IJCAI 2024 Papers

In offline Reinforcement Learning (RL), the pre-trained policies are utilized for initialization and subsequent online fine-tuning. However, existing methods suffer from instability and low sample efficiency compared to pure online learning. This paper identifies these limitations stemming from direct policy initialization using offline-trained policy models. We propose Continual Policy Revitalization (CPR) as a novel efficient, stable fine-tuning method. CPR incorporates a periodic policy revitalization technique, restoring the overtrained policy network to full learning capacity while ensuring stable initial performance. This approach enables fine-tuning without being adversely affected by low-quality pre-trained policies. In contrast to previous research, CPR initializes the new policy with an adaptive policy constraint in policy optimization. Such optimization keeps the new policy close to behavior policy constructed from historical policies. This contributes to stable policy improvement and optimal converged performance. Practically, CPR can seamlessly integrate into existing offline RL algorithms with minimal modification. We empirically validate the effectiveness of our method through extensive experiments, demonstrating substantial improvements in learning stability and efficiency compared to previous approaches. Our code is available at https://github.com/LAMDA-RL/CPR.

Rich textual and topological information of textual graphs need to be modeled in real-world applications such as webpages, e-commerce, and academic articles. Practitioners have been long following the path of adopting a shallow text encoder and a subsequent graph neural network (GNN) to solve this problem. In light of recent advancements in large language models (LLMs), it is apparent that integrating LLMs for enhanced textual encoding can substantially improve the performance of textual graphs. Nevertheless, the efficiency of these methods poses a significant challenge. In this paper, we propose ENGINE, a parameter- and memory-efficient fine-tuning method for textual graphs with an LLM encoder. The key insight is to combine the LLMs and GNNs through a tunable side structure, which significantly reduces the training complexity without impairing the joint model's capacity. Extensive experiments on textual graphs demonstrate our method's effectiveness by achieving the best model performance, meanwhile having the lowest training cost compared to previous methods. Moreover, we introduce two variants with caching and dynamic early exit to further enhance training and inference speed. Specifically, caching accelerates ENGINE's training by 12x, and dynamic early exit achieves up to 5x faster inference with a negligible performance drop (at maximum 1.17% relevant drop across 7 datasets). Our codes are available at: https://github.com/ZhuYun97/ENGINE.

This paper addresses the challenge of efficiently compressing screen content images (SCIs) – computer generated images with unique attributes such as large uniform regions, sharp edges, and limited color palettes, which pose difficulties for conventional compression algorithms. We propose a Superpixel-based Content Aggregation Block (SCAB) to aggregate local pixels into one super-pixel and aggregate non-local information via super-pixel transformer. Such aggregation enables the dynamic assimilation of non-local information while maintaining manageable complexity. Furthermore, we enhance our channel-wise context entropy model with a Dynamic Feature Fusion (DFF) mechanism. This mechanism integrates decoded slices and side information dynamically based on their global correlation, allowing the network to dynamically learn the optimal weights for global information usage. Extensive experiments on three SCI datasets (SCID, CCT, and SIQAD) show our method’s superior RD performance and inference time, making it the first network comparable with the advanced VVC-SCC standard.

Zero-shot adaptation is crucial for agents facing new tasks. Offline Meta-Reinforcement Learning (OMRL), utilizing offline multi-task datasets to train policies, offers a way to attain this ability. Although most OMRL methods construct task representations via contrastive learning and merge them with states for policy input, these methods may have inherent problems. Specifically, integrating task representations with states for policy input limits learning efficiency, due to failing to leverage the similarities among tasks. Moreover, uniformly sampling an equal number of negative samples from different tasks in contrastive learning can hinder differentiation of more similar tasks, potentially diminishing task representation robustness. In this paper, we introduce an OMRL algorithm to tackle the aforementioned issues. We design a network structure for efficient learning by leveraging task similarity. It features shared lower layers for common feature extraction with a hypernetworks-driven upper layer, customized to process features per task's attributes. Furthermore, to achieve robust task representations for generating task-specific control policies, we utilize contrastive learning and introduce a novel method to construct negative sample pairs based on task similarity. Experimental results show that our method notably boosts learning efficiency and zero-shot adaptation in new tasks, surpassing previous methods across multiple challenging domains.

