Generalized federated learning seeks to develop robust models across distributed source domains that generalize well to the unseen target domain. Mainstream methods make strict assumptions about the availability of target domain data, limiting the flexibility and adaptability of real-world applications. In this work, we tackle a real-world challenge that has never been addressed before: federated multi-source domain adaptation for an unseen target domain. We propose federated cross-domain semantic alignment with adversarial feature augmentation, a method that enhances model generalization across domains. Our method operates in the feature space to capture both diversity and invariance between source and target domains through a two-stage local training strategy. In the adversarial training phase, a domain identifier and feature discriminator constrain the generated features to extract target-relevant information. During the contrastive learning stage, a semantic representation alignment loss (SRA) is incorporated to align class prototype distributions between source and target domains, ensuring uniform classification standards. Federated aggregation consolidates model knowledge across clients, facilitating collaborative evolution and rapid adaptation to the unseen target domain. Extensive experimental results on four prevalent datasets demonstrate that our approach outperforms existing benchmarks across different backbones, showcasing its effectiveness in scenarios with data silos.

Non-Markovian Tasks (NMTs) are distinguished by their dependence on long-term memory and state-dependent dynamics, setting them apart from the traditional Markovian models typically employed in Reinforcement Learning (RL). NMTs not only suffer from reward sparseness but also rely on historical information, making their resolution considerably more challenging. In this paper, we propose a novel RL framework T4NMTD (Transition-centric framework for NMT Decomposition), designed specifically for learning NMTs which are specified by temporal logic. The core of T4NMTD is a task decomposition mechanism along with a parallel training approach for NMTs. An NMT is first decomposed as basic units based on the transitions of the automata which are derived from temporal logic formulae. The units are then modularized into sub-tasks according to their semantic similarity under logical interpretation. The training strategy of T4NMTD adopts a dual-level structure: the high-level learns to shape the boundaries and coordinate arrangement of the sub-tasks from a global perspective, while the low-level learns those sub-tasks in parallel. In addition, we invent a dynamic policy intervention scheme to mitigate the policy myopic issue during parallel training. A comprehensive evaluation is conducted on benchmark problems with respect to various metrics. The experimental results demonstrate that T4NMTD effectively addresses NMTs, achieving significant performance improvements compared with related studies.

In recent years, multi-label image recognition has emerged as a crucial task in computer vision, requiring simultaneous detection of multiple objects or attributes within images. Unlike single-label classification, this task demands explicit modeling of complex label correlations. Existing methods primarily focus on low-order pairwise relationships, failing to capture higher-order dependencies critical for real-world scene understanding. Additionally, the long-tail distribution of labels often causes models to prioritize frequent head labels while neglecting rare tail labels with limited training samples. To address these challenges, we propose a Dynamic Multi-modal Hypergraph Learning (DMHL) framework for semi-supervised multi-label recognition. DMHL constructs adaptive hypergraphs by fusing visual features, co-occurrence statistics, and textual embeddings. The framework dynamically refines these hypergraphs through three novel modules: HyperPrune, which prunes redundant hyperedges; HyperTransform, which generates dynamic hyperedges from node features; and HyperTune, which optimizes hypergraph weights via feature similarity alignment. These dynamic optimization modules enable DMHL to capture intricate high-order label correlations. Furthermore, DMHL employs hypergraph residual concatenation to enhance deep feature representations, which are leveraged for dynamic pseudo-label generation to alleviate label imbalance. Extensive experiments demonstrate DMHL achieves state-of-the-art results across four benchmarks: 86.0% mAP on MS-COCO (0.7% gain over prior SOTA), 96.3%/96.5% on Pascal VOC 2007/2012, and 64.0% on NUS-WIDE. Notably, in semi-supervised settings with 5% labeled data, DMHL surpasses previous methods by over 20% mAP on MS-COCO (70.5%), highlighting its effectiveness in capturing intricate label relationships and improving tail label recognition.

