Reproduce “The Shape of Events: Edge-Based Inductive Biases via Cross-Domain Distillation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Soshun Kihara, Shunsuke Yasuki, Masato Taki
The authors transfer an event-trained teacher’s predictions to an RGB ResNet-34 student, combining soft targets with ground-truth labels. Tests cover color, shape-conflict, corruption, and transfer. ED_Student improves on all four axes, including 57.5% accuracy under high-frequency noise, but ImageNet accuracy falls from 73.9% to 69.7%; conclusions are limited to tested setups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.