Neuroscience Enables AI to Reach New Accuracy Milestone

A new training model aimed at mimicking the way the human brain learns has given AI the capacity to recognize unfamiliar or out-of-distribution (OOD) data, an important milestone in improving the accuracy of this technology, according to Fujitsu Limited and MIT. This judgement ability resembles human-like flexibility and could theoretically enable commercial applications that require the agility to respond to changes along a spectrum of observation conditions.

“This achievement marks a major milestone for the future development of AI technology that could deliver a new tool for training models that can respond flexibly to different situations and recognize even unknown data that differs considerably from the original training data with high accuracy, and we look forward to the exciting real-world possibilities this opens up” said Seishi Okamoto, Fellow, Fujitsu Limited.

While the advent of deep neural networks (DNNs) has enabled AI models to demonstrate performance equivalent and even greater to that of humans, its recognition accuracy tends to deteriorate as environmental conditions, such as lighting and perspective, begin to deviate from its training process. In recognition of this problem, researchers were tasked with enabling AI to recognize OOD data in differing and evolving environmental circumstances. This process drew inspiration from neuroscience, specifically how humans synthesize information in development, usually through digestible units like shape and color. This cognitive teaching method has allowed AI to achieve highest recognition accuracy when measured against the “CLEVR-CoGenT” benchmark.

This milestone comes at a time when concern over unintended dataset bias, which can lead to risks ranging from privacy infringement to ethnic discrimination, mounts. As outlined by the World Economic forum, it is imperative for C-suite executives to consider and manage the possible risks associated with AI, this discovery—at the very least—is a starting point.

“There is a significant gap between DNNs and humans when evaluated in out-of-distribution conditions, which severely compromises AI applications, especially in terms of their safety and fairness. Research inspired by neuroscience may lead to novel technologies capable of overcoming dataset bias. The results obtained so far in this research program are a good step in this direction,” said Tomaso Poggio, MIT Professor and Director of Center for Brains, Minds and Machines (CBMM).