Graph Neural Networks (GNNs) often suffer from degraded performance when encountering out-of-distribution (OOD) samples, particularly in multi-domain graph scenarios. Existing graph OOD detection methods typically require extensive modifications to data or model architectures, resulting in high computational costs and limited generalization. Moreover, prior approaches largely overlook local structural semantics and community-level patterns, leading to biased representations and suboptimal detection performance. To overcome these limitations, we propose community enhanced graph out-of-distribution detection (CEGOOD), a novel framework that incorporates community structure into GNN-based OOD detection. Specifically, we propose two community-aware view generation strategies: intra-community attribute aggregation (ICAA) to distill fine-grained feature coherence and inter-community edge dropping (ICED) to fortify structural robustness by pruning non-critical cross-community edges. Furthermore, We also design three community-level loss functions (compactness, separability, and balance) to optimize community hierarchical structures and improve community representation. Experimental results on various datasets show that CEGOOD outperforms state-of-the-art baselines by an average of 1.8% AUC, with notable gains of 2.4% on AIDS+DHFR and 2.8% on BBBP+BACE, demonstrating superior adaptability and effectiveness in graph OOD detection tasks.

Graph matching aims to establish node correspondences between graphs, which is a classic combinatorial optimization problem. In recent years, (deep) learning-based methods have emerged as a superior alternative to traditional graph matching solvers. However, these methods typically rely on node-level correspondence labels, which can be prohibitively expensive or unrealistic. Inspired by contrastive learning that is a prevalent paradigm for self-supervised representation learning, we develop a Contrastive Learning Network for Unsupervised Graph Matching (CUGM), which is an end-to-end differentiable pipeline to learn node permutations. Specifically, we propose three-level augmentation including raw image augmentation, graph augmentation and model augmentation for generating diverse enough contrastive views to enrich training instances. Then a contrastive learning network is constructed to capture the higher-order structural information in graphs and learn the final node representations for yielding the affinity matrix to directly solve a linear assignment problem. More importantly, we propose a node-level contrastive loss with false negative cancellation for optimizing the whole network to extract the tailored node feature representations to improve graph matching accuracy. Experimental results on standard graph matching benchmarks demonstrate that our end-to-end unsupervised method achieves the competitive performance compared with state-of-the-art supervised and unsupervised graph matching methods.

Federated semi-supervised learning presents a pragmatic scenario wherein a centralized model is trained utilizing a server with access to labeled data, while participating clients lack any labeled data. In this context, the inaccuracy of real-world labels on the server available for training poses a huge challenge to the federated semi-supervised learning. These inaccuracies can have a detrimental impact on the overall performance of the system and impose limitations on its use. In this paper, we propose a novel Federated Semi-supervised learning framework with Contrastive Representations, called FedCR, with the aim of addressing the aforementioned ubiquitous problems in the field of image classification tasks. Firstly, our approach employs contrastive representation learning to build memory representations of images, which can learn an image's general features from an augmented view without relying on negative pairs and prevent the model from memorizing noise. Then we take a cautious approach during model updates to prevent any potential leakage to ensure the privacy and security of the clients' information. Additionally, for the sake of improving robustness of the model, a contrastive regularization function is applied to preserve information connected to true labels while filtering out information associated with wrong labels. Furthermore, we mitigate the negative impact of mislabeled data during supervised learning by utilizing an improved cross-entropy loss function. Extensive experiments on prevalent datasets for image classification tasks show that the proposed method surpasses previously established state-of-the-art federated semi-supervised learning algorithms and efficiently alleviates the issue of model over-fitting to erroneous labels, especially when label noise is present.

Methods based on deep learning for intelligent fault diagnosis have shown good results in general diagnostic tasks. Nevertheless, these methods largely depend on the sufficient labeled data, limiting their application in the actual scenarios where the availability of labeled data is limited. Moreover, the distribution of testing data is inconsistent with that of training data because bearings operate in various working conditions, leading to the performance degradation of these approaches. To tackle these two entangled problems, we propose a novel unsupervised domain adaptation network, which presents clustering-guided prototypical contrastive learning for cross-domain fault diagnosis. More specifically, k-means clustering is first used to aggregate similar source samples and target samples separately, acquiring the centroid of each cluster and the cluster index of each sample. Then, we propose in-domain and cross-domain contrastive learning strategies based on clustering results to achieve class alignment and domain alignment across source domain and target domain. By applying in-domain contrastive learning, we make the intra-class distance smaller while making the inter-class distance larger within each domain, effectively reducing the number of samples on the class boundaries. By applying cross-domain contrastive learning, class-to-class semantic similarity across two different domains is considered, which not only retains class discriminability in each domain but aligns these two domains at both the class level and the domain level. Detailed experiments on three bearing datasets reveal that our method outperforms in fault diagnosis across diverse working conditions, achieving average accuracy improvements of 2.10%, 7.44%, and 1.17% on the JNU, HUST, and Ottawa datasets, respectively.

