Researchers from Georgia Tech, Mila, Université de Montréal, and McGill University present a training framework and architecture for modeling neuronal population dynamics across diverse large-scale neuronal recordings. It tokenizes individual spikes to capture fine-grained temporal neuronal activity and uses cross-attention and a PerceiverIO backbone. A large-scale multi-session model is built using data from seven non-human primates with over 27,000 neuronal units and over 100 hours of recordings. The model demonstrates rapid adaptation to new sessions, enabling performance within a few hits in various tasks, presenting a scalable approach for neural data analysis.
Their study introduces a scalable framework for modeling neuronal population dynamics in various large-scale neuronal recordings using Transformers. Unlike previous models that worked on fixed sessions with a single set of neurons, this framework can train subjects and data from different sources. It leverages PerceiverIO and cross-attention layers to efficiently represent neural events, enabling few-shot performance for new sessions. The work presents the potential of transformers in neural data processing and introduces an efficient implementation for enhanced computations.
Recent advances in machine learning have highlighted the potential for scaling with large pre-trained models like GPT. In neuroscience, there is a demand for a fundamental model to connect diverse data sets, experiments, and topics for a more complete understanding of brain function. POYO is a framework that enables efficient training on diverse neural recording sessions, even when dealing with different sets of neurons and no known matches. It uses a unique tokenization scheme and the PerceiverIO architecture to model neural activity, showcasing its improvements in transferability and brain decoding across sessions.
The framework models the dynamics of neural activity across various recordings using tokenization to capture temporal details and utilizing cross-attention and the PerceiverIO architecture. A large multi-session model, trained on large primate datasets, can adapt to new sessions with unspecified neural matching for learning in just a few steps. Rotational position integrations improve the transformer’s attention mechanism. The approach uses 5ms binning for neuronal activity and has achieved accurate results on benchmark datasets.
The effectiveness of decoding neural activity from the NLB-Maze dataset was demonstrated by achieving an R2 of 0.8952 using the framework. The pre-trained model provided competitive results on the same dataset without weight changes, indicating its versatility. The ability to rapidly adapt to new sessions with unspecified neural matching for performance within a few shots was demonstrated. The large-scale multi-session model showed promising performance in various tasks, highlighting the potential of the framework for comprehensive analysis of large-scale neural data.
In conclusion, a unified and scalable framework for neuronal population decoding provides rapid adaptation to new sessions with unspecified neuronal correspondence and achieves strong performance on diverse tasks. The large-scale multi-session model, trained on data from non-human primates, showcases the framework’s potential for comprehensive neural data analysis. The approach provides a robust tool for advancing neuronal data analysis and enables large-scale training, thereby deepening knowledge of neuronal population dynamics.
Check Paper, And Project. All credit for this research goes to the researchers of this project. Also don’t forget to register our SubReddit 32k+ ML, More than 40,000 Facebook communities, Discord Channel, And E-mailwhere we share the latest AI research news, interesting AI projects and much more.
If you like our work, you will love our newsletter.
We are also on Telegram And WhatsApp.
Sana Hassan, Consulting Intern at Marktechpost and a dual degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-world solutions.
