adversary to develop distinct attack strategies for diverse goals, thus limiting precise control over victim model behavior in real-world scenarios. To address these issues, we present a gradient-free generalizable adversary that injects a single malicious node to manipulate the classification result of a target node in the black-box evasion setting. We propose Gradient-free Generalizable Single Node Injection Attack, namely G$^2$-SNIA, a reinforcement learning framework employing Proximal Policy Optimization. By directly querying the victim model, G$^2$-SNIA learns patterns from exploration to achieve diverse attack goals with extremely limited attack budgets. Through comprehensive experiments over three acknowledged benchmark datasets and four prominent GNNs in the most challenging and realistic scenario, we demonstrate the superior performance of our proposed G$^2$-SNIA over the existing state-of-the-art baselines. Moreover, by comparing G$^2$-SNIA with multiple white-box evasion baselines, we confirm its capacity to generate solutions comparable to those of the best adversaries.

Delving into Shape-aware Zero-shot Semantic Segmentation

Thanks to the impressive progress of large-scale vision-language pretraining, recent recognition models can classify arbitrary objects in a zero-shot and open-set manner, with a surprisingly high accuracy. However, translating this success to semantic segmentation is not trivial, because this dense prediction task requires not only accurate semantic understanding but also fine shape delineation and existing vision-language models are trained with image-level language descriptions. To bridge this gap, we pursue \textbf{shape-aware} zero-shot semantic segmentation in this study. Inspired by classical spectral methods in the image segmentation literature, we propose to leverage the eigen vectors of Laplacian matrices constructed with self-supervised pixel-wise features to promote shape-awareness. Despite that this simple and effective technique does not make use of the masks of seen classes at all, we demonstrate that it out-performs a state-of-the-art shape-aware formulation that aligns ground truth and predicted edges during training. We also delve into the performance gains achieved on different datasets using different backbones and draw several interesting and conclusive observations: the benefits of promoting shape-awareness highly relates to mask compactness and language embedding locality. Finally, our method sets new state-of-the-art performance for zero-shot semantic segmentation on both Pascal and COCO, with significant margins. Code and models will be accessed at https://github.com/Liuxinyv/SAZS.

HPN: Personalized Federated Hyperparameter Optimization

Numerous research studies in the field of federated learning (FL) have attempted to use personalization to address the heterogeneity among clients, one of FL's most crucial and challenging problems. However, existing works predominantly focus on tailoring models. Yet, due to the heterogeneity of clients, they may each require different choices of hyperparameters, which have not been studied so far. We pinpoint two challenges of personalized federated hyperparameter optimization (pFedHPO): handling the exponentially increased search space and characterizing each client without compromising its data privacy. To overcome them, we propose learning a \textsc{H}yper\textsc{P}arameter \textsc{N}etwork (HPN) fed with client encoding to decide personalized hyperparameters. The client encoding is calculated with a random projection-based procedure to protect each client's privacy. Besides, we design a novel mechanism to debias the low-fidelity function evaluation samples for learning HPN. We conduct extensive experiments on FL tasks from various domains, demonstrating the superiority of HPN.