As data with diverse representations become high-dimensional, multi-view unsupervised feature selection has been an important learning paradigm. Generally, existing methods encounter the following challenges: (i) traditional solutions either concatenate different views or introduce extra parameters to weight them, affecting the performance and applicability; (ii) emphasis is typically placed on graph construction, yet disregarding the clustering information of data; (iii) exploring the similarity structure of all samples from the original features is suboptimal and extremely time-consuming. To solve this dilemma, we propose an efficient multi-view unsupervised feature selection (EMUFS) to construct bipartite graphs between samples and anchors. Specifically, a parameter-free manner is devised to collaboratively fuse the membership matrices and graphs to learn the compatible structure information across all views, naturally balancing different views. Moreover, EMUFS leverages the similarity relations of data in the feature subspace induced by l2,0-norm to dynamically update the graph. Accordingly, the cluster information of anchors can be accurately propagated to samples via the graph structure and further guide feature selection, enhancing the quality of selected features and the computational costs in solution processes. A convergent optimization is developed to solve the formulated problem, and experiments demonstrate the effectiveness and efficiency of EMUFS.

Multi-view clustering is a popular unsupervised multi-view learning method. Real-world multi-view data are often distributed across multiple entities, presenting a challenge for performing multi-view clustering. Federated learning provides a solution by enabling multiple entities to collaboratively train a global model. However, existing federated multi-view clustering methods usually conduct feature extraction and clustering in separate steps, potentially leading to a degradation in clustering performance. To address this issue and for the sake of efficiency, we propose a novel Federated Multi-View Clustering method with Integrated Matrix Factorization and K-Means (FMVC-IMK), which integrates matrix factorization and multi-view K-means into one step. Additionally, an adaptive weight is employed to balance the influence of data from each view. FMVC-IMK further incorporates a graph-based regularizer to preserve the original data's geometric structure within the learned global clustering structure. We also develop a federated optimization approach to collaboratively learn a global clustering result without sharing any original data. Experimental results on multiple datasets demonstrate the effectiveness of FMVC-IMK.

Current Event Stream Super-Resolution (ESR) methods overlook the redundant and complementary information present in positive and negative events within the event stream, employing a direct mixing approach for super-resolution, which may lead to detail loss and inefficiency. To address these issues, we propose an efficient Recursive Multi-Branch Information Fusion Network (RMFNet) that separates positive and negative events for complementary information extraction, followed by mutual supplementation and refinement. Particularly, we introduce Feature Fusion Modules (FFM) and Feature Exchange Modules (FEM). FFM is designed for the fusion of contextual information within neighboring event streams, leveraging the coupling relationship between positive and negative events to alleviate the misleading of noises in the respective branches. FEM efficiently promotes the fusion and exchange of information between positive and negative branches, enabling superior local information enhancement and global information complementation. Experimental results demonstrate that our approach achieves over 17% and 31% improvement on synthetic and real datasets, accompanied by a 2.3x acceleration. Furthermore, we evaluate our method on two downstream event-driven applications, i.e., object recognition and video reconstruction, achieving remarkable results that outperform existing methods. Our code and Supplementary Material are available at https://github.com/Lqm26/RMFNet.

We introduce and study various models for satisfying electrical energy demands of prosumers in a microgrid, while optimizing their costs. Each prosumer has individual demands of electrical energy, which can vary day-by-day, and which they can satisfy by either generating electrical energy through a self-operated mini power plant like a solar panel, through buying from an external energy provider, such as the main grid or by trading with other prosumers. Our models take into account two key aspects motivated by real-life scenarios: first, we consider a daily volatility of prices for buying and selling the energy, and second, the possibility to store the self-generated energy in a battery of finite capacity to be either self-consumed or sold to other prosumers in the future. We provide a thorough complexity analysis, as well as efficient algorithms, so that prosumers can minimize their overall cost over the entire time horizon. As a byproduct, we also solve a new, generalized version of the KNAPSACK problem which may be of independent interest. We complement our theoretical findings by extensive experimental evaluations on realistic data sets.

Correlated subgraph searches (CSSs) are essential building blocks for AI-powered drug discovery. Given a query molecule modeled as a graph, CSS finds top-k molecules correlated to the query in a database. However, the cost increases exponentially with the molecule size. Herein we present Corgi, a framework to accelerate CSS methods while ensuring top-k search accuracy. Corgi dynamically excludes unnecessary subgraphs to overcome the expensive cost without sacrificing search accuracy. Our experimental analysis confirms that Corgi has a shorter running time and improved accuracy compared to existing state-of-the-art methods, while a case study demonstrates that Corgi is suitable for practical AI-powered drug discovery.

