Starting with Donald Hebb in the 1940s, neuroscientists have speculated that memories are encoded in the activity of groups of cells, or cell assemblies, that fire in precise sequences of activation. Thus, the memory representation of the title of the book you just read may correspond in the brain to the activation of neuron 7, then neuron 15, then neuron 3, and so on. Instead, the memory of a scene when your aunt greeted you with a hug may be encoded as a neural sequence starting with 15-43-5-7. This encoding scheme has big theoretical advantages as it would enable the storage of a combinatorially huge number of patterns, in a flexible way. Data from recording of large population of neuron is providing first evidence for these sequences to actually occur in the brain. They even occur during sleep, and other “quiet” times, as a “replay” of past experiences, in a process that may help stabilize and reprocess memories.
However, detecting these replays is very difficult, especially because they may reflect memory for an experience far in the past, or other situations for which it is difficult to form a “template” to search for. Research led by Francesco Battaglia and Martin Vinck (Ernst Strungmann Institute, Frankfurt, Germany) have devised a new mathematical procedure for this. They used a measure based on Optimal Transport theory, a branch of mathematics with applications ranging from economics to computer vision to physics, to define a “distance” between neural patterns, enabling the mapping of these patterns into a geometrical space, so that their structure and regularity can be studied. Initial applications to neural data showed that the method could reconstruct, based on very little extra information, which stimulus an animal was looking at.
A new TOP grant funded by the NWO, Exact Sciences department, will allow the team to explore this idea in much greater depth, by increasing its robustness to noise and other perturbation and considerably increase its applicability. The goal is to characterize the geometry of neural activity, crossing the bridge between very different brain states such as sleep and wakefulness, so that the representations of memory in the brain can be identified and visualized.
The new project, at the boundary between mathematics, computer science and neuroscience will fund 3 new positions at the postdoc/PhD student level, with diverse expertise.