Past Recording
Share
Star()
New methods for identifying latent manifold structure from neural data
Tuesday May 19 2020 16:00 GMT
Please to join the live chat.
New methods for identifying latent manifold structure from neural data
Why This Is Interesting

Numerous studies in neuroscience posit that large-scale neural activity reflects noisy high-dimensional observations of some underlying, low-dimensional signals of interest. Discovering such low-dimensional signals or structures can help shed light on how information is encoded at the population level, and provide significant scientific insight into the brain. In this talk, Dr. Anqi Wu will present her work on developing Bayesian methods to identify such latent manifold structures.

Full abstract: Numerous studies in neuroscience posit that large-scale neural activity reflects noisy high-dimensional observations of some underlying, low-dimensional signals of interest. Discovering such low-dimensional signals or structures can help shed light on how information is encoded at the population level, and provide significant scientific insight into the brain. In this talk, I will present my work on developing Bayesian methods to identify such latent manifold structures, which are referred to as latent manifold tuning models. Firstly, I will describe the latent manifold tuning model to discover low-dimensional latent dynamics in multi-neuron spike train data with an application to hippocampal place cells. Secondly, I will present a similar latent model to learn interpretable latent embeddings in calcium imaging data with an application to olfactory neurons. We show that the models are able to reveal the underlying signals of neural populations as well as uncovering interesting topography of neurons where there is a lack of knowledge and understanding about the brain.

Discussion Points

-Capturing low-dimensional latent (hidden) phenomena from high-dimensional recorded neural data -Nonlinearity and the need for non-linear tools for neurological signal processing and analysis -Latent manifold tuning models and their application in hippocampal place cells & calcium imaging data

Takeaways

-Nonlinear neural data can be modeled in an explainable manner using GPLVM/LMT -Latent manifold tuning benefits from domain knowledge, but provides the opportunity and data to expand beyond what is known with typical methods -It is often necessary to create new method (GPLVM/LMT) to apply some constraints that are required in the data (e.g., continuity of GPLVM/LMT vs. Bayesian Neural Networks) -A great example of how to interpret high-dimensional, noisy neural data over time (e.g., in hippocampal place cells)

Time of Recording: Tuesday May 19 2020 16:00 GMT