Exploring Shallow Architectures for Image Classification

Explore shallow architectures for image classification in this 28-minute conference talk by Edouard Oyallon from Institut des Hautes Etudes Scientifiques (IHES). Delve into the comparison between Deep Convolutional Neural Networks (CNNs) and Scattering Networks, a two-layer deep CNN architecture derived from cascaded complex wavelet transforms and modulus pointwise non-linearity. Examine the central question that drove Oyallon's PhD research: "Is it possible to derive competitive representations for image classification using geometric arguments?" Discover how this inquiry, although not yielding the desired outcome, led to an intriguing research direction focused on the potential of shallow architectures in tackling the ImageNet dataset. Review the findings and discuss potential challenges in the area of shallow learning, as presented by Edouard Oyallon, a researcher from CNRS & Sorbonne Université.