Modeling the insect mushroom bodies: application to a delayed match-to-sample task

Area of Science:

• Neuroscience and Artificial Intelligence
• Computational Neuroscience
• Robotics

Background:

• Insects exhibit advanced learning and task-solving despite small brains, inspiring neuroscience and robotics.
• Mushroom Bodies (MBs) are crucial for memory and learning in insects, particularly in olfactory conditioning.
• Fruit flies offer insights into MB neuroanatomy, though honeybee MBs are more complex.

Purpose of the Study:

• To present a novel bio-inspired neural architecture modeling a generalized insect Mushroom Body (MB).
• To improve upon existing artificial neural networks by mimicking MB functions and architecture.
• To demonstrate the model's capability in replicating insect learning and abstract association tasks.

Main Methods:

• Developed a multi-layer spiking neural network modeling key insect brain regions: antennal lobes, lateral horn, and MBs.
• Focused on MB-lobes and spatio-temporal pattern formation for processing mechanisms.
• Modeled olfactory conditioning and delayed matching-to-sample tasks.

Main Results:

• The neural network successfully models learning mechanisms like olfactory conditioning in honeybees and flies.
• The model demonstrated the ability to perform complex, abstract associations, such as delayed matching-to-sample tasks.
• Simulation results support the biological plausibility and potential applications of the computational model.

Conclusions:

• The proposed neural architecture provides a generalized model of insect MBs.
• This bio-inspired network can replace and enhance traditional artificial neural networks for learning tasks.
• The architecture is suitable for implementation in robot control systems for autonomous learning.