Deep Learning
RecursiveMix: Mixed Learning with History Lingfeng Y ang
Mix-based augmentation has been proven fundamental to the generalization of deep vision models.
Neural Basis Models for Interpretability
However, these models are typically difficult to train, require numerous parameters, and are difficult to scale.
Zonotope Domains for Lagrangian Neural Network Verification Matt Jordan
Under the dual lens, verification is treated as a stagewise optimization problem and Lagrangian relaxation is applied to yield a dual function that always provides valid lower bounds.
Zonotope Domains for Lagrangian Neural Network Verification Matt Jordan
Under the dual lens, verification is treated as a stagewise optimization problem and Lagrangian relaxation is applied to yield a dual function that always provides valid lower bounds.
KERPLE: Kernelized Relative Positional Embedding for Length Extrapolation T a-Chung Chi
RPEs effectively model the relative distance among tokens and enable length extrapolation.
Vision GNN: An Image is Worth Graph of Nodes Kai Han 1,2 Yunhe Wang
These progresses are pushing the vision models towards an unprecedented height. Different networks treat the input image in different ways. As shown in Figure 1, the image data is usually represented as a regular grid of pixels in the Euclidean space.