Motor neural null space


The investigation of neural populations dynamics has shed light on the existence of low-dimensional spaces capable of explaining most of the neural activity. These spaces are now commonly called neural manifolds. Neural manifolds are in general identified through optimization techniques based on dimensionality reductions that can be both linear and non-linear. These insights allow us to understand how similar patterns of activity (neural modes) are used to generate different output in terms of behavioral response. Some of the strongest results obtained through this approach include the finding that preparatory movement activity lies on an orthogonal plane with respect to executory activity, the fact that a single consistent neural manifold can explain different hand/fingers movements or yet the overlapping between a neural manifold spanning executory activity with the manifolds explaining observed or imagined movements.The identification of such low-dimensional manifolds not only help elucidate neural correlates of behavior, but it also holds great translational potential in what it can help the development of innovative brain-machine interfaces (BMIs) both in the frames of motor restoration and motor augmentation. However, most of these findings have relied on multi-unit neural activities, often recorded in non-human primates. To increase the translational.

Previous findings in the literature have shown that neural activity linked to motor preparation and to motor execution lies on two orthogonal spaces. We would like to investigate this aspect at the mesoscale level, that is with Local Field Potential (LFP) activity recorded via stereo-electroencephalography (sEEG) and electrocorticography (ECoG).Our data consists of intracortical recordings obtained in two different tasks. In both tasks, two different movements were performed (two types of grasps in one case, a reaching and a wrist extension movements in the second case).
  • Epoching of data
  • Filtering and spectral analysis of signals
  • Performing dimensionality reduction techniques to identify neural manifolds
  • Analyzing the variance captured by these manifolds (within and across manifolds)
  • Using machine learning models to decode the movement (prepared or executed) after projection onto the low-dimensional manifolds
  • Investigate the neural trajectories of different movements after projection of neural data onto the low-dimensional manifolds
  • Validations of results through statistical analysis
  • Good level in programming on Python
  • Theoretical knowledge of basic dimensionality reduction methods (such as PCA)
  • Experience with machine learning methods (Random Forests, SVM, LDA, …) for classification
  • (Previous experience in analyzing neural data is a plus.)

[email protected]

Best for:
Master’s thesis/Semester’s project

Campus Biotech (Geneva). It’s possible to perform the project remotely.

  • S. Saxena and J. P. Cunningham, “Towards the neural population doctrine,” Current Opinion in Neurobiology, 2019
  • J. A. Gallego et al., “Neural Manifolds for the Control of Movement,” Neuron, 2017
  • J. P. Cunningham and B. M. Yu, “Dimensionality reduction for large-scale neural recordings,” Nat Neuroscience, 2014
  • M. T. Kaufman et al., “Cortical activity in the null space: permitting preparation without movement,” Nat Neuroscience, 2014


If none of the projects suit you but you are interested in motor neural data in general, please feel free to contact us to discuss potential opportunities.

Leonardo Pollina ([email protected])