DARPA agency finds help in his brand new project that will allow computers to simulate human neocortex — the section of the brain that we use to comprehend language, reasoning, and perception.
Although we already have a machine-learning techniques that allow computers to make decisions based on constantly changing disk imaging, DARPA wants more. Military wish to obtain a «model of cortical computation,» which will allow computers to recognize new data when they are relevant to their current task, and adapt the latest information from their environment. And although the «cortical processors» sounds like something out of a sci-fi movie 80s, they — it’s specifically what will be required to bot or drone to gather intelligence in the field criteria and decision-making on the basis of what he vyznat.
Here is what the agency itself DARPA:
Although a thorough understanding of the principles of human cortex today lies beyond our knowledge, we, nevertheless, have the ability to identify and put in machine learning and neural networks basic algorithmic principles. Methods inspired neural models, namely, the neocortex, can recognize overarching spatial and temporal patterns, and can adapt to changing environmental conditions. As a consequence, these methods represent very promising approach to filtering and processing of data streams, and can provide a new level of performance problems in determination of information. Cortical model calculations should be resistant to errors caused by gaps in the available data to be parallel, the maximum energy-efficient, scalable and simple. It is also required to have small requirements for arithmetical accuracy, and ensure the ultra-dense, low-energy embodiments.