Search

Deep Learning Anomaly Detection | ByteJournal

Deep learning methods improve anomaly detection in financial time-series by capturing temporal patterns; common techniques include autoencoders, CNNs, and RNNs, each with respective strengths. While offering real-time detection and handling complex patterns, they face data and overfitting challenges, with structural entropy proposed as a mitigation approach and future work focusing on robustness and broader finance applications.

[2610.28560] Agentic Ensemble with Vision-Language Models for Visual Anomaly Detection

AnomalyAgent is a post-training fusion framework for visual anomaly detection that aligns scores and anomaly maps from multiple detectors, treats fusion as action selection with score and map actions, uses a constrained vision-language model to pick actions on hard cases, and achieves state-leading performance across multiple benchmarks.

Train, Evaluate, and Deploy Anomaly Detection Models with Roboflow

Roboflow now supports anomaly detection models for cloud or edge deployment. These models learn from normal, defect-free examples to flag unusual images with a heatmap, aiding QA/QC across industries; PatchCore and FoundAD offer different trade-offs, with speed and data requirements influenced by image resolution and memory usage.

Classification with decision boundaries, k-nearest neighbors, and decision trees – OER Informatik

Translated title. Original: Klassifizieren mit Trenngeraden, k-nächsten Nachbarn und Entscheidungsbäumen – OER Informatik

At k = 1, a single anomaly can change the decision, causing overfitting in the model.

PatchCore Anomaly Detection: Tutorial and FoundAD Comparison

The document explains PatchCore anomaly detection for PCB inspection, detailing how it learns from normal images, generates anomaly heatmaps, and compares PatchCore with FoundAD using a Roboflow workflow. It includes dataset preparation, labeling, and dataset versioning steps for a tutorial setup.

A Taiwanese startup speeds up AI-based inference, detecting anomalies with over 90% accuracy.

Translated title. Original: Una startup taiwanese accelera l'inferenza basata sull'intelligenza artificiale, rilevando anomalie con una precisione superiore al 90%.

Chimes AI’s Tukey platform combines machine learning and generative AI to detect industrial equipment anomalies and help teams assess causes and maintenance actions. After moving to NVIDIA H100 GPUs at FPT AI Factory, the company reported inference about 2.9 times faster in a 2,000-token test.

KI: Rossmann employees can query company data in natural language - Artificial Intelligence (AI) | News | LOGISTIK HEUTE - The German logistics magazine

Translated title. Original: KI: Rossmann-Mitarbeiter können Unternehmensdaten in natürlicher Sprache abfragen - Künstliche Intelligenz (KI, AI) | News | LOGISTIK HEUTE - Das deutsche Logistikmagazin

A logistics-tech solution, built on operational data, delivers tour-status insights, flags anomalies, and triggers alerts. It is already in daily use and aims to scale 'Chat with your Data' across other business areas, emphasizing solid data foundations and rapid query responses.

GitHub - Minxuan-Hu/streaming-integrity-alerting: Reproducible evaluation of streaming anomaly detection and alerting under operational workload constraints. · GitHub

This repository provides code and archived outputs to reproduce a study on streaming integrity alerting under workload constraints, detailing workflows to regenerate manuscript results and to rerun experiments, with data-panel construction from CFPB and NASA SMAP datasets and a full reproducibility checklist.

AI in production at Bolt

Three Bolt engineers discuss lessons from using AI in production, including scam detection, unreliable generated data, and coding agents.

Chance or risk: Collaboration of German companies with Big Tech | tagesschau.de

Translated title. Original: Chance oder Risiko: Zusammenarbeit deutscher Unternehmen mit Big Tech | tagesschau.de

Bosch collaborates with Microsoft to advance AI across production, testing an agent in Feuerbach that automates risk detection and recommendations, reducing manual workload. While boosting productivity, experts warn of increased dependence on US tech and regulatory hurdles in Europe that could affect competitiveness.