Deep Convolutional Neural Networks (CNNs) have achieved remarkable success in various tasks, particularly in image classification. In contrast, Scattering Networks, a two-layer deep CNN architecture derived from cascaded complex wavelet transforms and modulus pointwise non-linearity, have shown promise but lag behind deep CNNs in terms of performance on the widely recognized ImageNet dataset In this talk, we revisit the central question that drove my PhD research: “Is it possible to derive competitive representations for image classification using geometric arguments?” Although this inquiry did not yield the desired outcome, it sparked an intriguing research direction focusing on the potential of shallow architectures in tackling the ImageNet dataset. We will review these findings and discuss potential challenges in the area of shallow learning.
De Eero Simoncelli
De Naoki Saito
De Maarten de Hoop
De Martin Vetterli
De Maureen Clerc
De Irène Waldspurger
De Jérôme Kalifa
De Gitta Kutyniok
De Michael Unser
De Edouard Oyallon
De Remi Gribonval
De Ronnie Pavlov
De Viviane Pons
De Wolfgang Arendt
De David Hewett
