A Unified Framework for Reservoir Computing and Extreme Learning Machines based on a Single Time-delayed Neuron

Area of Science:

• Computational Neuroscience
• Machine Learning Hardware

Background:

• Extreme learning machines (ELMs) and reservoir computing (RC), specifically echo state networks (ESNs), are powerful computational models.
• Implementing these models efficiently in hardware remains a challenge.

Purpose of the Study:

• To present a unified framework for ELMs and ESNs.
• To demonstrate a hardware-efficient implementation of this unified framework.

Main Methods:

• A unified framework is proposed for ELMs and ESNs.
• The framework utilizes a single nonlinear neuron with delayed feedback to create a reservoir of "virtual" neurons.
• Random projections from the input layer are used to feed information to these virtual neurons.

Main Results:

• The proposed framework can be physically implemented using a single nonlinear neuron with delayed feedback.
• The reservoir computing (optoelectronic) implementation demonstrates the capability to realize extreme learning machines.
• Both software and hardware demonstrations validate the unified framework.

Conclusions:

• A unified framework for ELMs and ESNs is successfully presented.
• Efficient hardware implementation is achievable using a single nonlinear neuron with delayed feedback.
• The unified framework bridges the gap between theoretical models and practical hardware applications.