A new type of neurons for machine learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Artificial neural networks (ANNs) are a cornerstone of modern machine learning.
• Traditional ANNs utilize neurons with an inner product of input vectors and trainable weights.
• This structure limits the computational capacity of individual neurons.

Purpose of the Study:

• To explore the potential of enhancing artificial neural network neurons.
• To introduce a novel 'second-order neuron' by modifying the neuron's core computation.
• To improve the efficiency and effectiveness of neural network models.

Main Methods:

• Proposed replacing the standard inner product in neurons with a quadratic function of the input vector.
• Developed the theoretical framework for second-order neurons.
• Conducted numerical experiments to validate the concept.

Main Results:

• Demonstrated the feasibility of implementing second-order neurons.
• Showcased the enhanced capabilities of individual second-order neurons.
• Illustrated the benefits for neural network optimization through examples.

Conclusions:

• Second-order neurons offer a promising advancement over traditional first-order neurons.
• The proposed modification can lead to more powerful and easier-to-optimize neural networks.
• Further research into second-order neuron architectures and applications is warranted.