Scientists at EPFL have developed a novel artificial intelligence method that can identify common patterns in brain activity across different animals performing the same tasks, potentially transforming our understanding of how the brain processes information.
The research, published today in Nature Methods, introduces a new way to interpret complex neural signals and could advance the development of more effective brain-machine interfaces.
The new method, called MARBLE (Manifold Representation Basis Learning), uses advanced mathematics and artificial intelligence to analyze neural recordings in a way that reveals previously hidden similarities in brain activity patterns between different animals as they perform the same tasks.
“Suppose you and I both engage in a mental task, such as navigating our way to work. Can signals from a small fraction of neurons tell us that we use the same or different mental strategies to solve the task?” explains Pierre Vandergheynst, head of the Signal Processing Laboratory LTS2 in EPFL’s School of Engineering. “This is a fundamental question to neuroscience, because experimentalists often record data from many animals, yet we have limited evidence as to whether they represent a given task using the same brain patterns.”
The research team demonstrated MARBLE’s capabilities by analyzing brain recordings from both macaques performing reaching tasks and rats navigating mazes. The results showed that different animals use remarkably similar neural patterns when approaching the same tasks, a finding that had been difficult to prove with previous methods.
Adam Gosztolai, the study’s lead author and now an assistant professor at the AI Institute of the Medical University of Vienna, explains what makes their approach unique: “Inside the curved spaces, the geometric deep learning algorithm is unaware that these spaces are curved. Thus, the dynamic motifs it learns are independent of the shape of the space, meaning it can discover the same motifs from different recordings.”
Traditional deep learning methods have struggled to analyze dynamic systems like neural activity, which changes constantly over time. MARBLE overcomes this limitation by working within curved mathematical spaces that naturally represent complex patterns of brain activity. This approach allows it to identify common elements in neural activity even when recording from different subjects or under varying conditions.
The researchers found that MARBLE’s analysis of neural recordings was significantly more interpretable than existing methods. In practical tests, it showed superior accuracy in decoding arm movements from brain signals compared to current techniques, suggesting potential applications in brain-machine interfaces for assistive devices.
While the immediate applications focus on neuroscience, Vandergheynst emphasizes the broader potential of their work. “The MARBLE method is primarily aimed at helping neuroscience researchers understand how the brain computes across individuals or experimental conditions, and to uncover – when they exist – universal patterns,” he says. “But its mathematical basis is by no means limited to brain signals, and we expect that our tool will benefit researchers in other fields of life and physical sciences who wish to jointly analyze multiple datasets.”
The breakthrough comes at a crucial time in neuroscience research, as scientists increasingly seek to understand how different individuals’ brains process information and whether there are universal patterns in neural computation. This understanding could have far-reaching implications for treating neurological conditions and developing more effective brain-computer interfaces.
By providing a mathematical framework to compare neural activity across different subjects and conditions, MARBLE opens new possibilities for understanding the fundamental principles of brain function. The method’s ability to decode complex neural patterns could accelerate the development of advanced prosthetics and other assistive technologies that rely on interpreting brain signals.
The research demonstrates that despite individual variations in brain structure and neural recordings, there are underlying commonalities in how different brains process information during similar tasks. This finding suggests a level of universality in neural computation that could have profound implications for our understanding of cognition and consciousness.