Bio-Plausible Multimodal Learning with Emerging Neuromorphic Devices.

Area of Science:

• Artificial Intelligence
• Neuroscience
• Computer Engineering

Background:

• Multimodal machine learning (ML) aims to mimic human learning but faces challenges in complexity and energy consumption.
• Neuromorphic devices offer a potential solution by efficiently processing spatio-temporal data.

Purpose of the Study:

• To compare multimodal ML with multimodal neuromorphic computing.
• To examine the characteristics, principles, and learning abilities of multimodal neuromorphic devices.

Main Methods:

• Comparative analysis of multimodal ML and neuromorphic computing.
• Review of heterogeneous and homogeneous multimodal neuromorphic device architectures.
• Examination of neuromorphic circuits' learning capabilities and applications.

Main Results:

• Multimodal neuromorphic devices enable low-complexity, energy-efficient multimodal learning.
• Devices preprocess diverse physical signals into unified electrical signals.
• Neuromorphic circuits demonstrate significant multimodal learning capabilities.

Conclusions:

• Multimodal neuromorphic computing presents a promising, bio-plausible approach for enhanced AI.
• Further research is needed to address current limitations and challenges in the field.