Motivation, Learning, and Memory: A System-Level Brain Modeling Approach Lund, Sweden
December 5-6, 2005
Scott Makeig , Swartz Center for Computational Neuroscience, Institute for Neural Computation, University of California San Diego, La Jolla, CA
Viewing event-related brain dynamics from the top down.
Increasing evidence as well as simple logic point to the predominance of 'top-down' over 'bottom-up' roles for macroscopic field potentials and BOLD signals recorded non-invasively from the human brain. Unfortunately, standard analysis techniques based on response-locked averaging are ideal only for modeling stimulus time- and phase-locked bottom-up brain 'impulse' responses. Furthermore, response averaging methods both proceed from and force the assumption that the set of events used to select the data epochs to be averaged must evoke or induce equivalent brain respones.
A quite different point of view leads to a different analytic approach, It begins with the assumption that the ultimate and actual purpose of brain activity is to prepare the subject (or organism) to respond appropriately to the anticipated consequences of events, while actively guiding perception to best anticipate those consequences. The role of macroscopic brain activity in active 'top-down' perception involves active moment-to-moment regulation of the distribution of attention, both among sensory modalities and between sensory and associative channels, within a moving 'present context' sustained by a porrly understood contextual or working 'memory' system. From this point of view, each experimental event in any event-related task paradigm poses a fresh and potentially unique challenge. Superficially identical events (e.g. a 'target') in a behavioral task paradigm arrive with different perceptual and behavioral contexts, and the brain cannot afford to wait to build an average to respond with a renewed or shifted attentional focus and motor plan.
The challenge for computational cognitive neuroscience is, therefore, to develop analysis methods capable of discovering relationships between high-dimensional brain activity recordings and the significance-in-context of both sensory events and subject actions. The growing subject of information-based signal processing suggests a promising approach to this problem that I will illustrate with simple examples. The consequences of this approach may lead to better understanding of integrative ('top-down') brain function -- understanding unlikely to develop from analysis of microscale recordings alone. These methods could also be more readily appllied to buildng soon-feasible clinical and workplace brain monitoring systems.