# LLM-Generated Feature Pools for Time Series Anomaly Detection

arXiv: https://arxiv.org/abs/2609.21801 · HTML: https://mlsanity.com/paper/2609.21801/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 24 references that are in the map.
- First posted on arXiv 18 Sep 2026.

## Builds on (most cited first, 10 of 10)

- [Chronos: Learning the Language of Time Series](https://mlsanity.com/paper/2403.07815/) (2024)
- [Time-LLM: Time Series Forecasting by Reprogramming Large Language Models](https://mlsanity.com/paper/2310.01728/) (2023)
- [Large Language Models Are Zero-Shot Time Series Forecasters](https://mlsanity.com/paper/2310.07820/) (2023)
- [MOMENT: A Family of Open Time-series Foundation Models](https://mlsanity.com/paper/2402.03885/) (2024)
- [TSPulse: Tiny Pre-Trained Models with Disentangled Representations for Rapid Time-Series Analysis](https://mlsanity.com/paper/2505.13033/) (2025)
- [Towards Foundation Models for Zero-Shot Time Series Anomaly Detection: Leveraging Synthetic Data and Relative Context Discrepancy](https://mlsanity.com/paper/2509.21190/) (2025)
- [STREAM-VAE: Dual-Path Routing for Slow and Fast Dynamics in Vehicle Telemetry Anomaly Detection](https://mlsanity.com/paper/2511.15339/) (2025)
- [Time to Embed: Unlocking Foundation Models for Time Series with Channel Descriptions](https://mlsanity.com/paper/2505.14543/) (2025)
- [Tree-Structured Synergy of Large Language Models and Bayesian Optimization for Efficient CASH](https://mlsanity.com/paper/2601.12355/) (2026)
- [Matrix Profile for Time-Series Anomaly Detection: A Reproducible Open-Source Benchmark on TSB-AD](https://mlsanity.com/paper/2604.02445/) (2026)

## Abstract (arXiv)

We study how far a simple statistical pipeline can go on univariate time series anomaly detection under a strict selection protocol. The method extracts a small pool of statistics over sliding windows, scores each window with a transductive robust (MAD) model, and selects a feature subset per domain on a held-out tuning split. On TSB-AD-U it reaches 0.529 per-series VUS-PR, above the best neural (0.45) and statistical (0.44) entries on the public leaderboard and within 0.06 of the strongest pretrained foundation model, several of which use more supervision than ours. Ablations locate the cause: across three selection strategies and a hindsight oracle the score moves by 0.031, and across the aggregation grid by 0.096, while changing the candidate pool moves it by 0.226. The candidate pool sets the ceiling; the search over it is second-order. We therefore generate a pool per domain by prompting a multimodal LLM with in-context example windows from that domain. The generated pools match the hand-crafted one under matched selection, and the two cover different domains: selecting over their union improves on the generated pool in all twelve generator-seed pairs and lifts the pipeline to 0.588, matching the performance of the best entry on the leaderboard.