Time Matters in Regularizing Deep Networks: Weight Decay and Data Augmentation Affect Early Learning Dynamics, Matter Little Near Convergence.
Aditya GolatkarAlessandro AchilleStefano SoattoPublished in: NeurIPS (2019)
Keyphrases
- network structure
- social networks
- hidden variables
- learning algorithm
- image data
- learning process
- data sets
- knowledge discovery
- high quality
- database
- prior knowledge
- supervised learning
- data processing
- missing data
- data mining techniques
- original data
- learning tasks
- neural network
- experimental data
- spatial data
- data analysis
- training data
- data sources
- active learning
- data structure
- mobile devices