Deep Neural Collapse Is Provably Optimal for the Deep Unconstrained Features Model.
Peter SúkeníkMarco MondelliChristoph H. LampertPublished in: NeurIPS (2023)
Keyphrases
- high level
- computational model
- probabilistic model
- prior knowledge
- feature space
- cost function
- neural model
- learning models
- formal model
- recurrent neural networks
- experimental data
- classification method
- neural network
- management system
- low level
- statistical model
- theoretical analysis
- conceptual model
- image features
- classification models
- probability distribution
- object model