Artificial intelligence can now solve advanced math, perform sophisticated reasoning, and even operate personal computers, but today’s algorithms can still learn a thing or two from microscopic bugs.
Fluid artificial intelligencean MIT startup will today present several recent artificial intelligence models based on a novel type of “fluid” neural network that can be more productive, less energy-intensive and more clear than those on which all chatbots, image generators, facial recognition systems are based .
Liquid AI’s recent models include one for detecting fraud in financial transactions, another for controlling autonomous cars and a third for analyzing genetic data. The company showed off recent models it licenses to third-party companies at an event held today at MIT. The company has received funding from investors that include Samsung and Shopify, which are also testing its technology.
“We are expanding our business,” he says Ramin Hasanico-founder and CEO of Liquid AI, who co-invented Liquid Networks as an MIT student. Hasani’s research drew inspiration from C. elegancea millimeter-long worm usually found in soil or rotting vegetation. The worm is one of the few creatures whose nervous system has been completely mapped, and despite having only a few hundred neurons, it is capable of extremely sophisticated behavior. “It used to be just a research project, but the technology is fully commercialized and fully ready to provide value to enterprises,” Hasani says.
Inside a regular neural network, the properties of each simulated neuron are defined by a inert value or “weight” that influences its firing. IN fluid neural networkthe behavior of each neuron is governed by an equation that predicts its behavior over time, and the network solves a cascade of related equations as the network operates. The design makes the network more productive and pliant, allowing it to learn even after training, unlike a conventional neural network. Fluid neural networks are also amenable to inspection unlike existing models because their behavior can essentially be rewound to see how it generated the result.
In 2020, researchers showed that such a network containing just 19 neurons and 253 synapses, which is extremely petite by state-of-the-art standards, could control a simulated autonomous car. While a regular neural network can only analyze visual data at inert intervals, a fluid network is very effective at capturing how visual information changes over time. In 2022, the founders of Liquid AI I came up with a shortcut this made the mathematical work needed to create fluid neural networks practical.