Identification models of the nervous system

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Neuroscience

Background:

• Artificial neural networks (ANNs) trained via supervised learning often exhibit neuron behavior similar to biological neurons.
• The reasons for this resemblance and the utility of such models are not fully understood.

Purpose of the Study:

• To review and analyze recent developments in supervised learning models of the brain.
• To clarify the relationship between ANNs and neural computation.
• To establish a framework for generating realistic neural models.

Main Methods:

• Treating supervised learning models as a specific instance of system identification.
• Utilizing a general and well-established modeling paradigm.
• Incorporating high-level computational descriptions and neurobiological constraints.

Main Results:

• System identification provides a systematic approach to creating realistic neural models.
• This paradigm allows for the integration of detailed architectural and physiological constraints.
• It offers a structured method for model generation, avoiding ad hoc algorithm discovery.

Conclusions:

• Supervised learning models can be effectively understood through the lens of system identification.
• This approach facilitates the creation of neurobiologically plausible and data-fitting neural models.
• It offers a powerful tool for neuroscientists to build and validate computational models of the brain.