(2021). On the Cryptographic Hardness of Learning One-Hidden Layer Neural Networks. The Simons Institute for the Theory of Computing. https://simons.berkeley.edu/talks/cryptographic-hardness-learning-one-hidden-layer-neural-networks
MLA
On the Cryptographic Hardness of Learning One-Hidden Layer Neural Networks. The Simons Institute for the Theory of Computing, Dec. 07, 2021, https://simons.berkeley.edu/talks/cryptographic-hardness-learning-one-hidden-layer-neural-networks
BibTex
@misc{ scivideos_18851,
doi = {},
url = {https://simons.berkeley.edu/talks/cryptographic-hardness-learning-one-hidden-layer-neural-networks},
author = {},
keywords = {},
language = {en},
title = {On the Cryptographic Hardness of Learning One-Hidden Layer Neural Networks},
publisher = {The Simons Institute for the Theory of Computing},
year = {2021},
month = {dec},
note = {18851 see, \url{https://scivideos.org/Simons-Institute/18851}}
}
In this short talk, I will share some recent progress on the hardness of learning shallow RELU neural networks (Relu-NN) and polynomially small adversarial noise. We will present a result that efficiently learning an 1-hidden layer Relu-NN under Gaussian input and adversarial noise is "cryptographically hard", in the sense that it implies a polynomial-time quantum algorithm for the worst-case shortest vector problem.
