Feature Reconstruction From Outputs Can Mitigate Simplicity Bias in Neural Networks.
Sravanti AddepalliAnshul NaseryVenkatesh Babu RadhakrishnanPraneeth NetrapalliPrateek JainPublished in: ICLR (2023)
Keyphrases
- neural network
- pattern recognition
- feed forward
- back propagation
- artificial neural networks
- high resolution
- image features
- three dimensional
- feature set
- genetic algorithm
- neural nets
- training process
- data sets
- hopfield neural network
- shape recovery
- reconstruction method
- reconstruction error
- fuzzy systems
- recurrent neural networks
- multilayer perceptron
- neural network model
- feature vectors