MIT created an AI/ML algorithm last year that can learn and adapt to new information on the job. This was not only during its initial training phase. These “liquid” neural networks(in the Bruce Lee sense) literally play 4D chess — their models requiring time-series data to operate — which makes them ideal for use in time-sensitive tasks like pacemaker monitoring, weather forecasting, investment forecasting, or autonomous vehicle navigation. However, data throughput has become an issue and scaling these systems is prohibitively expensive computationally.
Tuesday’s announcement by MIT researchers was that they had found a way to overcome this restriction. They did not expand the data pipeline, but instead solved a differential equation that has puzzled mathematicians ever since 1907. Specifically, the team solved, “the differential equation behind the interaction of two neurons through synapses… to unlock a new type of fast and efficient artificial intelligence algorithms.”
“The new machine learning models we call ‘CfC’s’ [closed-form Continuous-time] replace the differential equation defining the computation of the neuron with a closed form approximation, preserving the beautiful properties of liquid networks without the need for numerical integration,” MIT professor and CSAIL Director Daniela Rus said in a Tuesday press statement. “CfC models are causal, compact, explainable, and efficient to train and predict. They open the way to trustworthy machine learning for safety-critical applications.”
For those who don’t have a PhD in Really Hard Math, differential equations can be used to describe the state of a system at different points or steps during the process. You can use a differential equation, for example, to find the position of a robot arm as it moves from A to B. These equations can be very time-consuming to solve for each step. MIT’s “closed form” solution end-arounds that issue by functionally modeling the entire description of a system in a single computational step. As the MIT team explains,
Imagine an end-to–end neural network receiving driving input from a camera mounted to a car. The network is then trained to produce outputs like the car’s steering angle. This was solved by the team using liquid neural networks that had 19 nodes. The 19 neurons and a small perception module were enough to drive a car in 2020. Each node in the system is described by a differential equation. With the closed-form solution, if you replace it inside this network, it would give you the exact behavior, as it’s a good approximation of the actual dynamics of the system. This would allow them to solve the problem faster and with fewer neurons.
By solving this equation at the neuron-level, the team is hopeful that they’ll be able to construct models of the human brain that measure in the millions of neural connections, something not possible today. The team also notes that this CfC model might be able to take the visual training it learned in one environment and apply it to a wholly new situation without additional work, what’s known as out-of-distribution generalization. That’s not something current-gen models can really do and would prove to be a significant step towards the generalized AI systems of tomorrow.
Engadget recommends only products that have been reviewed by our editorial staff. This is independent from our parent company. Some stories may contain affiliate links. Affiliate commissions may be earned if you make a purchase through any of the links. All prices correct as of the date of publication.
