Xing Wu

Multimodal anomaly detection benefits from complementary RGB and 3D evidence, yet auxiliary RGB reconstruction is not equally reliable across product categories and class-wise test-time policy selection is usually unavailable. We propose TC-MAF, a base-anchored multi-evidence fusion design that combines a multimodal detector, complementary Dinomaly evidence, and a small cross-modal consistency cue under one fixed pixel-level fusion formula. A lightweight training-dispersion confidence (TDC) term scales auxiliary participation using only normal training statistics. On MVTec-3D, TC-MAF reaches 0.979 image-level AUROC and 0.990 pixel-level AUPRO, achieving the best mean results on both detection and localization among the compared multimodal methods. Systematic ablations show that the fusion structure itself is the dominant factor, while TDC provides a smaller but reproducible calibration gain over no calibration or arbitrary calibration. Additional experiments show that the same design remains effective under a pooled-statistics variant, auxiliary-branch and backbone substitutions, few-shot settings, a missing-3D setting, and cross-dataset evaluation on Eyecandies. Code is available at https://github.com/Chranos/TC_MAF.

Existing multi-agent systems for automated writing assessment (AWA) suffer from three structural limitations: uniform agent weighting, fixed-depth computation, and lack of empirical auditing. To resolve these challenges, we propose CAMA, a Confidence-Aware Multi-Agent architecture consisting of three synergistic layers: Confidence-Weighted Aggregation (CWA), Adaptive Depth Routing (ADR), and Contribution-Guided Auditing (CGA). We evaluate its human alignment on 31 EFL essays and extend structural robustness checks to 151 samples across multiple datasets and 8 writing prompts. CAMA addresses the accuracy–efficiency–transparency trilemma. The experimental results show the following: (1) CWA provides a formal error containment bound in simulated agent failure, and multi-agent role decomposition achieves substantial improvement over a holistic baseline (QWK = +0.246); (2) ADR optimizes efficiency by routing 48.4% of standard essays to a low-cost Fast path while scaling up for difficult tasks; and (3) CGA provides an interpretable audit mechanism to quantify per-agent contributions. This framework offers a promising, transparent and principled paradigm for intelligent educational evaluation, with the caveat that broader validation across larger, more diverse populations remains a critical avenue for future work.

The escalating antimicrobial resistance (AMR) crisis necessitates the development of innovative anti-infectives with novel mechanisms of action. Nevertheless, research on natural products remains constrained by low-throughput screening and limited mechanistic insights. Artificial intelligence (AI) is catalyzing a pivotal paradigm shift—from the mere isolation of active compounds to precisely deciphering their modes of action. This review highlights AI’s transformative role in bridging ethnopharmacological knowledge and modern pharmacology to decode the mechanisms of plant-derived anti-infectives. Case studies on berberine, baicalein, danshensu derivatives, and rosmarinic acid derivatives from Coleus amboinicus illustrate AI’s capacity to map traditional therapeutic concepts to specific pathways (e.g., biofilm inhibition, inflammasome modulation) and to predict precise binding interactions and pharmacophores with high precision. Leveraging statistical correlations between ethnobotanical usage patterns and chemical similarity, we propose a “Knowledge–Data–Mechanism” three-layer framework centered on deep mechanistic insight. Integrating Chinese initiatives, such as the CNDR (China’s National Drug Repository) database and the TCM-AI platform, with global traditional medicine wisdom, this strategy provides an actionable roadmap for modernizing anti-infective discovery. Validated applications of this paradigm have demonstrated order-of-magnitude acceleration in mechanistic characterization, rapidly yielding structurally novel agents with well-defined, target-specific actions—a critical advancement in addressing the urgent global threat of antimicrobial resistance.