Mingming Gong

Domain generalization aims to address distribution shifts between training and testing data, enabling models to generalize well to unseen domains. Recently, gradient matching methods have shown excellent performance. These methods learn gradients from different source domains through meta-learning strategy and seek invariant directions by maximizing the inner products between gradients from different domains to match ideal generalization gradient direction. However, existing approaches typically use empirical risk minimization to obtain gradient information on different source domains , which may cause gradients in different learning tasks to overfit the source domain information. It results in significant gradient conflict between gradient in different source domains in some scenarios, making it difficult to optimize the gradient inner product. Moreover, when source-domain samples lack sufficient diversity, the learned invariant gradient direction may fail to generalize to potential unseen domains. Therefore, invariant gradient matching based on empirical risk minimization is often ineffective in non-ideal gradient scenarios. To address these limitations, we propose an independence regularization gradient matching method. During each meta-learning task, it suppresses spurious correlations among learned features and reduces overfitting to domain-specific information through sample weighting. This encourages the gradient learned from each source domain to approach the ideal generalization direction and provides a better initialization for subsequent gradient matching. We conduct extensive experiments under diverse gradient environments. The experimental results consistently demonstrate the effectiveness and robustness of our method for domain generalization.

Randomized Controlled Trials (RCTs) represent the gold standard for causal inference yet remain a scarce resource. While large-scale observational data is often available, it is typically used only for retrospective fusion, and remains discarded in prospective trial design due to bias concerns. We argue that this ''tabula rasa'' data acquisition strategy is inefficient when observational models contain useful structural information. In this work, we propose Active Residual Learning, a new paradigm that leverages the observational model as an informative but biased prior. This shifts the experimental focus from learning target causal quantities from scratch to estimating residual corrections that debias the observational model. To operationalize this, we introduce the R-Design framework. Theoretically, we characterize two key advantages: (1) a conditional structural efficiency gap, showing that estimating lower-complexity residual contrasts can admit faster convergence rates than reconstructing full outcomes; and (2) information efficiency, where we quantify the redundancy in standard parameter-based acquisition, demonstrating that such baselines can waste budget on task-irrelevant nuisance uncertainty. We propose R-EPIG (Residual Expected Predictive Information Gain), a unified criterion that directly targets the downstream causal quantity, reducing residual uncertainty for estimation or clarifying decision boundaries for policy. Experiments on synthetic and semi-synthetic benchmarks show that R-Design significantly outperforms baselines in the intended informative-but-biased regime, supporting the effectiveness of correcting a biased model rather than learning from scratch.

Adversarial training (AT) has been shown to effectively enhance a network's resilience against adversarial attack. However, conventional AT, which relies on a fixed pre-specified perturbation budget, suffers from several limitations when training robust models. First, enforcing the same perturbation budget across networks with different capacities leads to varying levels of robustness disparity between natural and robust accuracies, which deviates from the desired outcome of a robust network. Second, because the perturbation budget is fixed throughout training, the attack strength fails to scale adaptively with the evolving robustness of the model. This mismatch often results in robust overfitting and further degradation of adversarial robustness. To address these limitations, we propose a novel technique called Strength-Adaptive Adversarial Training (SAAT). In SAAT, the adversary incorporates an adversarial-loss constraint to guide the generation of adversarial training data. This constraint allows the perturbation budget to adapt dynamically based on the current training state, which effectively mitigates robust overfitting. Moreover, by explicitly regulating the attack strength through the adversarial loss, SAAT enables precise control over the robustness disparity between natural accuracy and adversarial robustness. Extensive experiments demonstrate that SAAT substantially improves adversarial robustness over standard AT.

The rapid proliferation of Internet of Things (IoT) applications across heterogeneous Cloud-Edge-IoT environments presents significant challenges in distributed scheduling optimization. Existing approaches face issues, including fixed neural network architectures that are incompatible with computational heterogeneity, non-Independent and Identically Distributed (non IID) data distributions across IoT scheduling domains, and insufficient cross-domain collaboration mechanisms. This pa per proposes KD-AFRL, a Knowledge Distillation-empowered Adaptive Federated Reinforcement Learning framework that addresses multi-domain IoT application scheduling through three core innovations. First, we develop a resource-aware hybrid architecture generation mechanism that creates dual-zone neural networks enabling heterogeneous devices to participate in collaborative learning while maintaining optimal resource utilization. Second, we propose a privacy-preserving environment-clustered federated learning approach that utilizes differential privacy and K-means clustering to address non-IID challenges and facilitate effective collaboration among compatible domains. Third, we introduce an environment-oriented cross-architecture knowledge distillation mechanism that enables efficient knowledge transfer between heterogeneous models through temperature-regulated soft targets. Comprehensive experiments with real Cloud-Edge IoT infrastructure demonstrate KD-AFRL's effectiveness using diverse IoT applications. Results show significant improvements over the best baseline, with 21% faster convergence and 15.7%, 10.8%, and 13.9% performance gains in completion time, energy consumption, and weighted cost, respectively. Scalability experiments reveal that KD-AFRL achieves 3-5 times better performance retention compared to existing solutions as the number of domains increases.

Policy inference plays an essential role in the contextual bandit problem. In this paper, we use empirical likelihood to develop a Bayesian inference method for the joint analysis of multiple contextual bandit policies in finite sample regimes. The proposed inference method is robust to small sample sizes and is able to provide accurate uncertainty measurements for policy value evaluation. In addition, it allows for flexible inferences on policy comparison with full uncertainty quantification. We demonstrate the effectiveness of the proposed inference method using Monte Carlo simulations and its application to an adolescent body mass index data set.

3D Gaussian Splatting (3DGS) has become a mainstream representation for real-time 3D scene synthesis, enabling applications in virtual and augmented reality, robotics, and 3D content creation. Its rising commercial value and explicit parametric structure raise emerging intellectual property (IP) protection concerns, prompting a surge of research on 3DGS IP protection. However, current progress remains fragmented, lacking a unified view of the underlying mechanisms, protection paradigms, and robustness challenges. To address this gap, we present the first systematic survey on 3DGS IP protection and introduce a bottom-up framework that examines (i) underlying Gaussian-based perturbation mechanisms, (ii) passive and active protection paradigms, and (iii) robustness threats under emerging generative AI era, revealing gaps in technical foundations and robustness characterization and indicating opportunities for deeper investigation. Finally, we outline six research directions across robustness, efficiency, and protection paradigms, offering a roadmap toward reliable and trustworthy IP protection for 3DGS assets.

Significant progress has been made in the field of video question answering (VideoQA) thanks to deep learning and large-scale pretraining. Despite the presence of sophisticated model structures and powerful video-text foundation models, most existing methods focus solely on maximizing the correlation between answers and video-question pairs during training. We argue that these models often establish shortcuts, resulting in spurious correlations between questions and answers, especially when the alignment between video and text data is suboptimal. To address these spurious correlations, we propose a novel training framework in which the model is compelled to acknowledge its ignorance when presented with an intervened question, rather than making guesses solely based on superficial question-answer correlations. We introduce methodologies for intervening in questions, utilizing techniques such as displacement and perturbation, and design frameworks for the model to admit its lack of knowledge in both multi-choice VideoQA and open-ended settings. In practice, we integrate a state-of-the-art model into our framework to validate its effectiveness. The results clearly demonstrate that our framework can significantly enhance the performance of VideoQA models with minimal structural modifications.