The generalization of neural networks has been a major focus of research in deep learning. It is often interpreted as an implicit bias towards solutions with specific properties. Especially, in practical applications, it has been observed that linear neural networks (LNN) tend to favor low-rank solutions for matrix completion tasks. However, most existing methods rely on increasing the depth of the neural network to enhance the low rank of solutions, resulting in higher complexity. In this paper, we propose a new explicit regularization method that calibrates the implicit bias towards low-rank trends in matrix completion tasks. Our approach automatically incorporates smaller singular values into the training process using a self-paced learning strategy, gradually restoring matrix information. By jointly using both implicit and explicit regularization, we effectively capture the low-rank structure of LNN and accelerate its convergence. We also analyze how our proposed penalty term interacts with implicit regularization and provide theoretical guarantees for our new model. To evaluate the effectiveness of our method, we conduct a series of experiments on both simulated and real-world data. Our experimental results clearly demonstrate that our method has better robustness and generalization ability compared with other methods.

Credit card fraud imposes significant costs on both cardholders and issuing banks. Fraudsters often disguise their crimes, such as using legitimate transactions through several benign users to bypass anti-fraud detection. Existing graph neural network (GNN) models struggle with learning features of camouflaged, indirect multi-hop transactions due to their inherent over-smoothing issues in deep multi-layer aggregation, presenting a major challenge in detecting disguised relationships. Therefore, in this paper, we propose a novel High-order Graph Representation Learning model (HOGRL) to avoid incorporating excessive noise during the multi-layer aggregation process. In particular, HOGRL learns different orders of \emph{pure} representations directly from high-order transaction graphs. We realize this goal by effectively constructing high-order transaction graphs first and then learning the \emph{pure} representations of each order so that the model could identify fraudsters' multi-hop indirect transactions via multi-layer \emph{pure} feature learning. In addition, we introduce a mixture-of-expert attention mechanism to automatically determine the importance of different orders for jointly optimizing fraud detection performance. We conduct extensive experiments in both the open source and real-world datasets, the result demonstrates the significant improvements of our proposed HOGRL compared with state-of-the-art fraud detection baselines. HOGRL's superior performance also proves its effectiveness in addressing high-order fraud camouflage criminals.

We consider an agent acting in a complex environment modeled through a multi-tiered specification, in which each tier adds nondeterminism in the environment response to the agent actions. In this setting, we devise an effective approach to best-effort synthesis, i.e., synthesizing agent strategies that win against a maximal set of possible environment responses in each tier. We do this in a setting where both the multi-tier environment and agent goal are specified in the linear temporal logic on finite traces (LTLf). While theoretical solution techniques based on automata on infinite trees have been developed previously, we completely side-step them here and focus on a DFA-based game-theoretic technique, which can be effectively implemented symbolically. Specifically, we present a provably correct algorithm that is based on solving separately DFA-based games for each tier and then combining the obtained solutions on-the-fly. This algorithm is linear, as opposed to being exponential, in the number of tiers and thus, it can graciously handle multi-tier environments formed of several tiers.

Motivated by the superior performance of image diffusion models, more and more researchers strive to extend these models to the text-based video editing task. Nevertheless, current video editing tasks mainly suffer from the dilemma between the high fine-tuning cost and the limited generation capacity. Compared with images, we conjecture that videos necessitate more constraints to preserve the temporal consistency during editing. Towards this end, we propose EVE, a robust and Efficient zero-shot Video Editing method. Under the guidance of depth maps and temporal consistency constraints, EVE derives satisfactory video editing results with an affordable computational and time cost. Moreover, recognizing the absence of a publicly available video editing dataset for fair comparisons, we construct a new benchmark named ZVE-50 dataset. Through comprehensive experimentation, we validate that EVE achieves a satisfactory trade-off between performance and efficiency. Codebase, datasets, and video editing demos are available at https://github.com/alipay/Ant-Multi-Modal-Framework/blob/main/prj/EVE.

Semantic recognition is pivotal in virtual reality (VR) applications, enabling immersive and interactive experiences. A promising approach is utilizing millimeter-wave (mmWave) signals to generate point clouds. However, the high computational and memory demands of current mmWave point cloud models hinder their efficiency and reliability. To address this limitation, our paper introduces ESP-PCT, a novel Enhanced Semantic Performance Point Cloud Transformer with a two-stage semantic recognition framework tailored for VR applications. ESP-PCT takes advantage of the accuracy of sensory point cloud data and optimizes the semantic recognition process, where the localization and focus stages are trained jointly in an end-to-end manner. We evaluate ESP-PCT on various VR semantic recognition conditions, demonstrating substantial enhancements in recognition efficiency. Notably, ESP-PCT achieves a remarkable accuracy of 93.2% while reducing the computational requirements (FLOPs) by 76.9% and memory usage by 78.2% compared to the existing Point Transformer model simultaneously. These underscore ESP-PCT's potential in VR semantic recognition by achieving high accuracy and reducing redundancy. The code and data of this project are available at \url{https://github.com/lymei-SEU/ESP-PCT}.

