Zhiwen Chen

MSc Health Data Science @ UCL · BSc Computer Science @ Birmingham (2:1) · Targeting UK early-career roles in AI, ML, NLP and Health Data.

London, United Kingdom

About

I am finishing an MSc in Health Data Science at UCL after a BSc in Computer Science at Birmingham (2:1, Jun 2025). My work sits at the intersection of machine learning, natural language processing, and healthcare data — from ontology-aware clinical entity linking with SNOMED CT, to a novel three-phase evolutionary algorithm for multi-objective Neural Architecture Search.

I am looking for an internship, graduate, or junior role in the UK where I can apply ML and NLP to real problems — particularly in healthcare, digital health, and applied AI.

Featured Project

BSc Dissertation · Sep 2024 – May 2025

SENAS: A Cyclical Three-Phase Evolutionary Algorithm for Multi-Objective Neural Architecture Search

University of Birmingham · Supervisor Dr. Jizheng Wan · Inspector Prof. Russell Beale · Topic introduced by Dr. Miqing Li

SENAS is a cyclic three-phase evolutionary NAS algorithm: SOM-guided diversity exploration, dual-archive convergence–diversity co-evolution, and ε+-indicator-guided refinement. A lightweight Monte-Carlo hypervolume estimator handles phase transitions without computing exact HV.

Benchmarked against NSGA-II, MOEA/D, HypE, RVEA and a SOM-SEMO baseline on the C10/MOP suite (NAS-Bench-101 / 201, NATS-Bench, DARTS), 30 independent runs per task, fixed budget of 10,000 architecture queries.

Task	SENAS	NSGA-II	MOEA/D	HypE	RVEA
C10/MOP1	0.9368 ✓	0.9156	0.8965	0.8060	0.9158
C10/MOP2	0.9174 ✓	0.8987	0.8714	0.7714	0.8903
C10/MOP3	0.8260 ✓	0.8054	0.7883	0.7924	0.8013
C10/MOP4	0.7568 ✓	0.7452	0.7231	0.7483	0.7419

Result: SENAS wins on 6 of 9 C10/MOP benchmarks. Honest limitation: SENAS underperforms on the DARTS-based MOP8–9, where the continuous architecture space penalises discrete BMU-based perturbation.

Stack: Python · PyTorch · EvoXBench

Selected Work

Ontology-aware Clinical Entity Linking with SNOMED CT

UCL MSc research · Ongoing

Two-stage entity linking framework for clinical text: candidate generation followed by hierarchy-informed reranking.
Reranker combines semantic similarity with SNOMED CT structural signals — hierarchy distance, shared ancestors, structural consistency.
Evaluating disambiguation accuracy against a non-hierarchical baseline.

Heart Failure Prediction from Clinical Features

UCL MSc coursework · 2026

End-to-end ML workflow on structured clinical variables, with EDA, missingness review, and comparative feature visualisation.
Compared classifiers using ROC-AUC, precision, recall, and F1-score, prioritising clinically interpretable feature importance.

NLP Algorithm Intern — Shanghai Zhiyu Information Technology

Beijing, Jun 2025 – Sep 2025

Built an OCR + NLP pipeline for automated identification of lending-related clauses in scanned corporate charter documents.
Designed page-level pre-processing (text extraction, region localisation, document cleaning) and iterated on model + data refinement.

Skills

Languages

Python · R · SQL · Java · JavaScript · C · Haskell

ML / NLP

PyTorch · Hugging Face Transformers · PaddleNLP · Information Extraction · Multi-objective Optimisation · Neural Architecture Search

Data & Genomics

EDA · Statistical evaluation · Benchmarking · PLINK · GWAS · Variant QC

Tools

Git / GitHub · Linux / Unix · OCR tooling · LaTeX

Education

University College London — MSc Health Data Science Sep 2025 – Sep 2026 (expected)

AI in Healthcare · Advanced Methods in Data Science and Statistics · Principles of Health Data Science · Applied Computational Genomics · Epidemiology for Data Science · Programming with Python for Health Research.

University of Birmingham — BSc Computer Science · 2:1 Sep 2021 – Jun 2025

Artificial Intelligence · Natural Language Processing · Evolutionary Computation · Data Structures and Algorithms · Operating Systems · Software Engineering · Mathematics and Logic Foundations.