ConvNN
Convolutional Nearest Neighbor (ConvNN) for Neural Networks
TL;DR
Convolutional Nearest Neighbor (ConvNN) is a novel method that reinterpreting convolutional operations as a form of nearest neighbor search. Nearest neighbor search can be from either spatial or feature space. This approach allows neural networks to enhance their feature representation capabilities by leveraging the strengths of both convolutional operations and nearest neighbor search techniques.
Paper
The paper on Attention Via Convolutional Nearest Neighbors is available on arXiv: here.
Presentation & Poster
Presented at MIT URTC (Undergraduate Research Technology Conference) 2025.
Shahd Hekal (Bowdoin '27) and I at MIT URTC 2025
Photos from the trip to Boston and Cambridge:
Newbury Street (Boston), Charles River (Cambridge), and MIT (Cambridge)