Federated Learning enables clients to train a joint model collaboratively without disclosing raw data. However, learning over non-IID data may raise performance degeneration, which has become a fundamental bottleneck. Despite numerous efforts to address this issue, challenges such as excessive local computational burdens and reliance on shared data persist, rendering them impractical in real-world scenarios. In this paper, we propose a novel federated knowledge transfer framework to overcome data heterogeneity issues. Specifically, a model segmentation distillation method and a learnable aggregation network are developed for server-side knowledge ensemble and transfer, while a client-side consistency-constrained loss is devised to rectify local updates, thereby enhancing both global and client models. The framework considers both diversity and consistency among clients and can serve as a general solution for extracting knowledge from distributed nodes. Extensive experiments on four datasets demonstrate our framework’s effectiveness, achieving superior performance compared to advanced competitors in high-heterogeneity settings.

Greybox fuzzing is a powerful testing technique. Given a set of initial seeds, greybox fuzzing continuously generates new test inputs to execute the program under test and gravitates executions towards rarely explored program regions with code coverage as feedback. Seed prioritization is an important step of greybox fuzzing that prioritizes promising seeds for input generation. However, mainstream greybox fuzzers like AFL++ and Zest tend to slight the importance of seed prioritization and plainly pick seeds according to the order of the seeds being queued, or rely on an approach with randomness, which may consequently degrade their performance. In this paper, we propose a novel distance-based seed prioritization approach named DiPri to facilitate greybox fuzzing. Specifically, DiPri calculates the distances among seeds and selects the ones that are farther from the others in priority to improve the probabilities of discovering previously unexplored regions. To make a preliminary evaluation, we integrate DiPri into AFL++ and Zest and conduct experiments on eight (four in C/C++ and four in Java) fuzz targets. We also consider six configurations, i.e., three prioritization modes multiplied by two distance measures, in our evaluation to investigate how different prioritization timings and measures affect DiPri. The experimental results show that, compared to the default seed prioritization approaches of AFL++ and Zest, DiPri covers 1.87%∼13.86% more edges in three out of four C/C++ fuzz targets and 0.29%∼4.97% more edges in the four Java fuzz targets with certain configurations. The results highlight the potential of facilitating greybox fuzzing with distance-based seed prioritization.

Federated Learning (FL) offers a promising solution for semantic segmentation in scenarios involving data distribution across isolated clients. Despite recent advances, federated semantic segmentation (FSS) continues to face key challenges. One major issue is the shift from centralized to decentralized training, where diverse and limited local data hinder consistent pixel-level representation learning. Another challenge is data heterogeneity from imbalanced class distributions across clients, which weakens feature consistency and degrades global performance. These limitations often lead to inconsistent feature learning and degraded global performance. To address the challenges of class heterogeneity and insufficient pixel-level representation learning in FSS, we propose a novel pixel-aware FSS framework that improves local adaptation and semantic consistency. Specifically, we design a fine-tuning strategy that initializes each client with a lightweight pre-trained model and performs local updates over multiple epochs. This improves model adaptability to local distributions while reducing communication overhead. To further enhance semantic consistency across heterogeneous clients, we introduce a client clustering strategy based on pixel-level semantic features. Clients with similar class distributions are grouped to encourage consistent feature learning within clusters. Cluster-level training and aggregation are then followed by a global aggregation step, promoting more robust and aligned semantic understanding. Empirical evaluation across multiple benchmark datasets confirms that our method achieves consistently high segmentation precision and enhanced model adaptability in highly heterogeneous federated scenarios.

Multi-label classification is crucial in image recognition. However, data class imbalance can lead to poor performance in classification algorithms, particularly for minority classes, thereby impacting overall accuracy. Existing augmentation methods for minority class data typically adopt a dataset-wide approach, failing to enhance minority class samples based on the model’s performance for each class during training. In this paper, we propose an end-to-end Self-paced Deep Multi-label data Augmentation (SDMA) method which is capable of dynamically generating image data for minority classes in multi-label classification. Our method selects a batch of relatively simple samples as the training set during each training iteration. The number of samples from each category in this batch reflects the model’s training performance on those categories. By generating data for the minority classes within these samples, we can dynamically generate samples that better meet the model’s needs based on its training progress. Additionally, to mitigate the impact of noisy data in the generated set, we compute the predicted similarity of the generated data with the seed and reference instances, excluding low-similarity data from influencing the model. Extensive experiments on multi-label datasets demonstrate that SDMA is competitive compared to other state-of-the-art methods.

