Kai-Xiang Shu

The results support response geometry as a conditional principle for structured pruning, with its benefit determined jointly by the observed response and the architecture in which that response is realized.

Source-grounded feature inversion reveals how neural networks extract features from inputs, providing unprecedented insights into their decision-making processes.

CNN classifiers don't just select from cleaned features, they actively cancel out shared background information via destructive interference, rewriting our understanding of how these networks actually "see".