I am a PhD researcher in Applied Mathematics at the University of Birmingham. My research focuses on temporal representation theory and methods, with applications to non-stationary time-series learning, foundation models, and world models.

I study how representations of temporal structure support comparison, retrieval, forecasting, and classification. My current scope includes time series, action histories, VLA trajectories, model rollouts, and temporal memory.

• Decomposition: temporal component recovery, mechanism-driven evaluation, and signal feature extraction.
• Similarity and retrieval: structural comparison and stationarity-aware retrieval for forecasting.
• Prediction and regression: output dependencies, online learning, hierarchical forecasting, and interpretable grouped regression.
• Current directions: time-series classification, symbolic representation and compression, and temporal tokenization for foundation and world models. TOMC Memory extends the representation and compression direction to conversational context and task memory.

My research centres on temporal representation. Three columns group work on decomposition, similarity and retrieval, and prediction and regression. Current directions appear below, with TOMC Memory inside symbolic representation and compression.

• Decomposition: the ICML 2026 component-recovery benchmark and the MIIR 2025 report on battery ultrasound signals. DeTime and TSDecompose-Benchmark are research software outputs.
• Similarity and retrieval: SARAF (KDD 2026) uses stationarity-aware historical retrieval for forecasting. EchoTime provides explainable structural comparison.
• Prediction and regression: dynamic regressor chains for taxi demand (IJCNN 2020), online kNN regressor chains (ICONIP 2023), and adaptive COVID-19 prediction (Heliyon 2023) address output dependencies and online learning. Hierarchical load forecasting (ICONIP 2022) and iTARGET (BIBM 2024) develop interpretable prediction. iTARGET estimates age from methylation profiles.

Current work includes time-series classification, symbolic encoding and compression, and temporal tokenization for foundation and world models. The map also labels Semantics-Enhanced Retrieval-Augmented Time Series Forecasting as accepted ICML 2026 workshop work within Similarity and Retrieval.

TOMC Memory belongs to the Symbolic Representation & Compression direction. It prepares compact context for AI assistants by computing task-oriented records and combining them with selected source text. Its software also saves and recalls explicit task notes for reuse across assistants. TOMC Memory is research software; the GitHub repository provides the source code, plugin downloads and installation guides.

PAA, SAX, and SAA-SAX are representative literature methods for the symbolic/compression direction. Paper summaries and connections explain how the individual works fit the programme.

Related work · Editable PowerPoint

• UKRI/AIRR compute resources (2026): two Gateway Project allocations on Isambard-AI: Time Series Language and Foundation Model and Language-Action Time-Series Tokenization for Efficient VLA Policies. Each project provides 10,000 GPUHR with nominal compute-resource value GBP 45,000; together they total 20,000 GPUHR with nominal compute-resource value GBP 90,000. Compute resources only, not direct cash funding.
• University of Birmingham College PhD Scholarship: full scholarship support for Applied Mathematics PhD study from Jan 2024 to Jul 2027, covering tuition fees and stipend/living costs.
• ICML 2026: first-author main-conference paper, Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark.
• KDD 2026: co-author on Stationarity-Aware Retrieval-Augmented Time Series Forecasting.

• ICML 2026 | CORE/ICORE A*: Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark. First author.
• KDD 2026 | CORE/ICORE A*: Stationarity-Aware Retrieval-Augmented Time Series Forecasting. Co-author.
• MIIR 2025 | Technical report: Non-contact ultrasound testing of batteries. Co-author. Institutional record.
• BIBM 2024 | CORE/ICORE B: iTARGET: Interpretable Tailored Age Regression for Grouped Epigenetic Traits. First author.
• Heliyon 2023 | JCR Q1: A novel online multi-task learning for COVID-19 multi-output spatio-temporal prediction. First author.
• ICONIP 2023 | Oral presentation | CORE/ICORE B: Correlated Online k-Nearest Neighbors Regressor Chain for Online Multi-output Regression. First author.
• ICONIP 2022 | Oral presentation | CORE/ICORE B: An Interpretable Multi-target Regression Method for Hierarchical Load Forecasting. First author.
• IJCNN 2020 | Oral presentation | CORE/ICORE B: A Novel Dynamically Adjusted Regressor Chain for Taxi Demand Prediction. First author.

NeurIPS 2026 Institution Atlas — Explore institutional paper participation through global top-200 treemaps and ten country editions. Search full institution names and download SVG charts or CSV data. An unofficial analysis of a 26 September 2026 snapshot; counting rules and affiliation limitations are documented in the repository.

Open the interactive atlas · UK edition

Full publication list and CV: academic homepage / publications / CV.