Heuristic search provides an efficient way to automatically explore the state space in planning, while Control Knowledge (CK) also has the potential to significantly increase the performance of planners. Currently, most of the state-of-the-art planners primarily rely on sophisticated heuristic mechanisms. However, these planners fail to scale up and to provide (high-quality) solutions in a range of problems. The objective of this paper is to incorporate CK with heuristic search in order to leverage the advantages of both, thus leading planners to achieve much higher efficiency. To achieve this, we introduce a novel CK which is specified by a variant of Linear Temporal Logic (LTL), referred to as LTL.. We propose an encoding methodology that translates LTL. into standard planning models. Consequently, we can directly use existing heuristic planners to solve the augmented problem, and avoid tailoring the planners in order to deal with CK implicitly. The novelty of this approach lies in that we define a useful CK LTL. with a concise encoding methodology, that can significantly improve the efficiency of heuristic search. In this paper, the encoding process is formally presented, and theoretical results on the complexity and soundness of the encoding are strictly proved. We find that appropriate CK is a good complement to heuristic search, and is capable of making hard problems easy to solve. Experiments demonstrate that our approach shows highly competitive results versus heuristic search and other CK-based techniques on many intractable benchmark problems, benefiting in improving the coverage and quality of plans.

Recently, significant strides have been made in the field of representation learning. Nevertheless, prior methods have predominantly centered on supervised learning, which necessitates a reliance on costly labeled datasets. In response to this challenge, graph contrastive learning leverages the potential of unlabeled data by maximizing the consistency between similar graph pairs, extracting effective representations of graph data. However, existing methods lack the consideration of multi-scale information and tend to focus on extracting local graph features, especially overlooking the interdependencies between graph-level representations. To overcome these challenges, we present an innovative framework, Multi-Scale Contrast for Global-Interaction Graph Representation Learning (MSCGI) to capture more abundant features. Firstly, our framework models the given graph at multiple scales, with subgraphs at each scale encoded by independent encoders. Subsequently, we design a Global Interaction module that generates edges among graphs based on the similarity of graph-level representations, thereby constructing a “global graph” to capture the interrelations among graphs. Finally, the framework maximizes the mutual information across the various scales to capture hierarchical information. Comprehensive experiments demonstrate that MSCGI outperforms the state-of-the-art unsupervised methods.

Contract-based design is a useful software engineering paradigm and has been exploited in many literatures for taming the complexity of embedded system development. A system is divided hierarchically into components in a top-down manner, where each component is associated with contracts structured in pairs of assumptions and guarantees. The strength of contract-based design enables stepwise refinement, compositional verification, and reuse of components etc. In this paper, we present a verification framework that builds upon contract-based design to ensure the correctness of embedded software. This framework can be integrated with many formal techniques, e.g., compositional verification, model checking, static analysis, runtime verification, that are shaped in a formal verification chain from the design perspective to the implementation perspective. First, a design level hierarchical model made up of components is abstracted for embedded software, and contracts relevant to components are specified and verified in temporal logic. Second, the behaviour models are constructed for the components located at the bottom of the hierarchical model, and verified using the same contracts at the first step. Third, the implementation of the bottom components are verified, still using the same contracts. Consequently, the consistency between design and implementation of embedded software is strictly proved. We provide a practical case study in the field of aerospace to illustrate the feasibility of our approach, where temporal logic LTL (Linear Temporal Logic) is employed for reasoning as contracts throughout the verification chain.

Federated unsupervised representation learning aims at leveraging unlabeled data from multiple parties to learn visual representations without compromising the data privacy and tackle the non-IID challenge by aligning diverse representation spaces. However, model heterogeneity and communication overhead will directly impact the convergence rate and model accuracy of federated unsupervised learning. And it is challenging to learn visual features for downstream tasks under the premise of compatibility with heterogeneous models and reducing communication overhead. To address these issues, we propose a novel communication-efficient federated unsupervised representation learning framework based on prototype similarity distillation (FLPD). In this framework, the global model builds the feature representation space based on the global dataset and steers the optimization of the prototype relations of the client models. In addition to employing discriminative self-supervised learning for model training, each client fine-tunes the local representation space with global prototype similarity via knowledge distillation, which facilitates local models to fit both the local data distribution and the global representation space. In order to maintain the compactness of prototypes within the same category and enhance the separability between prototypes of different categories, a prototype-based consistency constraint is introduced to alleviate the conflict between local and global representation space. Experimental results demonstrate that our framework outperforms other alternative approaches in terms of communication efficiency and accuracy in the federated settings with statistical heterogeneity and model heterogeneity.