Data augmentation plays a critical role in improving model performance across various domains, but it becomes challenging with graph data due to their complex and irregular structure. To address this issue, we propose EPIC (Edit Path Interpolation via learnable Cost), a novel interpolation-based method for augmenting graph datasets. To interpolate between two graphs lying in an irregular domain, EPIC leverages the concept of graph edit distance, constructing an edit path that represents the transformation process between two graphs via edit operations. Moreover, our method introduces a context-sensitive cost model that accounts for the importance of specific edit operations formulated through a learning framework. This allows for a more nuanced transformation process, where the edit distance is not merely count-based but reflects meaningful graph attributes. With randomly sampled graphs from the edit path, we enrich the training set to enhance the generalization capability of classification models. Experimental evaluations across several benchmark datasets demonstrate that our approach outperforms existing augmentation techniques in many tasks.

Offline reinforcement learning (RL) is a learning paradigm where an agent learns from a fixed dataset of experience. However, learning solely from a static dataset can limit the performance due to the lack of exploration. To overcome it, offline-to-online RL combines offline pre-training with online fine-tuning, which enables the agent to further refine its policy by interacting with the environment in real-time. Despite its benefits, existing offline-to-online RL methods suffer from performance degradation and slow improvement during the online phase. To tackle these challenges, we propose a novel framework called ENsemble-based Offline-To-Online (ENOTO) RL. By increasing the number of Q-networks, we seamlessly bridge offline pre-training and online fine-tuning without degrading performance. Moreover, to expedite online performance enhancement, we appropriately loosen the pessimism of Q-value estimation and incorporate ensemble-based exploration mechanisms into our framework. Experimental results demonstrate that ENOTO can substantially improve the training stability, learning efficiency, and final performance of existing offline RL methods during online fine-tuning on a range of locomotion and navigation tasks, significantly outperforming existing offline-to-online RL methods.

Empathy allows us to assume others are like us and have goals analogous to our own. This can also at times be applied to multi-agent games - e.g. Agent 1's attraction to green balls is analogous to Agent 2's attraction to red balls. Drawing inspiration from empathy, we propose EMOTE, a simple and explainable inverse reinforcement learning (IRL) approach designed to model another agent's action-value function and from it, infer a unique reward function. This is done by referencing the learning agent's own action value function, removing the need to maintain independent action-value estimates for the modelled agents whilst simultaneously addressing the ill-posed nature of IRL by inferring a unique reward function. We experiment on minigrid environments showing EMOTE: (a) produces more consistent reward estimates relative to other IRL baselines (b) is robust in scenarios with composite reward and action-value functions (c) produces human-interpretable states, helping to explain how the agent views other agents.

We consider the problem of user-adaptive 3D gaze estimation. The performance of person-independent gaze estimation is limited due to interpersonal anatomical differences. Our goal is to provide a personalized gaze estimation model specifically adapted to a target user. Previous work on user-adaptive gaze estimation requires some labeled images of the target person data to fine-tune the model at test time. However, this can be unrealistic in real-world applications, since it is cumbersome for an end-user to provide labeled images. In addition, previous work requires the training data to have both gaze labels and person IDs. This data requirement makes it infeasible to use some of the available data. To tackle these challenges, this paper proposes a new problem called efficient label-free user adaptation in gaze estimation. Our model only needs a few unlabeled images of a target user for the model adaptation. During offline training, we have some labeled source data without person IDs and some unlabeled person-specific data. Our proposed method uses a meta-learning approach to learn how to adapt to a new user with only a few unlabeled images. Our key technical innovation is to use a generalization bound from domain adaptation to define the loss function in meta-learning, so that our method can effectively make use of both the labeled source data and the unlabeled person-specific data during training. Extensive experiments validate the effectiveness of our method on several challenging benchmarks.

Because of the wide deployment of smart contracts, smart contract vulnerabilities pose a challenging risk to blockchain security. Currently, deep learning-based vulnerability detection is a very attractive solution due to its ability to identify complex patterns and features. The existing methods mainly consider the contract code content features, expert knowledge patterns, and contract code modalities. To further enhance smart contract vulnerability detection, this paper attempts to identify community features from smart contracts with similar semantic and syntactic structures, and shared features from two related vulnerability detection tasks, vulnerability classification and localization. The experimental results verify that the proposed approach significantly outperforms the state-of-the-art methods in terms of accuracy, recall, precision, and F1-score.