Researchers develop AI system that mimics biological models to function with small number of neurons

A team of researchers from the Massachusetts Institute of Technology (MIT), the Technische Universitat Wien (TU Wien), and the Institute of Science and Technology Austria (IST Austria) have developed an artificial intelligence (AI) system able to control a vehicle using only 19 artificial neurons. The system – presented in the Nature Machine Intelligence journal – is inspired by the brains of very small animals such as threadworms and is composed of two parts. A convolutional neural network processes input from a camera, decides which parts of the camera image are important, and transmits signals to the core element – a neural circuit policy control system (consisting of only 19 neurons) that steers a vehicle. The researchers found that the system achieved comparable results with previous AI models in terms of training and test errors. They argue that the major advantages of their proposed model are interpretability and robustness, plus the fact that it can reduce training time and make it possible to implement AI in relatively simple systems.

Related issues