The Leaky Integrate-and-Fire Neuron Is a Change-Point Detector for Compound Poisson Processes

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Statistical Signal Processing

Background:

• Animal nervous systems detect environmental changes via abrupt shifts in neural activity.
• Change-point detection (CPD) algorithms are commonly used to analyze these shifts.
• Few studies explore spiking neurons as inherent online CPD agents.

Purpose of the Study:

• To demonstrate that a leaky integrate-and-fire (LIF) neuron implements an online CPD algorithm.
• To analyze the performance of LIF neuron CPD across its parameter space.
• To investigate if neural networks of LIF neurons can detect changes in spiking rates.

Main Methods:

• Modeling a leaky integrate-and-fire (LIF) neuron as a CPD algorithm for compound Poisson processes.
• Quantifying LIF neuron CPD performance across parameter variations.
• Analyzing a feedforward network of LIF neurons for detecting input rate changes.

Main Results:

• An LIF neuron was shown to implement an online CPD algorithm.
• A feedforward network of LIF neurons detected a 5% change in input rates within 20 ms with rare false positives.
• Key electrophysiological features of LIF neurons were statistically interpreted in the context of CPD.

Conclusions:

• Spiking neurons, specifically LIF neurons, can function as sophisticated online statistical change-point detectors.
• Neural networks of LIF neurons exhibit efficient and reliable detection of subtle input rate changes.
• This suggests a re-evaluation of neurons not as noisy units but as implementers of optimal statistical algorithms.