Depth Self-Optimized Learning Toward Data Science

TL;DR

This paper introduces Depth Self-Optimized Learning (DSOL), a two-stage method for automatic neural network depth configuration and optimization using reinforcement learning, demonstrated on small datasets.

Contribution

The paper presents a novel two-stage framework for automatic neural network depth configuration and optimization without manual intervention.

Findings

DSOL effectively configured neural network depth for Iris and Boston datasets.

Reinforcement learning improved neural network performance through continuous optimization.

The approach reduces manual tuning in neural network design.

Abstract

We propose a two-stage model called Depth Self-Optimized Learning (DSOL), which aims to realize ANN depth self-configuration, self-optimization as well as ANN training without manual intervention. In the first stage of DSOL, it will configure ANN of specific depth according to a specific dataset. In the second stage, DSOL will continuously optimize ANN based on Reinforcement Learning (RL). Finally, the optimal depth is returned to the first stage of DSOL for training, so that DSOL can configure the appropriate ANN depth and perform more reasonable optimization when processing similar datasets again. In the experiment, we ran DSOL on the Iris and Boston housing datasets, and the results showed that DSOL performed well. We have uploaded the experiment records and code to our Github.