MIT operates in the fluid neural network

Artificial intelligence has grown exponentially in recent years Deep learning Based on learning Neural networks. To adapt to different conditions, the Massachusetts Institute of Technology Has developed a new genre in the United States (MIT) “Fluid” neural network, Can change its own parameters Equations.

In the AI ​​training phase, the neural network is used to generate algorithms by processing large amounts of data. This new approach takes into account chronological data for processing scenes, not fixed points. The researchers were impressed நூற்புழு Gynecomastia elegans Consisting of only 302, including the nervous system Neurons, Capable of complex operations.

Artificial intelligence suitable for data

Neural network Liquid uses Derivative equations The nest is built to change the parameters of the equations over time, which makes it more flexible. This allows for better processing of noisy data such as the video stream of an autonomous car when it is raining. Neural networks are compared to black boxes where inputs and outputs can be observed, but not what is happening inside. The ability to change these equations allows researchers to better understand how the network works, which is generally not possible. In addition, this approach uses a lower number of neurons than conventional networks, which reduces the required computing power.

The researchers say that fluid neuronal network improvement should be implemented in all areas where conditions can change rapidly. This includes Autonomous cars, Robot control, natural language processing, automatic Diagnostic Medical or video processing.