Closed-Loop Multisensory Brain-Computer Interface for Enhanced Decision Accuracy
Lead Research Organisation:
University College London
Department Name: Gatsby Computational Neuroscience Unit
Abstract
Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
Publications
Baldassarre L
(2017)
Sparsity Is Better with Stability: Combining Accuracy and Stability for Model Selection in Brain Decoding.
in Frontiers in neuroscience
Carlo Cliberto
(2017)
Consistent Multitask Learning with Nonlinear Output Relations
Ciliberto C
(2023)
Reexamining low rank matrix factorization for trace norm regularization
in Mathematics in Engineering
Ciliberto C
(2017)
Connecting YARP to the Web with Yarp.js
in Frontiers in Robotics and AI
Ciliberto C.
(2017)
Consistent multitask learning with nonlinear output relations
in Advances in Neural Information Processing Systems
Combettes P
(2018)
Learning with optimal interpolation norms
in Numerical Algorithms
Denevi G.
(2020)
The advantage of conditional meta-learning for biased regularization and fine tuning
in Advances in Neural Information Processing Systems
Denevi G.
(2019)
Online-within-online meta-learning
in Advances in Neural Information Processing Systems
Giulia Denevi
(2020)
The Advantage of Conditional Meta-Learning for Biased Regularization and Fine-Tuning
| Description | We have been both analyzing data and and developing theories relevant to multi-sensory integration. First, the data analysis. To perform optimal multi-sensory integration, the brain must have some notion of uncertainty -- that allows it to give more weight to more certain cues. We have spent the last year looking for signatures of uncertainty in data supplied by Anne Churchland's lab. On the theoretical front, we have been investigating how networks of neurons could perform the click task. We believe this is especially relevant, as the click task has a feature common to high level information processing: all the information is available, but subjects don't perform this task perfectly. Our goal is to understand why. |
| Exploitation Route | Although our work has been mostly theoretical, with additional effort theory can be brought into practice and meta-learning can have broad applications. We indeed started to explore in collaboration with Essex some of these applications, in the context of the project. We also aim to develop opensource software implementation of our algorithms. |
| Sectors | Other |