Home Press Room Awards
Howard Poizner and Team awarded $4.5M ONR MURI grant

Howard Poizner (PI, UCSD), and co-PI's Gary Lynch (UC Irvine) and Terry Sejnowski (Salk and UCSD), together with team leaders Hal Pashler, Sergei Gepshtein, Deborah Harrington, Tom Liu, Eric Hlagren, and Ralph Greenspan were recently awarded a $4.5M ONR MURI grant, with a $3M option period, to study the brain bases of unsupervised learning and training. (October 1, 2009) Dr. Poizner, a Research Scientist in the Institute for Neural Computation, is a member of TDLC, which partly supports his lab, and he is also a member of the Center's Executive Committee.

The study, “How Unsupervised Learning Impacts Training: From Brain to Behavior”, involves the following:

Principal Investigator: Howard Poizner; Co-PI’s: Gary Lunch (UC Irvine) and Terry Sejnowski (Salk and UCSD)
Agency: ONR (Office of Naval Research)
Funding: $4.5M (3yr base period) [started Oct 1, 2009]; $3.0M (2yr option period); $7.5M (5 year total period)

ONR #10124957- Poizner (PI) - $4,500,000 - 10/1/09-9/1/12

The goal of this multidisciplinary grant is to examine the neurobiological, genetic, brain dynamic, and neural circuit correlates of unsupervised learning and training. The proposed studies utilize the new capabilities for creating 3D immersive environments and simultaneous EEG-fMRI recordings recently established through ONR-DURIP grant # N000140811114.

Additional Information
The cerebral cortex is able to create rich representations of the world that are much more than just reinforcement learning and reflexes. Learning is often self-supervised without feedback, a type of learning referred to as unsupervised learning. Such learning, and memory, is (i) commonplace in naturalistic settings, (ii) critical to humans, (iii) encoded by LTP-type mechanisms, and (iv) of direct relevance to computational theories of learning. Using unsupervised learning, an individual builds up internal hierarchical structures and categorizations that model the statistical properties of the environment. These internal representations can be used flexibly and powerfully to acquire new information thereby creating situational
awareness and readiness to act in novel as well as in familiar environments. Yet, unsupervised learning and its neurobiological mechanisms are poorly understood. Our proposed projects will provide new understanding of the neurobiological, genetic, brain dynamic, and neural circuit correlates of this potentially powerful form of learning and training. We propose seven tasks that attack different aspects of the problem making use of parallel paradigms in rodents, flies, and humans. Task 1 maps memory during spatial learning in rats, seeking to uncover the neural engram of memory. Task 2 uses computational modeling to illuminate cortical processes of unsupervised learning in humans. Task 3 conducts studies of training, contrasting the rate and efficiency of both unsupervised and supervised learning. Task 4 explores the brain dynamics of unsupervised learning, using motion capture and virtual environments while recording cortical EEG. Tasks 5 and 6 investigate neuroimaging and genetic correlates of unsupervised learning bringing to bear the new methodology of simultaneous EEG-fMRI recording. Finally, Task 7 exploits the genetic cellular, and behavioral homologies of the fruit fly with humans to study the dopaminergic and genetic regulation of inter-regional coherence associated with learning.


These studies should provide insight into design of the best training environments for our modern military, and increase our understanding of the underlying neurobiological, genetic, brain dynamic, and neural circuit correlates of those environments. Moreover, the studies will open the way to asking if memory enhancing drugs such as ampakines or if particular learning regimens (e.g., extensive experience with diverse environments, short vs. long sessions) change the number and/or distribution of learning-related synaptic modifications and/or the nature of the neural networks and brain dynamics that underlie unsupervised learning. This issue is fundamental to development of mechanism-based strategies for improving
learning and performance in complex environments. Finally, the genetic studies will pave the way for development of individualized training techniques that optimize learning environments.