The commonly used Reinforcement Learning (RL) model, MDPs (Markov Decision Processes), has a basic premise that rewards depend on the current state and action only. However, many real-world tasks are non-Markovian, which has long-term memory and dependency. The reward sparseness problem is further amplified in non-Markovian scenarios. Hence learning a non-Markovian task (NMT) is inherently more difficult than learning a Markovian one. In this paper, we propose a novel Parallel and Modular RL framework, ParMod, specifically for learning NMTs specified by temporal logic. With the aid of formal techniques, the NMT is modulaized into a series of sub-tasks based on the automaton structure (equivalent to its temporal logic counterpart). On this basis, sub-tasks will be trained by a group of agents in a parallel fashion, with one agent handling one sub-task. Besides parallel training, the core of ParMod lies in: a flexible classification method for modularizing the NMT, and an effective reward shaping method for improving the sample efficiency. A comprehensive evaluation is conducted on several challenging benchmark problems with respect to various metrics. The experimental results show that ParMod achieves superior performance over other relevant studies. Our work thus provides a good synergy among RL, NMT and temporal logic.

Federated learning (FL) unites multiple participants to collaboratively learn a global consensus model on the centralized server by aggregating their individual models trained locally on clients. To meet the goal of obtaining an optimal model, sufficient labeled data and myriad communications are required during training. However, the major problems are the limited budget for manually annotating unlabeled instances and the restricted bandwidth of server and clients. This article presents a communication-efficient federated active semi-supervised learning (CEFedASSL) framework that unites active learning (AL) clients and a semi-supervised learning (SSL) client to train models on unlabeled data while achieving communication efficiency. In each AL client, different query strategies are, respectively, applied for the local model to obtain a more robust model and query only the optimal samples which significantly reduces the cost of annotation. Subsequently, these optimal samples are encrypted as input to fine-tune the pretrained model of the SSL client by performing self-training, thereby enhancing the model performance while preserving the privacy of data. Furthermore, we propose an efficient selective aggregation strategy to reduce the communication cost between clients and the server. Empirical experiments on four different learning tasks demonstrate that the proposed CEFedASSL distinctively outperforms the common FL algorithms in terms of both model performance and communication costs.

Due to the unique advantages of collaborative learning on isolated yet unlabeled data, federated self-supervised learning has received increasing attention from both academic and industrial researchers. Most of the existing federated self-supervised approaches concentrate on the classical scenario, i.e., a large amount of unlabeled data is stored on the clients. However, in many real-world applications, partial labels may be available to the client user, while a large amount of unlabeled data remains on the server side. The existing federated self-supervised methods may usually have difficulty in addressing this scenario. In this paper, we propose a creative federated global self-supervised framework (FGSS) for large-scale unlabeled data that innovatively uses self-supervised learning on the server side, and during every round of communication, we use a small amount of labeled data from the client to facilitate the performance of self-supervised learning. To address the heterogeneity of local data from different clients, we designed an aggregation approach that can adjust the weight of each local model based on the frequency of participation in the communication and the size of its dataset. Experimental results show that our framework outperforms the most existing state-of-the-art methods in both IID and non-IID settings under certain conditions.

Documents on popular libraries such as pyOpenSSL do not specify parameters’ domains precisely. Inexact domains hinder efficiently generating valid test cases. In order to solve this problem, an approach for inferring exact domains, namely IED, is put forward. IED starts from crawling parameters’ data types from online official documents. Then, IED conducts exception testing to find tight domains based on slack domains of data types. Finally, IED attempts to detect the extensibility of tight bounds and outputs exact domains. Based on IED, experiments have been conducted on 29 basic parameters of application programming interfaces of pyOpenSSL. Experimental results show that the inferred exact domains vary dramatically even for the same data type and reduce the original domains of data types significantly. Thus, IED is effective and helpful in efficiently generating valid test cases.