The increasing sophistication and frequency of cyber attacks necessitate automated and intelligent response mechanisms that can adapt to evolving threats. This paper presents ARCS (Adaptive Reinforcement learning for Cybersecurity Strategy), a novel framework that leverages deep reinforcement learning to optimize automated incident response strategies in cybersecurity systems. Our approach uniquely combines state representation learning of security events with a hierarchical decision-making process to map attack patterns to optimal defense measures. The framework employs a custom reward mechanism that balances incident resolution time, system stability, and defense effectiveness. Using a comprehensive dataset of 20,000 cybersecurity incidents, we demonstrate that ARCS achieves 27.3% faster incident resolution times and 31.2% higher defense effectiveness compared to traditional rule-based approaches. The framework shows particular strength in handling complex, multi-stage attacks, reducing false positive rates by 42.8% while maintaining robust system performance. Through extensive experiments, we validated that our approach can effectively generalize across different attack types and adapt to previously unseen threat patterns. The results suggest that reinforcement learning-based automation can significantly enhance cybersecurity incident response capabilities, particularly in environments requiring rapid and precise defensive actions.

Detecting anomalies in distributed systems through log analysis remains challenging due to the complex temporal dependencies between log events, the diverse manifestation of system states, and the intricate causal relationships across distributed components. This paper introduces a TLAN (Temporal Logical Attention Network), a novel deep learning framework that integrates temporal sequence modeling with logical dependency analysis for robust anomaly detection in distributed system logs. Our approach makes three key contributions: (1) a temporal logical attention mechanism that explicitly models both time-series patterns and logical dependencies between log events across distributed components, (2) a multi-scale feature extraction module that captures system behaviors at different temporal granularities while preserving causal relationships, and (3) an adaptive threshold strategy that dynamically adjusts detection sensitivity based on system load and component interactions. Extensive experiments on a large-scale synthetic distributed system log dataset show that TLAN outperforms existing methods by achieving a 9.4% improvement in F1-score and reducing false alarms by 15.3% while maintaining low latency in real-time detection. The framework demonstrates particular effectiveness in identifying complex anomalies that involve multiple interacting components and cascading failures. Through comprehensive empirical analysis and case studies, we validate that TLAN can effectively capture both temporal patterns and logical correlations in log sequences, making it especially suitable for modern distributed architectures. Our approach also shows strong generalization capability across different system scales and deployment scenarios, supported by thorough ablation studies and performance evaluations.

Supply chain network design (SCND) is a critical strategic challenge that significantly influences operational efficiency and competitiveness in the global marketplace. This paper introduces Think4SCND, a novel deep reinforcement learning framework that addresses the dynamic complexities of SCND by integrating a Supply Chain Transformer Network (SCTN) with a Thinking Model. Our approach formulates the SCND problem as a Markov Decision Process, developing an architecture capable of handling mixed discrete-continuous action spaces. The core innovation lies in the Thinking Model, which enhances the SCTN’s ability to reason about future states and evaluate decision sequences, enabling more informed and forward-looking decision-making. We propose an end-to-end training algorithm that effectively combines model-free reinforcement learning with model-based planning. Extensive experiments on both synthetic and real-world datasets show that Think4SCND significantly outperforms state-of-the-art baselines, achieving near-optimal solutions with a fraction of the computational cost. The framework demonstrates superior adaptability to disruptions and strong generalization capabilities, successfully transferring knowledge from medium-sized problems to larger, unseen instances.

