Interview with Thi Kieu Khanh Ho: Time-series anomaly detection

2026-07-09 · Source: ΑΙhub · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, medium

Summary

Thi Kieu Khanh Ho's PhD research at McGill University and Mila – Québec AI Institute, supervised by Professor Narges Armanfard, focuses on time-series anomaly detection. Her work aims to enable AI systems to identify unusual events in complex, real-world data streams without relying on large amounts of labeled examples. Key contributions include EEG-CGS, a self-supervised framework incorporating local graph structures for anomalous channel detection in EEG recordings, effective for seizure detection without labeled seizures. She also developed TSAD-C, a graph- and diffusion-based framework addressing multivariate time-series anomaly detection with contaminated training data. Additionally, Ho contributed two surveys on self-supervised and graph-based time-series anomaly detection. Her recent work involves applying foundation models to physiological signals for identifying the epileptogenic zone in drug-resistant epilepsy patients, emphasizing trustworthiness for clinical decisions.

Key takeaway

For AI Scientists developing anomaly detection systems for safety-critical applications like clinical diagnostics, you should prioritize methods that effectively handle scarce, noisy, or contaminated training data. Focus on building models that demonstrate trustworthiness, explainability, and robustness, rather than solely optimizing for benchmark performance. This approach is crucial for bridging the significant gap between research results and genuinely useful, deployable clinical tools that inform critical decisions.

Key insights

Building AI models for reliable anomaly detection requires addressing scarce, noisy, or contaminated labeled data.

Principles

• Self-supervised learning enables anomaly detection without labeled examples.
• Real-world training data often contains anomalies, necessitating robust handling.
• Trustworthiness, explainability, and robustness are paramount for clinical AI deployment.

Method

Develop contrastive and generative self-supervised frameworks that integrate local graph structures for time-series anomaly detection.

In practice

• Use self-supervised methods for EEG seizure detection without labeled data.
• Implement graph- and diffusion-based models for contaminated time-series data.
• Prioritize model explainability for informing surgical decisions.

Topics

• Time-series Anomaly Detection
• Self-supervised Learning
• Graph-based Methods
• Foundation Models
• Clinical AI
• Explainable AI

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• I Predicted the T20 World Cup 2026 Final With an Anomaly Detection Pipeline. — Anomaly detection system failures reveal more about production challenges than successes, especially regarding evidence thresholds and cold-start problems.
• Fast and Accurate Anomaly Detection in Time Series — A novel unsupervised algorithm for time series anomaly detection, combining Haar wavelets and a $t$-test, significantly outperforms existing benchmarks.
• Data Machina #261 — Generative AI and LLMs are being adapted to achieve state-of-the-art performance in diverse time-series tasks.
• Out of Context: Reliability in Multimodal Anomaly Detection Requires Contextual Inference — Context-dependent anomalies require conditional inference in multimodal anomaly detection, separating context from observations.
• Unsupervised Anomaly Detection of Information Operations Users via Behavioral and Language Patterns — TENSOR uses multimodal data and LLM-adjusted Temporal Point Processes for unsupervised IO user anomaly detection.
• ProtoX-AD: Self-Explainable Time Series Anomaly Detection and Characterization — ProtoX-AD integrates interpretable prototypes into self-supervised time series anomaly detection for explainable results and anomaly characterization.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ΑΙhub.