Researchers from Georgia tech, Mila, Université de Montréal and McGill University present a training framework and architecture to model neural population dynamics across diverse large-scale neural recordings. It tokenizes individual spikes to capture fine temporal neural activity and employs cross-attention and a PerceiverIO backbone. A large-scale multi-session model is built from data from seven non-human primates with over 27,000 neural units and over 100 hours of recordings. The model demonstrates rapid adaptation to new sessions, enabling low-shot performance on various tasks, showing a scalable approach to neural data analysis.
Their study presents a scalable framework for modeling neuronal population dynamics in diverse large-scale neuronal recordings using Transformers. Unlike previous models that operated on fixed sessions with a single set of neurons, this framework can train across subjects and data from different sources. It leverages PerceiverIO and cross-attention layers to efficiently represent neural events, enabling low-shot performance for new sessions. The work shows the potential of transformers in neural data processing and introduces an efficient implementation to improve calculations.
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 that unites diverse data sets, experiments, and subjects for a more complete understanding of brain function. POYO is a framework that allows efficient training across multiple neural recording sessions, even when dealing with different sets of neurons and without known correspondences. It uses a unique tokenization scheme and the PerceiverIO architecture to model neural activity, showing its transferability and improvements in brain decoding between sessions.
The framework models the dynamics of neural activity in diverse recordings using tokenization to capture temporal details and employ cross-attention architecture and PerceiverIO. A large multi-session model, trained on vast primate data sets, can adapt to new sessions with unspecified neural correspondence for learning in a few sessions. Rotating position inlays improve the attention mechanism of the transformer. The approach uses 5 ms bins for neural activity and has achieved detailed results on benchmark data sets.
The efficiency 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 returned competitive results on the same data set without weight modifications, indicating its versatility. The ability to rapidly adapt to new sessions with unspecified neural correspondence was demonstrated for few-shot performance. The large-scale multi-session model showed promising performance on various tasks, emphasizing the potential of the framework for comprehensive analysis of neural data at scale.
In conclusion, a unified and scalable framework for decoding neural populations offers rapid adaptation to new sessions with unspecified neural correspondence and achieves robust performance on various tasks. The large-scale multi-session model, trained on non-human primate data, shows the potential of the framework for comprehensive analysis of neural data. The approach provides a robust tool to advance neural data analysis and enables training at scale, deepening insights into neural population dynamics.
Review the Paper, and Project. All credit for this research goes to the researchers of this project. Also, don’t forget to join. our 32k+ ML SubReddit, Facebook community of more than 40,000 people, Discord channel, and Electronic newsletterwhere we share the latest news on ai research, interesting ai projects and more.
If you like our work, you’ll love our newsletter.
we are also in Telegram and WhatsApp.
Sana Hassan, a consulting intern at Marktechpost and a dual degree student at IIT Madras, is passionate about applying technology and artificial intelligence to address real-world challenges. With a strong interest in solving practical problems, she brings a new perspective to the intersection of ai and real-life solutions.
<!– ai CONTENT END 2 –>
