TRCLA: A Transfer Learning Approach to Reduce Negative Transfer for Cellular Learning Automata.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Traditional machine learning assumes identical data distributions, which often fails in real-world scenarios.
• Transfer learning addresses distribution shifts but introduces negative transfer (NT) challenges.
• Existing research inadequately addresses the significant problem of NTs.

Purpose of the Study:

• To propose a novel transductive learning algorithm to alleviate the negative transfer (NT) issue.
• To introduce new decision criteria within cellular learning automata (CLA) to mitigate NTs.
• To enhance the performance of machine learning models in non-identical data distribution environments.

Main Methods:

• Developed a transductive learning algorithm based on cellular learning automata (CLA).
• Employed two established learning automata (LA) as estimator CLAs.
• Introduced novel 'merit' and 'attitude' parameters to CLA for NT limitation.

Main Results:

• The proposed CLA-based algorithm demonstrated reduced negative transfer (NT).
• Experimental results showed higher accuracy compared to existing methods.
• The algorithm effectively addressed challenges posed by non-identical data distributions.

Conclusions:

• The novel transductive learning algorithm based on CLA successfully mitigates negative transfer.
• The introduced decision criteria (merit and attitude parameters) are effective in limiting NTs.
• This approach offers improved accuracy and robustness for machine learning in diverse data environments.