M.Sc. student in electrical and computer engineering at the University of Alberta; B.E. in automation from Jilin University. Along the way: bio-inspired robot navigation at NII, Tokyo, and robotic manipulation at Western University.
A synapse-local predictive-coding learning rule that removes the Jacobian transpose from backpropagation's error transport — and, unlike previous predictive-coding methods, gets better as the network gets deeper.
The winning Track-1 entry to the ChaLearn UDIVA-HHOI challenge, with an analysis of why the ceiling is low — the binding limit is available context, not model capacity.
System paper for the shared task — first of 31 on seen targets, second of 22 on unseen — deriving the precision a repair must reach to pay off on a metric that excludes the NONE class.
A unified framework for 3D detection that survives the move out of the dataset — combining multi-modal fusion with contrastive-enhanced prototype learning to face domain shift and data scarcity at once.
Collaborative 3D object detection fusing vehicle and infrastructure sensing on the Mixed Signals V2X benchmark.
Assessing parkinsonian gait from clinical video — ranked first overall on the benchmark.
Semi-supervised teeth and pulp-canal segmentation with metal-artifact removal in CBCT, crown–root registration across CBCT and intraoral scans, and multimodal dental analysis.
Multimodal event recognition, anticipation, and causal grounding in human–human-object interaction, exocentric and egocentric — first across every track of the challenge.
Detecting the stance of Arabic text toward its target — first on Track 1, where targets are seen in training, and second on Track 2, where they are not.
Predicting the externalizing dimension of the psychopathology p-factor from EEG, invariant across subjects — first at 0.97843; tenth on Challenge 1, cross-task transfer of behavioural response time.
Second in both tracks entered — Sensor Placement (Track 3) and Cross-Platform 3D Object Detection (Track 5) — on robust robot sensing across platforms and viewpoints.
An esports-style, full-stack robotics league with fleets of up to nine robots — led the software team across a two-year cycle, building real-time armor-plate detection for turret auto-aiming in C++ and SLAM navigation with AI perception for the sentry.
我转身背离过去，面对火焰，把遗憾烧尽。
I turned from the past to face the flame, and burned the regret away.
人们用道路切割大地；高楼大厦就是人类的蜂巢。
People cut the earth apart with roads; the towers are the human hive.
On the way