“Liquid” machine-learning system adapts to changing conditions

MIT News  January 28, 2021
An international team of researchers (USA – MIT, Austria) designed a neural network that can adapt to the variability of real-world systems. They took inspiration from C.elegans which has only 302 neurons in its nervous system, yet it can generate unexpectedly complex dynamics. The equations they used to structure their neural network allowed the parameters to change over time based on the results of a nested set of differential equations. Most neural networks’ behavior is fixed after the training phase. The fluidity of their “liquid” network makes it more resilient to unexpected or noisy data and hence more robust. Flexibility makes it more interpretable. Just changing the representation of a neuron, with the differential equations, it is possible to explore some degrees of complexity that could not be explored otherwise. It is easier to peer into the “black box” of the network’s decision making and diagnose why the network made a certain characterization. The network excelled in a battery of tests. The advance could aid decision making based on data streams that change over time, including those involved in medical diagnosis and autonomous driving…read more. Open Access TECHNICAL ARTICLE 

… “liquid” network that varies its equations’ parameters, enhancing its ability to analyze time series data. Credits: Image: Jose-Luis Olivares, MIT

Posted in Liquid neural network and tagged , .

Leave a Reply