'Deep faking the mind' might enhance brain-computer interactions for those with impairments.

Health and Fitness Hits

For persons with impairments, USC Viterbi School of Engineering researchers has found a way to enhance brain-computer interactions by employing generative adversarial networks (GANs).

Generative adversarial networks (GANs) are computational frameworks that pit two neural networks against one another in order to produce new, synthetic instances of data that seem to be genuine. Image, video, and speech production are all common uses for them.

It was reported in Nature Biomedical Engineering that scientists have trained artificial intelligence (AI) to produce synthetic data on the brain's activity. Brain-computer interfaces may benefit from machine-learning methods based on neural signals known as spike trains (BCI).

A BCI system analyzes brain impulses and converts them into instructions, enabling users to operate digital devices like computer cursors only by mental commands. People with motor dysfunction or paralysis, even those with locked-in syndrome—when a person is fully conscious but unable to move or communicate—may benefit from these gadgets, enhancing their quality of life. There are a variety of BCI devices available, from caps that monitor brain impulses to devices that are implanted in the brain. From neurorehabilitation to the treatment of depression, new use cases are being discovered all the time. However, despite all of this potential, making these systems quick and robust enough for real-world applications has been challenging.

BCIs, in particular, need massive volumes of brain data and extended periods of training, calibration, and learning to make sense of their inputs.

Laurent Itti, a computer science professor, and research co-author noted that obtaining enough data for BCI algorithms may be difficult, costly, or even impossible if paralyzed persons cannot create enough strong brain signals.

Moreover, the technology is user-specific and must be trained from the start for each individual. As an alternative to obtaining real-world data, how about creating synthetic data—artificially computer-generated data—that may "serve as a substitute?"

Generative adversarial networks are the next frontier. GANs are well-known for their ability to produce "deep fakes," or pictures that are close to the original but not the same.

Shixian Wen, the study's lead author and a Ph.D. student mentored by Itti, wondered whether GANs might also provide training data for BCIs by producing synthetic neurological data indistinguishable from the real thing.

According to the paper, one session of data from a monkey reaching for an item was used to train a deep learning spike synthesizer. A vast quantity of identical, if fictitious, brain data were then generated using the synthesizer. That information was then sent into a BCI, which was subsequently trained using actual data from the same monkey on a different day or another animal. This strategy was able to get the system up and operating significantly more quickly than traditional procedures. According to the researchers, the overall training speed of a BCI increased up to 20 times using Gan-synthesized neural data.

In the words of Wen, "less than a minute's worth of actual data mixed with the synthetic data works as well as 20 minutes of genuine data"

"Using artificial spike trains, we've seen AI for the first time devise a blueprint for cognition or movement. This study is an important first step in making BCIs more practical in everyday life."

In addition, the system quickly responded to new sessions or people utilizing little extra brain data after training on one experimental session.

Creating artificial spike trains that seem like they came from this individual when they anticipate executing specific actions and then utilizing this data to help learn the next person is the key novelty here, according to Itti.

In addition to BCIs, synthetic data created by GANs might lead to advancements in other data-hungry fields of artificial intelligence by speeding up training and boosting performance.

Reference:

Wen, S., Yin, A., Furlanello, T. et al. Rapid adaptation of brain-computer interfaces to new neuronal ensembles or participants via generative modeling. Nat Biomed Eng (2021). https://doi.org/10.1038/s41551-021-00811-z

Comments / 0

Published by

First Year Medicine student

Los Angeles, CA
1105 followers

More from Health and Fitness Hits

Comments / 0