Skip to content

Reinforcement Learning Algorithm to perform Neural Architecture Search(NAS) in DNNs - Model 1

License

Notifications You must be signed in to change notification settings

wolflegend99/fyp

Repository files navigation

fyp

Description

Repo for final year project

Neural Architecture Search poses a problem for Deep Learning. We use a Reinforcement Learning method to solve this issue, with multiple worker agents. The approach uses multiple DDPG agents, which are run by a controller and treats the environment of the dataset as a input to work on.

  • Model: variable layers, variable number of nodes but same in each hidden layer
Framework: PyTorch
Mode of Update: Asynchronous
Algorithm: Multi-Agent DDPG
Environment: Datasets
Neural Network: DNNs
Sampling: Replay Buffer
Type of Learning: Actor-Critic Method

The controller can be trained on GPUs as well.

Dependencies

torch
numpy
pandas
sklearn

Installation

git clone https://github.com/wolflegend99/fyp.git

About

Reinforcement Learning Algorithm to perform Neural Architecture Search(NAS) in DNNs - Model 1

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages