HAMLET: A Hierarchical Agent-Based Machine Learning Platform For AI Research And Development

0
2260
Source: https://arxiv.org/pdf/2010.04894.pdf

Machine learning (ML) algorithms are widely used as computational tools for solving various real-world problems, including image, audio, and text classification tasks. New algorithms are developed regularly. Keeping a record of the massive volume of new algorithms and instantly accessing those presented in the past is becoming more challenging.

The researchers at Purdue University and the University of Cincinnati recently built a platform called HAMLET. It can help computer scientists and developers browse through existing machine learning models. This platform supports scientists in research and development by assisting them in evaluating and training their algorithms. It allows research teams to share their models and could eventually democratize ML models developed worldwide.

Advertisement

Ahmad Esmaeili, one of the researchers, said that organizing and managing records of the ML algorithms and datasets have always been a significant challenge for researchers in the domain. This becomes an even more tedious task as the number of ML solutions and components continues to grow. They aimed at creating a platform that administers the available ML contributions and assets in a distributed way and facilitates actions such as accessing, comparing, and evaluating those resources efficiently.

https://arxiv.org/pdf/2010.04894.pdf

HAMLET stands for Hierarchical Agent-based Machine LEarning plaTform. It comprises a group of AI tools trained to manage a large group of machine learning algorithms, related resources, and tasks that machine learning models are equipped to achieve. The researchers described the artificial agents’ skills that manage the platform, arranged at various levels of a hierarchy based on the data, algorithms, or tasks they represent. Hierarchical Multi-Agent Systems give a suitable and appropriate way to analyze, model, and simulate complicated systems in which a large number of entities are interacting at various levels of abstraction.

This platform starts with an empty structure and autonomously evolves by introducing new machine learning resources/queries over time. The platform is based on multi-agent systems and can be distributed over a network of computers and devices. Thus, there is no constraint on the size and type of algorithms and data they can host.

The platform has a flexible query structure and a user-friendly interface. Researchers can use it to perform various responsibilities, such as training and testing their algorithms individually and in batches.

The researchers used the HAMLET platform to complete 120 training and four batch testing assignments on a simulated environment generated with Smart Python Agent Development Environment SPADE, to test its effectiveness. They regularly examined and trained 24 machine learning algorithms using nine renowned datasets for training AI agents. The results recommend that HAMLET is a very assuring and useful tool for training and testing machine learning algorithms.

With HAMLET, machine learning solutions are made accessible to everyone. Thus the platform helps the machine learning research societies easily share and record their methods and resources, despite their geographical locations.

In the future, the researchers can use this platform worldwide to train new machine learning algorithms on multiple datasets, identify existing models for specific purposes or evaluate new algorithms and compare their performance to that of other existing ones. All these tasks can efficiently be executed through a single query on HAMLET.

The team aims to conduct studies further and work on supporting more complex algorithms, the platform’s survivability against failures, merging various platforms, and the privacy of accessing algorithms. This project is in its start and can be enhanced in many aspects to ensure that it adequately satisfies the modern research and industrial requirements.

Paper: https://arxiv.org/pdf/2010.04894.pdf

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.