Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· pp. 115-126· 0 citations· 18 references
Abstract
Neural network repair aims to correct prediction failures caused by multiple security threats—such as backdoor attacks, natural corruptions, and safety property violations—through limited adjustments to model parameters. However, most existing repair methods rely on single-sample, point-to-point correction strategies, overlooking the statistical regularities of the feature space. As a result, they are highly sensitive to the scale of faulty samples and struggle to simultaneously achieve repair generalization and original performance preservation under small-sample settings. To address these limitations, we propose a novel general neural network repair paradigm termed NCCDA (Neuron-wise Class-Conditional Distribution Alignment). The method is grounded in a key insight: prediction failures fundamentally arise from neuron-level internal representations deviating from the high-likelihood regions corresponding to their true classes. NCCDA constructs neuron-wise class-conditional distribution references and formulates the repair process as a joint optimization of distribution alignment and structure preservation. By guiding abnormal representations back to high-likelihood regions while anchoring the structure of normal samples, the method enables efficient and adaptive repair without explicit neuron localization. We theoretically prove a generalization error bound under small-sample settings based on Rademacher complexity, providing formal guarantees. Extensive experiments across 7 benchmark datasets and 38 models, covering three categories of repair tasks, demonstrate that NCCDA consistently outperforms existing methods in repair effectiveness, generalization repair capability (Gene), and original accuracy preservation.
VeRe is proposed, a verification-guided repair framework that leverages linear relaxation to precisely and efficiently estimate the repair significance of neurons and synthesizes ideal intervals that provide sound guarantees for correct behaviors, thereby facilitating surgical and targeted adjustments of neuron parameters to rectify faulty behaviors.
Jia-Nan Ma, Wei Chen, Pengfei Yang et al.· ACM Transactions on Software...· 0 citations
With the widespread deployment of deep neural networks (DNNs) in safety-critical domains, reducing the cost of model validation under limited testing budgets has become increasingly important. Existing test case prioritization techniques often rely on single-checkpoint confidence signals derived from output probabilities. However, DNNs can be confidently wrong, and the confidence margin between the predicted and competing classes is frequently small, which weakens early fault discovery. To address this limitation, we propose a Neural-Collapse-Inspired Prioritization (NCIP) framework that replaces absolute confidence with cross-checkpoint prediction variability in the terminal training regime, where model geometry becomes highly structured. NCIP introduces two key components. First, it selects an NC-guided representative subset of training checkpoints using an equiangularity score of classifier weights, quantified as the standard deviation of pairwise cosine similarities among class weight vectors. Second, it prioritizes test inputs by their prediction variability across the selected checkpoints, surfacing boundary-adjacent and failure-prone samples that are unstable under checkpoint-induced decision boundary shifts. Extensive experiments across multiple datasets and architectures show that NCIP achieves strong performance in early fault discovery compared with competitive baselines, with 1.5 to 16.6 percent RAUC-ALL gains and 4.9 to 20.6 percent RAUC-500 gains under the same testing budget. NCIP further attains the best average performance across all dataset-model pairs.
Chunyu Liu, Mingyuan Li, Yang Li et al.· arXiv.org· 0 citations
A fine-tuning-stage defense that simultaneously hardens LLMs against both attack classes by redistributing safety signals across a broader set of neurons, and provides a formal guarantee that NeuronGuard strictly reduces the attack success rate (ASR) upper bound.
Anjun Gao, Yueyang Quan, Yu Xia et al.· 0 citations
Fault diagnosis for complex industrial equipment plays a crucial role in safeguarding production safety and advancing the capabilities of intelligent operation and maintenance. Current deep learning approaches have demonstrated promising accuracy in fault classification tasks; however, their signal representations alone cannot provide a transparent interface for embedded large language models. To tackle the aforementioned challenges, we propose PGA-LLM, a novel fault diagnosis framework for industrial equipment that leverages large language models via probability-guided alignment. First, a variational autoencoder (VAE)-based signal encoder embedded with reconstruction constraints is established. Joint reconstruction and classification objectives balance discriminative representation learning and signal reconstruction. Second, the probability-guided alignment (PGA) module combines fault-class probability guidance with a residual feature path; a learned gate fuses both paths before continuous soft-prompt projection. Furthermore, a progressive three-stage training scheme is adopted, encompassing encoder pre-training, interface optimization, and low-rank adaptation (LoRA) of Qwen2.5-1.5B. Extensive experiments are carried out on four standard datasets, CWRU, Gear, Mixed, and MBHM, and the Stage 2 signal-side output achieves classification accuracies of 97.1%, 99.0%, 93.4%, and 96.3%, respectively. The report-generation branch provides a schema-constrained signal-to-language interface for maintenance-oriented reporting.
Tao Wang, Yanqiang Di, Shao-Chong Feng et al.· Technologies· 0 citations
Deep neural networks (DNNs) are known to produce erroneous results under real-world noisy inputs, presenting a major bottleneck to their use in applications where lives, safety, or significant resources are at stake. It has been commonly observed that humans are highly resilient to the noisy inputs that are challenging for DNNs. However, very few efforts have translated this observation into techniques to improve DNN robustness. We hypothesize that statistically aligning DNNs to human behavior during training could improve robustness. Based on this insight, we propose BrainTrain, a framework to create more robust DNNs through human behavior alignment and demonstrate its utility in the context of object recognition. BrainTrain captures human behavior in the form of a confusion matrix constructed from human responses to object recognition challenges and uses a composite loss function to co-optimize accuracy and human behavior alignment during stochastic gradient descent (SGD) based training. We also propose Similarity Driven Label Smoothing (SDLS), a regularization method that scales BrainTrain to applications where it is difficult or expensive to collect human behavioral data. DNNs trained with BrainTrain showed up to 26% higher accuracy under a wide range of noisy inputs and 2.1 times lower calibration error with negligible increase in training time. We also demonstrate that SDLS leads to improvements in noise robustness in ResNets trained on the ImageNet-1 K image classification dataset. We show that BrainTrain is complementary to conventional techniques like noise-added training and can provide further improvements over and above these techniques.
Bharath Anand, Sarada Krithivasan· Frontiers in Artificial Inte...· 0 citations
Testing three delivery mechanisms–supervised fine-tuning, a dual-head classification loss, and reinforcement learning with a dense reward derived from the normalised penalty finds that supervised approaches consistently regress below the zero-shot baseline under distribution shift, while GRPO succeeds and generalises across independent training seeds.
Muntasir Adnan, Manile Srun, Carlos C. N. Kuhn· Machine Learning and Knowled...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
