Time-RA reformulates time series anomaly detection as a reasoning-intensive generative task and provides the RATs40K multimodal benchmark to evaluate and improve LLM-based diagnosis.
hub
Anomaly transformer: Time series anomaly detection with association discrepancy
17 Pith papers cite this work. Polarity classification is still indexing.
17
Pith papers citing it
abstract
Unsupervised detection of anomaly points in time series is a challenging problem, which requires the model to derive a distinguishable criterion. Previous methods tackle the problem mainly through learning pointwise representation or pairwise association, however, neither is sufficient to reason about the intricate dynamics. Recently, Transformers have shown great power in unified modeling of pointwise representation and pairwise association, and we find that the self-attention weight distribution of each time point can embody rich association with the whole series. Our key observation is that due to the rarity of anomalies, it is extremely difficult to build nontrivial associations from abnormal points to the whole series, thereby, the anomalies' associations shall mainly concentrate on their adjacent time points. This adjacent-concentration bias implies an association-based criterion inherently distinguishable between normal and abnormal points, which we highlight through the \emph{Association Discrepancy}. Technically, we propose the \emph{Anomaly Transformer} with a new \emph{Anomaly-Attention} mechanism to compute the association discrepancy. A minimax strategy is devised to amplify the normal-abnormal distinguishability of the association discrepancy. The Anomaly Transformer achieves state-of-the-art results on six unsupervised time series anomaly detection benchmarks of three applications: service monitoring, space & earth exploration, and water treatment.
hub tools
citation-role summary
baseline 1
citation-polarity summary
roles
baseline 1polarities
baseline 1representative citing papers
TS-Reasoner is a domain-oriented agent using LLMs, computational tools, and error feedback for multi-step time series inference, showing better performance than general LLMs on understanding and reasoning benchmarks.
A physics-grounded dilution-fridge simulator with LLM agents achieves supervised-ML parity on cryogenic fault classification using six demonstrations and no training, validated on simulated telemetry plus a real-hardware false-alarm check.
A novel unsupervised anomaly detection method for time series using Haar wavelets and a designed t-test outperforms state-of-the-art benchmarks across 343 datasets.
A transformer model guided by a causal graph prior achieves state-of-the-art anomaly detection and root-cause attribution on ASD and SMD benchmarks by restricting main predictions to graph-supported causes while using an isolated shadow path for residual correlations.
DBR-AF decouples cross-variable correlations in reconstruction and applies autoregressive flows to model residual densities for improved anomaly detection in multivariate time series.
PG-TMT couples a physics-aligned tri-branch encoder with EVT-calibrated decision rules to achieve higher PR-AUC and shorter detection times at controlled false-alarm rates across multiple bearing datasets.
Neural CDEs serve as correctors that reduce error accumulation in multi-step forecasts from learned time-series models across synthetic, physics, and real-world data.
Zeus proposes a multi-scale Transformer with point-wise tokenization and Multi-Objective Temporal Masking to enable tuning-free performance on forecasting, interpolation, and other time series tasks.
Overlapping inference windows improve reconstruction-based time series anomaly detection by up to 28% relative gain across models on TSB-AD and UCR benchmarks and can alter rankings.
U²AD learns unified normal data representations via score-based generative modeling and a novel time-dependent score network to outperform prior methods in accuracy and early anomaly detection for multivariate time series.
The paper presents architecture variants for observers and controllers in self-organizing cyber-physical energy systems that account for information and control constraints.
A conditional attribution framework retrieves contextually similar normal states from learned VAE/UMAP embeddings to produce dependency-preserving root cause explanations for time-series anomalies.
Active learning with masked reconstruction and minimax training raises AUC by 12.39% across 28 test cases on four multivariate datasets and seven unsupervised backbones.
Introduces a cyclic-dynamics dataset for industrial MTSAD and benchmarks federated anomaly detection methods on it and a public dataset.
Fourier-KAN-Mamba combines Fourier features, KAN nonlinearities, and Mamba state-space modeling with a gating mechanism and reports better anomaly detection performance than prior methods on the MSL, SMAP, and SWaT benchmarks.
citing papers explorer
-
Time-RA: Towards Time Series Reasoning for Anomaly Diagnosis with LLM Feedback
Time-RA reformulates time series anomaly detection as a reasoning-intensive generative task and provides the RATs40K multimodal benchmark to evaluate and improve LLM-based diagnosis.