Automated food safety inspection systems rely heavily on the visual detection of contamination, spoilage, and foreign objects in food products. Current approaches typically require extensive labeled training data for each specific hazard type, limiting generalizability to novel or rare safety issues. We propose a zero-shot detection framework for visual food safety hazards that enables the identification of previously unseen contamination types without requiring explicit training examples. Our approach adapts and extends the Knowledge-Enhanced Feature Synthesizer (KEFS) methodology to the food safety domain by constructing a specialized knowledge graph that encodes visual safety attributes and their correlations with food categories. We introduce a Food Safety Knowledge Graph (FSKG) that models the relationships between 26 food categories and 48 visual safety attributes (e.g., discoloration, mold patterns, foreign material characteristics) extracted from food safety databases and expert knowledge. Using this graph as the prior knowledge, our system synthesizes discriminative visual features for unseen hazard classes through a multi-source graph fusion module and region feature diffusion model. Experiments on our newly constructed Food Safety Visual Hazards (FSVH) dataset demonstrate that our approach achieves 63.7% mAP in zero-shot hazard detection, outperforming state-of-the-art general zero-shot detection methods by 6.9%. Furthermore, our framework demonstrates robust generalization to fine-grained novel hazard categories while maintaining high detection performance (59.8% harmonic mean) in generalized zero-shot scenarios where both seen and unseen hazards may occur simultaneously. This work represents a significant advancement toward automated, generalizable food safety inspection systems capable of adapting to emerging visual hazards without a costly retraining process.

This study introduces a novel framework for the inverse design of sustainable food packaging materials using generative adversarial networks (GANs) and the recently released OMat24 dataset containing 110 million DFT-calculated inorganic material structures. Our approach transforms traditional material discovery paradigms by enabling end-to-end design from desired performance metrics to material composition. We developed a GAN-driven inverse design architecture specifically optimized for food packaging applications, integrating sensor-derived data on critical constraints such as biodegradability and barrier properties directly into the generative process. This integration occurs at three levels: (1) sensor-measured properties define conditioning targets for the GAN, (2) sensor data train the property prediction network, and (3) sensor-based characterization validates generated materials. An enhanced EquiformerV2 graph neural network was employed to accurately predict the formation energy, stability, and sensor-measurable properties of candidate materials. The model achieved a mean absolute error of 12 meV/atom for formation energy on the OMat24 test set (25% improvement over baseline models), while predictions of sensor-measured functional properties reached R2 values of 0.84–0.89 through the integration of experimental measurements and physics-based proxy models. The framework successfully generated over 100 theoretically viable candidate materials, with 20% exhibiting superior barrier properties and controlled degradation characteristics. Our computational approach demonstrated a 20–100× acceleration in screening efficiency compared to traditional DFT calculations while maintaining high accuracy. This work presents a significant advancement in computational materials discovery for sustainable packaging applications, offering a promising pathway to address the urgent global challenges of food waste and plastic pollution.

The present paper provides a novel hybrid computational framework that integrates Computational Fluid Dynamics (CFD) with advanced machine learning techniques to optimize solar thermal collectors employing micro-heat pipe arrays (MHPA) for food dehydration applications. The methodology addresses the fundamental challenge of balancing computational efficiency with prediction accuracy in thermal system design. A validated CFD model generated 935 numerical cases across diverse operational and design parameters, which were used to train and evaluate three machine learning algorithms: linear regression (LR), support vector regression (SVR), and artificial neural networks (ANN). While baseline LR achieved R² = 0.61, optimized SVR and ANN models demonstrated superior performance with R² values of 0.96 and 0.94 respectively. The study identifies a critical transition at 100–300 samples where error rates drop sharply, with optimal performance requiring more than 600 samples. Entropy analysis quantified information transfer between input parameters and thermal efficiency, identifying MHPA thermal conductivity as the most influential parameter (~ 20% mutual information), followed by air inlet temperature (~ 17%) and air velocity (~ 14%). This information-theoretic approach provided clear design priorities by measuring entropy reduction potential of each parameter. Interpretability analysis established optimal operating ranges for key parameters including MHPA equivalent thermal conductivity, power density, glass cover heat transfer coefficient, and air inlet temperature. The hybrid methodology shows promise for efficiently optimizing solar thermal collector designs at lower computational costs than traditional methods to provide valuable insights for solar food drying systems.