Haoyan Luo

Haoyan Luo (罗皓严)

I am a MRes (Master of Research) student in Machine Learning at Imperial College London, supervised by Prof. Lucia Specia. I obtained my B.Eng. degree in Computer Science from the Chinese University of Hong Kong, Shenzhen , advised by Prof. Haizhou Li and Prof. Shuang Li.

My research interests mainly lie in machine learning and natural language processing, especially in:

Explainable and interpretable machine learning models: develop understanding of large-scale language processing systems.
Applied machine learning: build fast and automate machine learning systems with a focus on robustness and improved performance.

Email / Github / Linkedin

Work Experience

Shanghai AI Lab	Research Intern	(July, 2023 - Jan, 2024)
Micosoft Research Asia	Research Intern	(July, 2022 - Feb, 2023)
Tencent Healthcare	Big Data Engineer Intern	(June, 2021 - Sep, 2021)
Seasun Entertainment	Software Engineer Intern	(June, 2020 - Sep, 2020)

Publications

	Depth-Adaptive Logit Refinement for Fine-tuning Large Language Models pdf
	Inference-Time Alignment of Large Language Models Haoyan Luo, Lucia Specia pdf
	From Understanding to Utilization: A Survey on Explainability for Large Language Models Haoyan Luo, Lucia Specia arXiv pdf
	OpenFE: Automated Feature Generation With Expert-level Performance Tianping Zhang, Zheyu Zhang, Haoyan Luo, Fengyuan Liu, Wei Cao, Jian Li International Conference on Machine Learning (ICML' 23) pdf / code / bibtex
	ActiveAD: Enhancing Anomaly Detection in Tabular Data through Active Learning Strategies Haoyan Luo, Xiaofan Gui, Wei Cao, Jiang Bian pdf

Selected Projects

	Dissecting token expression in language models Work in progress Probe and visualize the token representation in LLMs. Dissecting the knowledge and relation propogation in the next-token generating process with mechanistic interpretability and decomposition theory.
	Singer-adaptive Singing Voice Conversion pdf / code / demo Finding the probable weakness of the existing SVC systems in converting to professional singing voice and building a model with generalizability across various target vocalists while preserving high- quality voice conversion
	Steering the Networked Temporal Point Processes via Controlling the Network Graph code Developed a model-based reinforcement learning strategy for optimizing network dynamics, using neural ODEs to model multivariate temporal point processes. Created a novel approach that manipulates graph topology for effective intervention in networks, such as mitigating epidemics or easing traffic congestion.
	Deep-Learning-Lookup: Deep Learning Algorithms and Utils in Python code Implemented various deep learning code examples (notebooks and python scripts) and useful utility functions including data manipulation, visualization, training, and evaluation.
	Mindy: a web application for creating and managing online document report / code Mindy is a web application for creating and managing online document. Mindy (meaning ”map your mind”), aims to provide a platform for users to map their mind online instead of on a piece of paper, which has unlimited space, be easy to digitize, and make it searchable.

Misc.

During my undergraduate studies, I was a student coordinator in Shaw college and a host in the Communication and Public Relations Office. I was also the president of TEDxCUHKSZ (check our annual events!) and spent a lot of effort promoting our slogan: "ideas worth spreading" :p

During my free time, I am a huge piano and cello lover. I also enjoy playing football, basketball and tennis. Go Warriors and KTBFFH💙

Website Credits to Jon Barron source code and Siddhartha Gairola source code