New Brain-Computer Interface Transforms Thoughts to Images
University of Helsinki uses AI machine learning to imagine what you’re thinking.
Posted Sep 27, 2020
Achieving the next level of brain-computer interface (BCI) advancement, researchers at the University of Helsinki used artificial intelligence (AI) to create a system that uses signals from the brain to generate novel images of what the user is thinking and published the results earlier this month in Scientific Reports.article continues after advertisement
“To the best of our knowledge, this is the first study to use neural activity to adapt a generative computer model and produce new information matching a human operator’s intention,” wrote the Finnish team of researchers.
The brain-computer interface industry holds the promise of innovating future neuroprosthetic medical and health care treatments. Examples of BCI companies led by pioneering entrepreneurs include Bryan Johnson’s Kernel and Elon Musk’sNeuralink.
Studies to date on brain-computer interfaces have demonstrated the ability to execute mostly limited, pre-established actions such as two-dimensional cursor movement on a computer screen or typing a specific letter of the alphabet. The typical solution uses a computer system to interpret brain-signals linked with stimuli to model mental states. Seeking to create a more flexible, adaptable system, the researchers created an artificial system that can imagine and output what a person is visualizing based on brain signals. The researchers report that their neuroadaptive generative modeling approach is “a new paradigm that may strongly impact experimental psychology and cognitive neuroscience.”
The University of Helsinki researchers used a combination of a generative neural network with neuroadaptive brain interfacing to create a new BCI paradigm. Neuroadaptive generative modeling is the estimation of a person’s intentions via adapting a generative model to neural activity. To expand capabilities and not be limited to pre-defined categories, the researchers based the solution on a generative adversarial network (GAN) to generate novel information from a latent representation of an input space.
Generative adversarial networks are a relatively recent innovation in artificial intelligence machine learning where two artificial neural networks simultaneously train one another by competing. Backpropagation is applied to the dueling neural networks. GANs enable machines to imagine and create their own novel images. Brain activity is used to adjust the latent space to provide an unlimited output of possible samples.article continues after advertisement
In this study, 31 participants were instructed to passively watch images and mentally focus on the images that match certain criteria as their brain activity was recorded by non-invasive electroencephalogram (EEG). The participants were tasked to complete eight facial category recognition tasks by focusing on faces that matched the categories of smiling, not smiling, blond or dark hair, young, old, female, or male. The brain responses were divided by participant into training and testing data.
A classifier separated the brain responses in the testing set based on the criteria of relevant or irrelevant images. The vectors of the relevant images where fed to the intention model which generated a visualization of the mental target. Thus, the participant’s neural reactions parameterize an intentional model that can be used to generate novel images to illustrate a person’s perceptual categories. Then the computer-generated images were evaluated by the participants for validation.
“Our experiment provided strong evidence that neuroadaptive modelling is highly effective in generating previously non-existing information matching the human operator’s intended perceptual categories,” the researchers wrote.
The worldwide brain-computer interface market is projected to grow over the next seven years at a compound annual growth rate (CAGR) of 15.5 percent to reach of USD 3.7 billion in revenue by 2027 according to Grand View Research. Brain-computer interface advancements may one day help treat a variety of brain disorders and diseases such as dementia, epilepsy, paralysis, Alzheimer’s disease, Parkinson’s disease, and sleep disorders. With each new innovative discovery, science breaks existing limits with novel paradigms in hopes of a better future ahead.