{"id":1398,"date":"2025-10-06T20:27:20","date_gmt":"2025-10-06T20:27:20","guid":{"rendered":"https:\/\/scipapermill.com\/index.php\/2025\/10\/06\/anomaly-detection-unleashed-from-zero-shot-vision-to-explainable-time-series-and-resilient-networks\/"},"modified":"2025-12-28T21:59:37","modified_gmt":"2025-12-28T21:59:37","slug":"anomaly-detection-unleashed-from-zero-shot-vision-to-explainable-time-series-and-resilient-networks","status":"publish","type":"post","link":"https:\/\/scipapermill.com\/index.php\/2025\/10\/06\/anomaly-detection-unleashed-from-zero-shot-vision-to-explainable-time-series-and-resilient-networks\/","title":{"rendered":"Anomaly Detection Unleashed: From Zero-Shot Vision to Explainable Time Series and Resilient Networks"},"content":{"rendered":"

Latest 50 papers on anomaly detection: Oct. 6, 2025<\/h3>\n

Anomaly detection is the unsung hero of AI\/ML, diligently sifting through oceans of data to spot the subtle deviations that signal critical issues\u2014be it a cyberattack, a faulty machine, or a health concern. It\u2019s a field constantly evolving, driven by the need for more robust, efficient, and interpretable methods in increasingly complex domains. Recent breakthroughs, as highlighted by a compelling collection of new research, are pushing the boundaries of what\u2019s possible, from training-free approaches to models that inherently understand \u2018normal\u2019 from \u2018abnormal\u2019.<\/p>\n

The Big Idea(s) & Core Innovations<\/h3>\n

One of the most exciting trends is the move towards generalist and zero-shot anomaly detection<\/strong>, drastically reducing the need for extensive labeled data. Papers like \u201cA Single Image Is All You Need: Zero-Shot Anomaly Localization Without Training Data<\/a>\u201d by Mehrdad Moradi et al.\u00a0from Georgia Tech and Arizona State University introduces SSDnet, which performs robust anomaly localization on a single image without any prior training, leveraging convolutional neural networks\u2019 inductive bias and perceptual losses. This is echoed in industrial settings by the \u201cPatchEAD: Unifying Industrial Visual Prompting Frameworks for Patch-Exclusive Anomaly Detection<\/a>\u201d framework by Po-Han Huang et al.\u00a0from Inventec Corporation, providing training-free, patch-level anomaly detection crucial for quality control. This ability for models to implicitly understand \u2018normal\u2019 is further explored by Chun-Liang Li et al.\u00a0(MIT, Google Research, Stanford University) in \u201cFoundation Visual Encoders Are Secretly Few-Shot Anomaly Detectors<\/a>\u201d, revealing that foundation visual encoders inherently detect anomalies by leveraging the natural image manifold.<\/p>\n

Driving this generalization is the innovative use of Large Language Models (LLMs) and Vision-Language Models (VLMs). \u201cPANDA: Towards Generalist Video Anomaly Detection via Agentic AI Engineer<\/a>\u201d by Zhiwei Yang et al.\u00a0from Xidian University introduces an agentic AI engineer that enables generalist video anomaly detection without training data or manual involvement, leveraging self-adaptive strategy planning and self-reflection. Similarly, Shu Zou et al.\u00a0(Australian National University) in \u201cUnlocking Vision-Language Models for Video Anomaly Detection via Fine-Grained Prompting<\/a>\u201d presents ASK-HINT, a framework that enhances VAD by using fine-grained, action-centric prompts with frozen VLMs. For time series, \u201cAXIS: Explainable Time Series Anomaly Detection with Large Language Models<\/a>\u201d by Tian Lan et al.\u00a0(Tsinghua University, Huawei) makes LLMs explainable anomaly detectors, providing intuitive, context-aware rationales.<\/p>\n

Beyond vision and language, breakthroughs are enhancing the robustness and efficiency of anomaly detection in critical infrastructure. In cybersecurity, Oluwakemi Adebayo (University of Technology, Nigeria) presents an \u201cAdaptive Cybersecurity Architecture for Digital Product Ecosystems Using Agentic AI<\/a>\u201d that dynamically detects and responds to threats. For networked systems, PUL-Inter-slice Defender, introduced by John Doe et al.\u00a0from University of Technology, in \u201cPUL-Inter-slice Defender: An Anomaly Detection Solution for Distributed Slice Mobility Attacks<\/a>\u201d uses machine learning to detect subtle distributed slice mobility attacks. Guolei Zeng et al.\u00a0(University of Oxford, Singapore Management University) addresses semi-supervised graph anomaly detection with \u201cNormality Calibration in Semi-supervised Graph Anomaly Detection<\/a>\u201d (GraphNC), calibrating normality learning to reduce false positives\/negatives. In time series, \u201cPi-Transformer: A Physics-informed Attention Mechanism for Time Series Anomaly Detection<\/a>\u201d by Sepehr Maleki et al.\u00a0(University of Lincoln, Trainline) incorporates physical priors to detect subtle timing and phase irregularities.<\/p>\n

Crucially, improved data representation and context utilization are central. \u201cReTabAD: A Benchmark for Restoring Semantic Context in Tabular Anomaly Detection<\/a>\u201d by Sanghyu Yoon et al.\u00a0(LG AI Research, Sungkyunkwan University) introduces the first context-aware tabular AD benchmark, demonstrating how textual metadata significantly boosts performance and interpretability. \u201cUniMMAD: Unified Multi-Modal and Multi-Class Anomaly Detection via MoE-Driven Feature Decompression<\/a>\u201d by Yuan Zhao et al.\u00a0(Dalian University of Technology, Nanyang Technological University) offers a unified framework for multi-modal, multi-class anomaly detection using a Mixture-of-Experts approach, enhancing efficiency while reducing domain interference. For time series, \u201cScatterAD: Temporal-Topological Scattering Mechanism for Time Series Anomaly Detection<\/a>\u201d by Tao Yin et al.\u00a0(Chongqing University, University of Oxford) highlights that anomalies exhibit stronger scattering patterns and leverages this for improved detection using temporal and topological features.<\/p>\n

Under the Hood: Models, Datasets, & Benchmarks<\/h3>\n

Recent advancements in anomaly detection are heavily reliant on novel models, curated datasets, and robust benchmarks. Here\u2019s a glimpse into the foundational resources powering these innovations:<\/p>\n