A Winner-Take-All Approach to Emotional Neural Networks with Universal Approximation Property
This work addresses the need for efficient neural network architectures in applications requiring low computational complexity, such as real-world prediction and pattern recognition, though it appears incremental as it builds on existing emotional neural network concepts.
The authors tackled the problem of designing neural networks with high information capacity and low complexity by proposing a brain-inspired winner-take-all emotional neural network (WTAENN), proving its universal approximation property and demonstrating superior accuracy and lower model complexity on various tasks like classification and prediction.
Here, we propose a brain-inspired winner-take-all emotional neural network (WTAENN) and prove the universal approximation property for the novel architecture. WTAENN is a single layered feedforward neural network that benefits from the excitatory, inhibitory, and expandatory neural connections as well as the winner-take-all (WTA) competitions in the human brain s nervous system. The WTA competition increases the information capacity of the model without adding hidden neurons. The universal approximation capability of the proposed architecture is illustrated on two example functions, trained by a genetic algorithm, and then applied to several competing recent and benchmark problems such as in curve fitting, pattern recognition, classification and prediction. In particular, it is tested on twelve UCI classification datasets, a facial recognition problem, three real world prediction problems (2 chaotic time series of geomagnetic activity indices and wind farm power generation data), two synthetic case studies with constant and nonconstant noise variance as well as k-selector and linear programming problems. Results indicate the general applicability and often superiority of the approach in terms of higher accuracy and lower model complexity, especially where low computational complexity is imperative.