Brain-inspired reservoir computing framework for complex time-series prediction

Area of Science:

• Computational Neuroscience
• Machine Learning
• Time-Series Analysis

Background:

• Accurate time-series prediction is crucial across scientific domains.
• Achieving both high accuracy and computational efficiency in predictions remains a significant challenge.

Purpose of the Study:

• To introduce a brain-inspired neural network, the echo state network (ESN), for enhanced time-series prediction.
• To evaluate the performance of the proposed BINN-ESN against established deep learning baselines.

Main Methods:

• Developed a brain-inspired neural network-based echo state network (BINN-ESN).
• Utilized a modified continuous coupled neural network (MCCNN), inspired by the mammalian visual cortex, as the neural model.
• Compared BINN-ESN with GPU-accelerated long short-term memory (LSTM) baselines.

Main Results:

• BINN-ESN demonstrated superior long-term predictive stability across various systems.
• The proposed model required substantially less computation than LSTM for long-term predictions.
• A system-dependent trade-off was observed, with deep learning baselines showing stronger single-step accuracy.

Conclusions:

• BINN-ESN offers a computationally efficient and stable solution for long-term time-series prediction.
• The brain-inspired approach presents a viable alternative to traditional deep learning methods.
• Code availability facilitates further research and application of the BINN-ESN model.