Infinite wide (finite depth) Neural Networks benefit from multi-task learning unlike shallow Gaussian Processes - an exact quantitative macroscopic characterization.
Jakob HeissJosef TeichmannHanna WuttePublished in: CoRR (2021)
Keyphrases
- gaussian processes
- multi task learning
- multi task
- gaussian process
- multitask learning
- gaussian process regression
- dirichlet process
- covariance function
- natural language processing
- learning algorithm
- information extraction
- theoretical analysis
- semi supervised
- high order
- hyperparameters
- regression model
- high dimensional