Researchers develop materials that can recognize sound and can be applied to sensors of autonomous vehicle

Various forms of pattern recognition, especially machine learning, have been widely applied in various modern technologies. By discovering patterns in the data, machine learning can transform such patterns into understandable information, such as identifying objects on the path of autonomous vehicle, or identifying the thermal characteristics of intruders.

In a similar way to machine learning, an emerging framework called 'reservoir computing' can be used to learn patterns from data. The key to this form of calculation is to use a complex system that repeatedly calculates information. All of this can be accomplished through a recurrent neural network (RNN) that utilizes a dynamical system or "reserve pool" to perform pattern recognition tasks.

However, traditional RNNs are composed of a large number of independent, interconnected nodes or neurons with multiple parameters, which can make such systems clumsy, difficult to train, and even result in high energy consumption, thereby hindering their applications. Reserve pool computation simplifies this by randomly selecting parameters and only training output nodes, greatly reducing complexity.

Recent research has shifted towards using physical foundations capable of performing pattern recognition tasks for reservoir computation. As long as such materials are sufficiently complex, nonlinear (input changes may not necessarily be the same as output changes), and have gradually weakening memory characteristics, that is, the system's ability to recall past input signals gradually weakens over time, it can function as a reserve pool.