Alert button
Picture for Samuel L. Smith

Samuel L. Smith

Alert button

ConvNets Match Vision Transformers at Scale

Oct 25, 2023
Samuel L. Smith, Andrew Brock, Leonard Berrada, Soham De

Viaarxiv icon

Unlocking Accuracy and Fairness in Differentially Private Image Classification

Aug 21, 2023
Leonard Berrada, Soham De, Judy Hanwen Shen, Jamie Hayes, Robert Stanforth, David Stutz, Pushmeet Kohli, Samuel L. Smith, Borja Balle

Figure 1 for Unlocking Accuracy and Fairness in Differentially Private Image Classification
Figure 2 for Unlocking Accuracy and Fairness in Differentially Private Image Classification
Figure 3 for Unlocking Accuracy and Fairness in Differentially Private Image Classification
Figure 4 for Unlocking Accuracy and Fairness in Differentially Private Image Classification
Viaarxiv icon

On the Universality of Linear Recurrences Followed by Nonlinear Projections

Jul 21, 2023
Antonio Orvieto, Soham De, Caglar Gulcehre, Razvan Pascanu, Samuel L. Smith

Figure 1 for On the Universality of Linear Recurrences Followed by Nonlinear Projections
Figure 2 for On the Universality of Linear Recurrences Followed by Nonlinear Projections
Figure 3 for On the Universality of Linear Recurrences Followed by Nonlinear Projections
Figure 4 for On the Universality of Linear Recurrences Followed by Nonlinear Projections
Viaarxiv icon

Differentially Private Diffusion Models Generate Useful Synthetic Images

Feb 27, 2023
Sahra Ghalebikesabi, Leonard Berrada, Sven Gowal, Ira Ktena, Robert Stanforth, Jamie Hayes, Soham De, Samuel L. Smith, Olivia Wiles, Borja Balle

Figure 1 for Differentially Private Diffusion Models Generate Useful Synthetic Images
Figure 2 for Differentially Private Diffusion Models Generate Useful Synthetic Images
Figure 3 for Differentially Private Diffusion Models Generate Useful Synthetic Images
Figure 4 for Differentially Private Diffusion Models Generate Useful Synthetic Images
Viaarxiv icon

Unlocking High-Accuracy Differentially Private Image Classification through Scale

Apr 28, 2022
Soham De, Leonard Berrada, Jamie Hayes, Samuel L. Smith, Borja Balle

Figure 1 for Unlocking High-Accuracy Differentially Private Image Classification through Scale
Figure 2 for Unlocking High-Accuracy Differentially Private Image Classification through Scale
Figure 3 for Unlocking High-Accuracy Differentially Private Image Classification through Scale
Figure 4 for Unlocking High-Accuracy Differentially Private Image Classification through Scale
Viaarxiv icon

Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error

May 27, 2021
Stanislav Fort, Andrew Brock, Razvan Pascanu, Soham De, Samuel L. Smith

Figure 1 for Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error
Figure 2 for Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error
Figure 3 for Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error
Figure 4 for Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error
Viaarxiv icon

High-Performance Large-Scale Image Recognition Without Normalization

Feb 11, 2021
Andrew Brock, Soham De, Samuel L. Smith, Karen Simonyan

Figure 1 for High-Performance Large-Scale Image Recognition Without Normalization
Figure 2 for High-Performance Large-Scale Image Recognition Without Normalization
Figure 3 for High-Performance Large-Scale Image Recognition Without Normalization
Figure 4 for High-Performance Large-Scale Image Recognition Without Normalization
Viaarxiv icon

On the Origin of Implicit Regularization in Stochastic Gradient Descent

Jan 28, 2021
Samuel L. Smith, Benoit Dherin, David G. T. Barrett, Soham De

Figure 1 for On the Origin of Implicit Regularization in Stochastic Gradient Descent
Figure 2 for On the Origin of Implicit Regularization in Stochastic Gradient Descent
Figure 3 for On the Origin of Implicit Regularization in Stochastic Gradient Descent
Figure 4 for On the Origin of Implicit Regularization in Stochastic Gradient Descent
Viaarxiv icon

Characterizing signal propagation to close the performance gap in unnormalized ResNets

Jan 27, 2021
Andrew Brock, Soham De, Samuel L. Smith

Figure 1 for Characterizing signal propagation to close the performance gap in unnormalized ResNets
Figure 2 for Characterizing signal propagation to close the performance gap in unnormalized ResNets
Figure 3 for Characterizing signal propagation to close the performance gap in unnormalized ResNets
Figure 4 for Characterizing signal propagation to close the performance gap in unnormalized ResNets
Viaarxiv icon