Award Abstract # 2024526
NCS-FO: Identification and control of neural cognitive systems

NSF Org: IIS
Division of Information & Intelligent Systems
Recipient: UNIVERSITY OF CALIFORNIA, DAVIS
Initial Amendment Date: September 10, 2020
Latest Amendment Date: October 13, 2020
Award Number: 2024526
Award Instrument: Standard Grant
Program Manager: Kenneth Whang
IIS
 Division of Information & Intelligent Systems
CSE
 Directorate for Computer and Information Science and Engineering
Start Date: October 1, 2020
End Date: September 30, 2024 (Estimated)
Total Intended Award Amount: $995,777.00
Total Awarded Amount to Date: $995,777.00
Funds Obligated to Date: FY 2020 = $995,777.00
History of Investigator:
  • Jochen Ditterich (Principal Investigator)
    jditterich@ucdavis.edu
  • Zhaodan Kong (Co-Principal Investigator)
Recipient Sponsored Research Office: University of California-Davis
1850 RESEARCH PARK DR STE 300
DAVIS
CA  US  95618-6153
(530)754-7700
Sponsor Congressional District: 04
Primary Place of Performance: University of California-Davis
CA  US  95618-6134
Primary Place of Performance
Congressional District:
04
Unique Entity Identifier (UEI): TX2DAGQPENZ5
Parent UEI:
NSF Program(s): IntgStrat Undst Neurl&Cogn Sys
Primary Program Source: 01002021DB NSF RESEARCH & RELATED ACTIVIT
Program Reference Code(s): 8089, 8091, 8551
Program Element Code(s): 862400
Award Agency Code: 4900
Fund Agency Code: 4900
Assistance Listing Number(s): 47.070

ABSTRACT

Implantable devices provide an alternative for treating neurological disorders in patients who do not respond well to available drugs. A good example are deep brain stimulators for the treatment of Parkinson?s Disease, which have been very successful. Many neurological and psychiatric disorders are associated with cognitive deficits, which tend to be difficult to ameliorate through drug therapy. In the future, it might be possible to treat cognitive deficits with implantable devices, but this requires much more advanced technology than the currently available stimulators. This project will lay the foundation for interfacing technology with neural cognitive systems. Being able to steer or control a system of any kind requires a model or mathematical description of the system. The project will establish a framework for deriving such a model directly from neural activity. It will further develop strategies for moving the cognitive system into a desired target state through the application of electrical microstimulation, with the intent to allow for correction of disease-related maladaptive states.

This interdisciplinary project, which involves investigators from both neuroscience and engineering, has the following components: 1) System identification: Using a large number of implanted electrodes, the recorded neural activity will be used to identify a dynamical system that is able to capture (and predict) the temporal pattern of neural activity. The focus will be on piecewise linear models for approximating nonlinear dynamics. 2) State decoding: With the help of the identified model and based on the current and recent activity pattern, the current internal state of the neural system can be identified. 3) System control: In the case of the current state of the system deviating from a desired target state, the identified model in combination with control-theoretic strategies can be used to determine a stimulation pattern that needs to be applied to drive the system towards the desired state. The project will use simple cognitive tasks, like working memory and perceptual decision-making tasks, to provide a proof of principle that the state of a cognitive neural system can be adjusted in real time to improve, for example, overall performance on the task. This is an ambitious goal, which, in this form, has not been achieved before, but a successful outcome has the potential to revolutionize how mental disorders might be treated in the future. This work is supported by the Integrative Strategies for Neural and Cognitive Systems (NCS) Program.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

PROJECT OUTCOMES REPORT

Disclaimer

This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.

Deep Brain Stimulation (DBS) has been successful for treating motor disorders like Parkinson’s disease. The approach relies on placing a stimulation electrode at a particular location in the brain and providing a continuous stimulation pulse train, which has an effect that resembles creating a local lesion, i.e., disrupting information processing at that location. This approach works for counteracting relative overactivity in a particular pathway (like in the case of Parkinson’s disease), but likely is not applicable for all disorders. It would be desirable to have implantable stimulators available as a treatment option for cognitive deficits resulting from neurological and psychiatric diseases.

The brain is a complex dynamical system. How does an engineer approach controlling a complex dynamical system like, e.g., an airplane? By developing a mathematical model of the dynamics of the system, taking measurements using multiple sensors to be able to estimate the current state of the system, and using model-based control to decide what input needs to be provided to the system using multiple actuators to change its state from an undesired to a desired one. Could a similar approach work for the brain? It would require learning a suitable dynamical system model from observed brain activity (using a larger number of electrodes) and using this model to estimate the current state from neural activity and to determine which stimulation patterns (again, using a larger number of electrodes) have to be applied to change the state in a desired way.

The dynamical system models that have primarily been used for estimating brain states so far were either linear dynamical system (LDS) models, which are desirable from a mathematical point of view, but cannot capture any nonlinearities, which are expected to be present in the brain, or nonlinear models that are based on artificial neural networks, which are undesirable from a control point of view. In the context of controlling nonlinear technical systems, engineers have successfully taken advantage of piecewise-linear models, which are still desirable from a mathematical point of view, but can also approximate nonlinearities. We were therefore investigating whether such models are also useful in the context of capturing brain dynamics.

Starting with a nonlinear model of perceptual decision-making, we first generated synthetic neural data to probe whether piecewise-linear models could be learned from these observations and how their performance compared to linear models. The results indicated that piecewise-linear models could be estimated from the synthetic neural data, that they provided a better explanation for the neural observations than linear models, and that they were able to make better predictions how the observations (and the underlying brain state) would evolve in the near future, which is the important aspect for control.

Applying the approach to real neural data had different results depending on the dataset. When analyzing a publicly available dataset with neural activity that was recorded from frontal cortex during perceptual decision-making, we did not find a clear benefit of using piecewise-linear models over linear ones. The observation that different time periods of the decision process were best described by different linear models suggested that this dataset might be better captured by a time-variant linear model than by a time-invariant piece-wise linear model.

Using our own neural data, recorded from chronic electrode arrays in frontal cortex during different tasks, including working-memory and perceptual decision-making tasks, the results so far suggest that piecewise-linear models outperform linear ones in at least some of the task contexts. An implantable brain stimulator should be applicable regardless of what problem a subject currently solves or what goal a subject currently pursues. To address the question how task-specific neural dynamics are, we estimated separate models from the data collected in the different task contexts and used them to predict neural activity from either the same or different task contexts. The results so far suggest that the best predictions were always made by the model that was learned from data that were collected in the same task context, but the second best model was a hybrid model that was learned from pooled data from the different task contexts. Learning a task-agnostic hybrid dynamical system model that would be good enough to make useful predictions for different task contexts therefore seems possible.

Future work will have to show how these models can be combined with model-based control and closed-loop stimulation to successfully modulate brain states.


Last Modified: 01/28/2025
Modified by: Jochen Ditterich

Please report errors in award information by writing to: awardsearch@nsf.gov.

Print this page

Back to Top of page