Jiahao (Kevin) Cheng

Research

Computational pathology,
from images to structured language.

I study machine-learning methods for pathology images and related clinical information. My current interests include vision-language models, deep learning, representation learning, and computer vision.

I am a member of Atlas Analytics Lab and Mila.

Research themes

What I work on

01

Computational pathology

I apply machine learning to questions and workflows involving pathology images and related clinical information.

02

Visual representation learning

I am interested in how deep learning and computer vision can form useful representations of complex, gigapixel whole-slide images.

03

Vision-language models

I explore ways to connect slide-level visual evidence with structured clinical language and case-level reasoning.

Further exploration

More on computational pathology

Publications

Peer-reviewed work

Conference paper2026

Gated Spatial Redundancy Projection for Pathology Transformer Attentions

Zhiyuan Yang, Jiahao Cheng, Vincent Quoc-Huy Trinh, Mahdi S. Hosseini

British Machine Vision Conference (BMVC 2026), Oral Presentation

GatedSRP is a lightweight, drop-in correction for pathology transformer self-attention. For each patch token and attention head, it estimates a local redundancy direction from neighboring value vectors and learns a signed gate that decides how to correct the aligned component, adding about 0.02% trainable parameters.

Conference paper2026

Simple Token-Efficient Vision-Language Model for Case-level Pathology Synoptic Report Generation

Zhiyuan Yang, Jiahao Cheng, Vincent Quoc-Huy Trinh, Mahdi S. Hosseini

Zhiyuan Yang and Jiahao Cheng contributed equally.

Proceedings of the 7th International Conference on Deep Learning Theory and Applications (DeLTA 2026), pp. 514–537

PathoSynVLM is a token-efficient vision-language model for generating structured, case-level pathology reports from one or more whole-slide images. The work studies a frozen-encoder, lightweight-aligner, language-model design and the trade-off between visual detail, token count, memory, and report quality.