Search for: All records

Abstract Industrial environments demand accurate detection of anomalies to maintain product quality and ensure operational safety. Traditional industrial anomaly detection (IAD) methods often lack the flexibility and adaptability needed in dynamic production settings, where new defect types and operational changes continually emerge. Recent advancements in multimodal large language models (MLLMs) have shown promise by combining visual and textual processing capabilities, yet they are often limited by their lack of domain-specific expertise, particularly regarding industry-standard defect tolerances. To overcome limitations, we introduce Echo, a novel multi-expert framework designed to enhance MLLM performance for IAD. Echo integrates four specialized modules: the Reference Extractor retrieves similar normal images to establish contextual baselines; the Knowledge Guide provides critical, industry-specific insights; the Reasoning Expert enables structured, stepwise analysis for complex queries; and the Decision Maker synthesizes information from the preceding modules to deliver precise, context-aware responses. Evaluations on the MMAD benchmark reveal that Echo significantly improves adaptability, precision, and robustness compared to conventional approaches. Our results demonstrate that guided MLLMs, when augmented with expert modules, can effectively bridge the gap between general visual understanding and the specialized requirements of industrial anomaly detection, paving the way for more reliable and interpretable inspection systems. 

With the adoption of machine learning models in various practical domains, there is a growing need for evaluating and increasing model robustness. Hyperdimensional computing (HDC) is a neurosymbolic computational paradigm that represents symbols as high dimensional vectors and symbolic operations as vector operations, seamlessly interfacing between neuro- and symbolic components of a model. However, there is a notable gap in HDC research regarding the robustness of HDC models to input perturbations. This study presents a novel theoretical framework tailored to evaluate the robustness of hyperdimensional classifiers against perturbations in the input space. In particular, our proposed measure of robustness gives a theoretical upper bound for the magnitude of noise a model can tolerate without changing its prediction for any given data point. We also propose a method to enhance the robustness of the model based on our proposed measure of robustness. Our approach introduces several methods to calculate model robustness as a function of the specific dataset and type of hyperdimensional encoding used. The results show that the average robustness of HDC models increases under the proposed optimization scheme while maintaining accuracy by varying the variance of the Gaussian distribution used to encode hypervectors. The practical effectiveness of our proposed measure of robustness is also demonstrated.