Rapid online learning and robust recall in a neuromorphic olfactory circuit

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Neuromorphic Engineering

Background:

• Mammalian olfactory systems excel at rapid odor identification despite noisy environments.
• Existing algorithms often struggle with real-time learning and robust performance under interference.

Purpose of the Study:

• To develop a neural algorithm for rapid online learning and identification of odorant samples.
• To implement this algorithm on a neuromorphic system, inspired by the mammalian olfactory bulb.
• To demonstrate robust odor identification under noisy conditions and explore lifelong learning capabilities.

Main Methods:

• A spike timing-based neural algorithm utilizing distributed, event-driven computations.
• Implementation on the Intel Loihi neuromorphic system.
• One-shot online learning with spike timing-dependent plasticity over gamma-frequency packets.
• Use of chemosensor arrays in a wind tunnel for odorant data acquisition.

Main Results:

• Reliable identification of learned odorants despite strong destructive interference.
• Enhanced noise resistance through neuromodulation and contextual priming.
• Demonstration of lifelong learning capabilities, inspired by adult neurogenesis.

Conclusions:

• The developed neural algorithm effectively mimics biological olfaction for rapid odorant identification.
• The algorithm shows significant noise resistance and lifelong learning potential.
• This approach is broadly applicable to signal identification problems with high-dimensional signals in noisy backgrounds.