The regression Tsetlin machine: a novel approach to interpretable nonlinear regression

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Intelligence

Background:

• Tsetlin machines (TMs) have shown promise in pattern classification using bitwise operations.
• Existing TMs are primarily designed for discrete classification tasks.
• Nonlinear regression problems require specialized algorithms for continuous data handling.

Purpose of the Study:

• Introduce the regression Tsetlin machine (RTM) for nonlinear regression.
• Adapt Tsetlin machine principles to handle continuous input and output.
• Evaluate RTM performance against established regression techniques.

Main Methods:

• Convert continuous input data into a binary representation via thresholding.
• Utilize Tsetlin machine's propositional formula generation for regression.
• Aggregate the TM's output to produce a continuous prediction.

Main Results:

• RTM achieved competitive or superior performance on five benchmark datasets.
• Empirical comparisons validate RTM's effectiveness in nonlinear regression.
• The RTM approach offers a novel method for regression tasks.

Conclusions:

• The regression Tsetlin machine is a viable and effective tool for nonlinear regression.
• RTM offers an alternative to existing state-of-the-art regression methods.
• This work contributes to the advancement of energy-autonomous computing and intelligence.