Samuel Schoenholz
10 papers · 2017–2020 · 3 conferences · across top CS/AI conferences
Achievements
Jump to papers ↓+4 more ↓ Show less ↑
π£ Hot Topic Early Bird π§ Keyword Pioneer πΊοΈ Taxonomy Completionist (19) π Interdisciplinary Bridge π Conference Polyglot (3)
π
Cross-Pollinator
(4)
π
Renaissance Researcher
(6)
π
Keyword Champion
(3)
π
Century Club
(10)
Conferences
NIPS (6)
ICML (3)
AISTATS (1)
Top co-authors
Keywords
dynamical isometry
(3)
mean field theory
(3)
initialization scheme
(3)
convolutional neural network
(2)
gradient descent
(2)
free probability theory
(2)
vanishing gradient
(2)
neural tangent kernel
(2)
weight initialization
(2)
gaussian process
(1)
recurrent neural network
(1)
neural network optimization
(1)
generalization bound
(1)
gradient flow
(1)
random matrix theory
(1)
batch normalization
(1)
residual network
(1)
singular value distribution
(1)
linear model
(1)
molecular dynamics
(1)
Papers
Disentangling Trainability and Generalization in Deep Neural Networks
ICML 2020
JAX MD: A Framework for Differentiable Physics
NIPS 2020
Finite Versus Infinite Neural Networks: an Empirical Study
NIPS 2020
MetaInit: Initializing learning by learning to initialize
NIPS 2019
Wide Neural Networks of Any Depth Evolve as Linear Models Under Gradient Descent
NIPS 2019
Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks
ICML 2018
The emergence of spectral universality in deep networks
AISTATS 2018
Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural Networks
ICML 2018
Resurrecting the sigmoid in deep learning through dynamical isometry: theory and practice
NIPS 2017
Mean Field Residual Networks: On the Edge of Chaos
NIPS 2017