Tree DNN: A Deep Container Network

TL;DR

TreeDNN is a novel deep network architecture designed for multi-task learning across different datasets, enabling efficient training, reduced storage, and faster inference by selectively loading branches.

Contribution

The paper introduces TreeDNN, a new architecture that allows multi-dataset training with branch-specific data, improving efficiency and responsiveness over traditional methods.

Findings

Competitive performance with reduced parameter storage

Faster inference by loading specific branches

Effective multi-dataset training with TreeDNN

Abstract

Multi-Task Learning (MTL) has shown its importance at user products for fast training, data efficiency, reduced overfitting etc. MTL achieves it by sharing the network parameters and training a network for multiple tasks simultaneously. However, MTL does not provide the solution, if each task needs training from a different dataset. In order to solve the stated problem, we have proposed an architecture named TreeDNN along with it's training methodology. TreeDNN helps in training the model with multiple datasets simultaneously, where each branch of the tree may need a different training dataset. We have shown in the results that TreeDNN provides competitive performance with the advantage of reduced ROM requirement for parameter storage and increased responsiveness of the system by loading only specific branch at inference time.