Soft Parameter Sharing Multi-Task Learning. Parameter sharing Parameters ---------- mtl_layers: List of Layer instances

Cross-stitch block References ---------- [1] Cross-stitch Networks for Multi-task Learning, (2017) Ishan Misra et al

Mixture of Experts Layer Parameters ---------- expert_layers: List of Layers List of experts, each expert is expected to be

Standard neural network with dense layers Parameters ---------- architecture: List of ints Architecture of the network hi

Multi-Gate Mixture of Experts - multitask learning model that allows automatically learn measure of relationship between tasks and adapts so

One-Gate Mixture of Experts - similar to Multi-Gate Mixture of Experts. The only difference is that every task shares the same MOE layer, whi

Helper layer for Mixture of Experts Layer, allows to drop some of the experts and then renormalize other experts, so that we still have proba

Layer that preprocesses input features of UCI Income Census (1994), creates embeddings for categorical features. Note this layer assumes nume

Unit tests for Mixture of Experts Layer

Base class for testing multitask and single tak models

Loads and preprocesses data the way it is decribed in section 6.3 of the paper. Parameters ---------- local_path: str, optional

Test Multi-task learning through soft parameter sharing

Test cross-stitch block

Unit tests for Muli-Gate Mixture of Experts Layer

Test utils

Hyperparameter tuner for Multi-task models on Income Census dataset