Metaheuristic Artificial Bee Colony Optimization Algorithm for Neural Networks

Overview

Reinforcement learning, a machine learning training method based on rewarding desired behaviors and punishing undesired ones, has taken the field of artificial intelligence into a new age. A common method of reinforcement learning that is being used are Deep Q-Networks, or DQNs. There are still problems with the optimization of these DQNs as there are many hyperparameters to be set and the optimization simply follows a complex structure. In order to push the bounds of reinforcement learning, the optimization of these DQNs need to be simplified, but enhance performance at the same time. A novel metaheuristic swarm-based optimization algorithm, the Artificial Bee Colony, or ABC algorithm may provide the answer to the optimization problem presented by DQNs. This algorithm was introduced by Karaboga (2005) for optimizing numerical problems. It was inspired by the intelligent foraging behavior of honey bees. Throughout this project, a Python library was developed to implement the ABC algorithm in neural networks for RL tasks, in which the ABC model outperformed the standard DQN model by at least a factor of 15. However, there was a slight difference in the iteration time where the DQN model had an average of 9ms and the ABC model had an average of 12ms. This library could be applied to other RL tasks. For example, it could be used in real-world applications such as self-driving cars. With this new library, the training of neural networks used for RL tasks could be more accurate and finished within a shorter period of time.

Competition history

• AJAS 2024 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Robotics and Intelligent Machines projects
• AJAS Robotics and Intelligent Machines projects
• Every AJAS 2024 project