HELLO, I'M SHUMAN 

About Me

Hello! 👋 My name is Shuman. I am a Ph.D. student at Simon Fraser University.  Aside from research, I enjoy exploring new cities, discovering new coffee beans and cafés, reading, and taking photos with the 3x zoom on my phone. I also enjoy snowboarding on the local mountains whenever I get the chance.

Research

I am part of the Ester Lab, and my supervisor is Prof. Martin Ester. I was also a visiting PhD student at the Scalable Trustworthy AI (STAI) Lab at KAIST, hosted by Prof. Seong Joon Oh. I continue to work closely with Prof. Seong Joon Oh and the STAI Lab on my research. My research focuses on multimodal and visual understanding/reasoning, compositional generalization, and out-of-distribution (OOD) generalization. I have also worked on self-supervised representation learning and uncertainty quantification.

Publications & Manuscripts

* equal contribution (joint first author)

BindSplit: Benchmarking Seen and Unseen Compositional Generalization in VLMs
Shuman Peng, Arnas Uselis, Darina Koishigarina, Martin Ester, and Seong Joon Oh
Under review.

CLIP Models Generalize Less Than Compositional Benchmarks Suggest [Paper][Project page]
Shuman Peng, Arnas Uselis, Darina Koishigarina, Martin Ester, and Seong Joon Oh
Combining Theory and Benchmarks (CTB) Workshop at ICML 2026.
Compositional Learning Workshop at ICML 2026.

Informed Augmentation Selection Improves Tabular Contrastive Learning [Paper]

Arash Khoeini*, Shuman Peng*, and Martin Ester

Self-supervised Learning (SSL) Workshop at NeurIPS 2024.

The 29th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2025).

Improving OOD Generalization of Pre-trained Encoders via Aligned Embedding-Space Ensembles [Paper]

Shuman Peng, Arash Khoeini, Sharan Vaswani, and Martin Ester

Unifying Representations in Neural Models (UniReps) Workshop at NeurIPS 2024. 

Self-supervised Learning (SSL) Workshop at NeurIPS 2024.

Better Calibration Error Estimation for Reliable Uncertainty Quantification [Paper]

Shuman Peng*, Parsa Alamzadeh*, and Martin Ester

3rd Workshop on Interpretable Machine Learning in Healthcare (IMLH) at ICML 2023. 

Combining Domain-Specific Meta-Learners in the Parameter Space for Cross-Domain Few-Shot Classification [Paper]

Shuman Peng, Weilian Song, and Martin Ester.

arXiv preprint 2020. 

AITL: Adversarial Inductive Transfer Learning with input and output space adaptation for pharmacogenomics [Paper]

Hossein Sharifi-Noghabi, Shuman Peng, Olga Zolotareva, Colin C. Collins, and Martin Ester.

Bioinformatics, Volume 36, Issue Supplement_1, Pages i380–i388. (Presented at ISMB 2020)