no code implementations • 26 Jul 2023 • Amnon Geifman, Daniel Barzilai, Ronen Basri, Meirav Galun
We leverage the duality between wide neural networks and Neural Tangent Kernels and propose a preconditioned gradient descent method, which alters the trajectory of GD.
no code implementations • 27 Nov 2022 • Daniel Barzilai, Amnon Geifman, Meirav Galun, Ronen Basri
Over-parameterized residual networks (ResNets) are amongst the most successful convolutional neural architectures for image processing.
no code implementations • 17 Mar 2022 • Amnon Geifman, Meirav Galun, David Jacobs, Ronen Basri
We study the properties of various over-parametrized convolutional neural architectures through their respective Gaussian process and neural tangent kernels.
no code implementations • 7 Apr 2021 • Yuval Belfer, Amnon Geifman, Meirav Galun, Ronen Basri
Deep residual network architectures have been shown to achieve superior accuracy over classical feed-forward networks, yet their success is still not fully understood.
1 code implementation • NeurIPS 2020 • Amnon Geifman, Abhay Yadav, Yoni Kasten, Meirav Galun, David Jacobs, Ronen Basri
Experiments show that these kernel methods perform similarly to real neural networks.
no code implementations • ICML 2020 • Ronen Basri, Meirav Galun, Amnon Geifman, David Jacobs, Yoni Kasten, Shira Kritchman
Recent works have partly attributed the generalization ability of over-parameterized neural networks to frequency bias -- networks trained with gradient descent on data drawn from a uniform distribution find a low frequency fit before high frequency ones.
no code implementations • CVPR 2020 • Amnon Geifman, Yoni Kasten, Meirav Galun, Ronen Basri
Global methods to Structure from Motion have gained popularity in recent years.
no code implementations • ICCV 2019 • Yoni Kasten, Amnon Geifman, Meirav Galun, Ronen Basri
A common approach to essential matrix averaging is to separately solve for camera orientations and subsequently for camera positions.
1 code implementation • CVPR 2019 • Yoni Kasten, Amnon Geifman, Meirav Galun, Ronen Basri
First, given ${n \choose 2}$ fundamental matrices computed for $n$ images, we provide a complete algebraic characterization in the form of conditions that are both necessary and sufficient to enabling the recovery of camera matrices.