Jiayu Qin

PhD Student in Computer Science and Engineering

Specializing in deep learning, large language models, and multimodal representation learning

Jiayu Qin profile picture

Professional Summary

Motivated Ph.D. student in Computer Science and Engineering specializing in deep learning, large language models, and multimodal representation learning. Experienced in designing and scaling machine learning pipelines on large, noisy datasets for financial prediction and biomedical discovery.

Education

University at Buffalo, SUNY

PhD student in Computer Science and Engineering

08/2022 - Now

Fudan University

Bachelor of Engineering, majoring in Electrical Engineering

Second-class Scholarship (Top 5%) at graduation

09/2018 - 06/2022

Work Experience

Harvard University

02/2025-05/2025

Visiting Researcher

MBZUAI

08/2024-01/2025

Visiting Researcher

GALASPORTS CO.

06/2023-08/2023

Machine Learning Engineer Summer Intern

University at Buffalo

09/2022-Now

Ph.D Teaching/Research Assistant

Research Experience

LLM empowered Financial News Topic Extraction & Market Prediction

Ongoing Work

  • Developed LLM-driven pipeline to extract financial themes from 40+ years of news data
  • Created attention scores that improved stock market prediction (15.2% annual return)

A Probability Contrastive Learning Framework for Graph Representation Learning

NeurIPS 2024

  • Solved False Pairs problem in graph contrastive learning via Bayesian modeling
  • Achieved state-of-the-art results on MoleculeNet and QM9 benchmarks

A Unified Biomedical Knowledge Model for Molecule-Protein Interaction Prediction

Ongoing Work

  • Integrated multi-modal biological data using optimal transport
  • Improved accuracy and zero-shot generalization for interaction prediction

Other Works

Learning Unnormalized Statistical Models via Compositional Optimization (MECO)

ICML 2023 poster

KidSpeak: A General Multi-Purpose LLM for Kids' Speech Recognition and Screening

Arxiv

SE-3 Equivariant Mamba for Molecular Representation Learning

Arxiv

Prompt tuning based adapter for vision-language model adaptation

Ongoing Work