One-core-neuron-system encodes high-dimensional data into one-dimensional time-series representation

https://www.eurekalert.org/news-releases/1069796

"compact, high performing/ efficient deep learning... unlike traditional large models that rely on billions of parameters... based on delay embedding theorem, small model framework with spatiotemporal information transformation and multiple delayed feedback. This approach ensures precise forecasting while requiring, on average, only 0.035% of the parameters used in large models. Applications range from time-series prediction to image classification tasks"