-
TS-Reasoner: Domain-Oriented Time Series Inference Agents for Reasoning and Automated Analysis
TS-Reasoner is a domain-oriented agent using LLMs, computational tools, and error feedback for multi-step time series inference, showing better performance than general LLMs on understanding and reasoning benchmarks.
-
Onnes: A Physics-Grounded Multi-Agent LLM Simulator for Cryogenic Fault Diagnosis in Quantum Computing Infrastructure
A physics-grounded dilution-fridge simulator with LLM agents achieves supervised-ML parity on cryogenic fault classification using six demonstrations and no training, validated on simulated telemetry plus a real-hardware false-alarm check.
-
Fast and Accurate Anomaly Detection in Time Series
A novel unsupervised anomaly detection method for time series using Haar wavelets and a designed t-test outperforms state-of-the-art benchmarks across 343 datasets.
-
Causally-Constrained Probabilistic Forecasting for Time-Series Anomaly Detection
A transformer model guided by a causal graph prior achieves state-of-the-art anomaly detection and root-cause attribution on ASD and SMD benchmarks by restricting main predictions to graph-supported causes while using an isolated shadow path for residual correlations.
-
Multivariate Time Series Anomaly Detection via Dual-Branch Reconstruction and Autoregressive Flow-based Residual Density Estimation
DBR-AF decouples cross-variable correlations in reconstruction and applies autoregressive flows to model residual densities for improved anomaly detection in multivariate time series.
-
Physics-Guided Tiny-Mamba Transformer for Reliability-Aware Early Fault Warning
PG-TMT couples a physics-aligned tri-branch encoder with EVT-calibrated decision rules to achieve higher PR-AUC and shorter detection times at controlled false-alarm rates across multiple bearing datasets.
-
Neural CDEs as Correctors for Learned Time Series Models
Neural CDEs serve as correctors that reduce error accumulation in multi-step forecasts from learned time-series models across synthetic, physics, and real-world data.
-
Zeus: Towards Tuning-Free Foundation Model for Time Series Analysis
Zeus proposes a multi-scale Transformer with point-wise tokenization and Multi-Objective Temporal Masking to enable tuning-free performance on forecasting, interpolation, and other time series tasks.
-
Disjoint or Overlapping? Inference Windowing for Reconstruction-Based Time Series Anomaly Detection
Overlapping inference windows improve reconstruction-based time series anomaly detection by up to 28% relative gain across models on TSB-AD and UCR benchmarks and can alter rankings.
-
Learning Unified Representations of Normalcy for Time Series Anomaly Detection
U²AD learns unified normal data representations via score-based generative modeling and a novel time-dependent score network to outperform prior methods in accuracy and early anomaly detection for multivariate time series.
-
Architectures for Robust Self-Organizing Energy Systems under Information and Control Constraints
The paper presents architecture variants for observers and controllers in self-organizing cyber-physical energy systems that account for information and control constraints.
-
Conditional Attribution for Root Cause Analysis in Time-Series Anomaly Detection
A conditional attribution framework retrieves contextually similar normal states from learned VAE/UMAP embeddings to produce dependency-preserving root cause explanations for time-series anomalies.
-
Detecting the Undetectable: Enhancing Unsupervised time series Anomaly Detection via Active Learning
Active learning with masked reconstruction and minimax training raises AUC by 12.39% across 28 test cases on four multivariate datasets and seven unsupervised backbones.
-
Federated Learning for Multivariate Time Series Anomaly Detection in Industrial Automation
Introduces a cyclic-dynamics dataset for industrial MTSAD and benchmarks federated anomaly detection methods on it and a public dataset.
-
Fourier-KAN-Mamba: A Novel State-Space Equation Approach for Time-Series Anomaly Detection
Fourier-KAN-Mamba combines Fourier features, KAN nonlinearities, and Mamba state-space modeling with a gating mechanism and reports better anomaly detection performance than prior methods on the MSL, SMAP, and SWaT benchmarks.
- Modeling Spectral Energy Shifts in Spatio-Temporal Graph Anomaly Detection