Authors: Hitesh Vaidya, Travis Desell, Ankur Mali, Alexander Ororbia

Published on: February 19, 2024

Impact Score: 7.8

Arxiv code: Arxiv:2402.12465

Summary

• What is new: A new model called the continual SOM (CSOM) has been proposed, which has shown to significantly outperform traditional self-organizing maps (SOMs) in terms of learning accuracy on continuous data streams without suffering from catastrophic forgetting.
• Why this is important: The challenge of catastrophic forgetting in artificial neural networks (ANNs), especially in unsupervised architectures like SOMs when dealing with continuous data streams without clear task boundaries.
• What the research proposes: The introduction of the continual SOM (CSOM), an advanced version of the traditional SOM that is capable of online unsupervised learning with a low memory footprint.
• Results: CSOM achieves almost a two times increase in accuracy on benchmarks such as MNIST, Kuzushiji-MNIST, Fashion-MNIST, and state-of-the-art results on CIFAR-10 for online unsupervised class incremental learning.

Technical Details

Technological frameworks used: CSOM (Continual Self-Organizing Map)

Models used: Self-organizing maps (SOMs), Artificial Neural Networks (ANNs)

Data used: MNIST, Kuzushiji-MNIST, Fashion-MNIST, CIFAR-10

Potential Impact

AI and machine learning development platforms, educational tech companies, e-commerce, image and pattern recognition services

Want to implement this idea in a business?

We have generated a startup concept here: MindMapAI.