What is ‘liquid’ machine learning?
Machine learning has taken another step forward after researchers at MIT cracked what has been dubbed ‘liquid’ machine learning networks.
What is liquid machine learning?
Most AI works on the basis of a training phase, during which the algorithm is exposed to massive amounts of training data. Once sufficiently trained, the ML is implemented in a real-world situation, applying what it has learned to new data. With liquid machine learning, the algorithms are flexible and continue learning once they are operational. That means they can adapt to data streams that change over a period of time.
How is liquid machine learning useful?
The researchers say liquid machine learning could help with medical diagnosis and autonomous driving. The study’s lead author, Ramin Hasani, says, “This is a way forward for the future of robot control, natural language processing, video processing – any form of time series data processing. The potential is really significant.”
Is liquid machine learning ready to implement now?
Not yet. The research is being presented at the AAAI Conference on Artificial Intelligence in February 2021 by CSAIL researchers and others who have worked on the project. But assuming it gets a positive reception, it would then be moved forward. Hasani is excited about the possibilities: “The real world is all about sequences. Even our perception – you’re not perceiving images, you’re perceiving sequences of images,” he says. “So, time series data actually create our reality.”
So liquid machine learning hasn’t been tested yet?
It hasn’t been out to the wider research community but in tests predicting future values in datasets from atmospheric chemistry to traffic patterns, it bested other top-level time series algorithms. “In many applications we see the performance is really high,” Hasani says, “Everyone talks about scaling up their network. We want to scale down, to have fewer but richer nodes.”
What are nodes in liquid machine learning?
Liquid machine learning is heavily influenced by a microscopic nematode called C. Elegans. “It only has 302 neurons in its nervous system,” Hasani says, “yet it can generate unexpectedly complex dynamics… Just changing the representation of a neuron, you can really explore some degrees of complexity you couldn’t explore otherwise.”
Where next for liquid machine learning?
Hasani continues to improve the system in the hope of implementing it in industrial applications. “We have a provably more expressive neural network that is inspired by nature,” he says, “but this is just the beginning of the process. The obvious question is how do you extend this? We think this kind of network could be a key element of future intelligence systems.