An end-to-end model of active electrosensation

Area of Science:

• Neuroscience
• Sensory Biology
• Computational Biology

Background:

• Weakly electric fish navigate and identify objects using self-generated electric fields.
• Sensing object resistance and capacitance relies on subtle electric field distortions.
• Neural mechanisms for electrosensation remain largely unknown.

Purpose of the Study:

• To understand the neural computations behind electrosensation in fish.
• To develop computational models simulating electrosensory object detection.
• To investigate the potential of artificial neural networks (ANNs) in modeling sensory processing.

Main Methods:

• Measured electroreceptor afferent responses and developed a filter-based model.
• Created models of electric fields and their distortions by objects.
• Trained an ANN on simulated datasets of fish-object interactions.

Main Results:

• The ANN accurately extracted 3D object location, size, and electrical properties.
• ANN performance matched that of real fish in behavioral tasks.
• Optimal ANN performance involved a two-stage process: size/distance estimation, then property extraction.

Conclusions:

• End-to-end modeling offers a powerful approach for studying electrosensation.
• Electrosensory processing may involve modularity, with distinct stages for different computations.
• Findings suggest testable hypotheses for experimental investigation of sensory processing in fish.