Tim G. Zhou

Tim (in wind);

Abbotsford, BC.

I am a first-year PhD student in Computer Science at UBC, co-advised by Dr. Geoff Pleiss and Dr. Evan Shelhamer. I’m broadly interested in application-driven deep learning research — particularly self-supervised learning, uncertainty quantification, ensemble methods, and most recently multi-modal networks for remote sensing.

Previously, I was a research assistant in UBC SPIN Lab (HCI & Robotics), where I gave fluffy robots heartbeats and made them breathe.

Outside of research, I’m an enthusiastic, albeit mediocre, bird watcher. I also perform with UBC Acappella and play trumpet in a jazz band.

UBC Computer Vision Reading Group (CVRG): I organize the weekly CV reading group, which meets every Monday at 2pm. To join the mailing list, email cvrg-l-request@cs.ubc.ca with the word join in the subject or body.

news

May 19, 2026	Passed my Research Proficiency Evaluation (RPE), I’m allegedly proficient now :)
May 12, 2026	I started as a Mitacs Accelerate research intern with RBC Borealis, working on Adapting Neural Networks to an Ever-Changing Earth.
Mar 07, 2026	Delulu-Net: Changing Modalities By Cross-Band Transfer, Addition, and Peeking was accepted to ICLR 2026 ML4RS Workshop.
Sep 18, 2025	Asymmetric Duos was accepted to Neurips 25 as a spotlight paper.
May 06, 2025	I received the Martin Frauendorf Memorial Prize in Computer Science, awarded annually to one graduating UBC CS student for outstanding leadership and service to the department ($900).

selected publications

NeurIPS 25
Asymmetric Duos: Sidekicks Improve Uncertainty

Tim G. Zhou^*, Evan Shelhamer, and Geoff Pleiss

In NeurIPS, 2025

Abs Bib HTML PDF

We propose pairing large deep neural networks with smaller sidekick models to improve uncertainty quantification in a computationally efficient manner. Rather than ensembling multiple training runs, we combine predictions via learned weighted averaging. Our method achieves improved accuracy and uncertainty metrics across five image classification benchmarks with only 10-20% more computation, while the smaller model rarely degrades performance.
@inproceedings{zhou2025asymmetricduos, title = {Asymmetric Duos: Sidekicks Improve Uncertainty}, author = {Zhou, Tim G. and Shelhamer, Evan and Pleiss, Geoff}, booktitle = {NeurIPS}, year = {2025}, url = {https://arxiv.org/abs/2505.18636}, }
IEEE TAC
Haptically Experienced Animacy Facilitates Emotion Regulation: A Theory-Driven Investigation

Preeti Vyas, Bereket Guta, Tim G. Zhou, Noor Naila Himam, Andero Uusberg, and Karon E. MacLean

In IEEE Transactions on Affective Computing, 2026

Abs Bib HTML PDF

We developed CHORA, a zoomorphic robot featuring biomimetic breathing and heartbeat behaviors. Through a mixed-methods study with 30 participants, we gathered physiological data, self-reports, and interview feedback. Our findings demonstrate how haptically experienced animacy can support emotional regulation by enabling four different coping strategies.
@inproceedings{vyas2026chora, title = {Haptically Experienced Animacy Facilitates Emotion Regulation: A Theory-Driven Investigation}, author = {Vyas, Preeti and Guta, Bereket and Zhou, Tim G. and Himam, Noor Naila and Uusberg, Andero and MacLean, Karon E.}, booktitle = {IEEE Transactions on Affective Computing}, year = {2026}, url = {https://arxiv.org/abs/2602.07395}, }
ICLR 26 ML4RS
Changing Modalities by Cross-Band Transfer, Addition, and Peeking

Tim G Zhou, Anthony Fuller, Geoff Pleiss, and Evan Shelhamer

In 4th ICLR Workshop on Machine Learning for Remote Sensing (Main Track), 2026

Abs Bib HTML PDF

Machine learning models for remote sensing typically assume a static set of modalities. However, as we equip newer satellites with novel sensors and retire old ones, practitioners may wish to deploy a model on a substitution, superset, or subset of modalities given data availability or practical constraints. We formulate the setting of changing modalities and identify three main scenarios: Modality Transfer, Addition, and Peeking. We propose Delulu-Net, an architecture with modular components for all three changing modality scenarios. Delulu-Net learns a multi-modal model from a unimodal teacher and \emphun-labeled multimodal data, providing a practical alternative to re-labeling and re-training.
@inproceedings{zhouchanging, title = {Changing Modalities by Cross-Band Transfer, Addition, and Peeking}, author = {Zhou, Tim G and Fuller, Anthony and Pleiss, Geoff and Shelhamer, Evan}, booktitle = {4th ICLR Workshop on Machine Learning for Remote Sensing (Main Track)}, year = {2026}, url = {https://openreview.net/forum?id=rkGlI80m12}, }