Integration of inputs by cortical neurons provides the basis for the complex information processing performed in the cerebral cortex. Here, we propose a new analytic framework for understanding integration within cortical neuronal receptive fields. Based on the synaptic organization of cortex, we argue that neuronal integration is a systems--level process better studied in terms of local cortical circuitry than at the level of single neurons, and we present a method for constructing self-contained modules which capture (nonlinear) local circuit interactions. In this framework, receptive field elements naturally have dual (rather than the traditional unitary influence since they drive both excitatory and inhibitory cortical neurons. This vector-based analysis, in contrast to scalarsapproaches, greatly simplifies integration by permitting linear summation of inputs from both "classical" and "extraclassical" receptive field regions. We illustrate this by explaining two complex visual cortical phenomena, which are incompatible with scalar notions of neuronal integration.

Baylis & Driver (Nature Neuroscience, 2001) have recently presented data on the response of neurons in macaque inferotemporal cortex (IT) to various stimulus transformations. They report that neurons can generalize over contrast and mirror reversal, but not over figure-ground reversal. This finding is taken to demonstrate that ``the selectivity of IT neurons is not determined simply by the distinctive contours in a display, contrary to simple edge-based models of shape recognition'', citing our recently presented model of object recognition in cortex (Riesenhuber & Poggio, Nature Neuroscience, 1999). In this memo, I show that the main effects of the experiment can be obtained by performing the appropriate simulations in our simple feedforward model. This suggests for IT cell tuning that the possible contributions of explicit edge assignment processes postulated in (Baylis & Driver, 2001) might be smaller than expected.

In a recent experiment, Freedman et al. recorded from inferotemporal (IT) and prefrontal cortices (PFC) of monkeys performing a "cat/dog" categorization task (Freedman 2001 and Freedman, Riesenhuber, Poggio, Miller 2001). In this paper we analyze the tuning properties of view-tuned units in our HMAX model of object recognition in cortex (Riesenhuber 1999) using the same paradigm and stimuli as in the experiment. We then compare the simulation results to the monkey inferotemporal neuron population data. We find that view-tuned model IT units that were trained without any explicit category information can show category-related tuning as observed in the experiment. This suggests that the tuning properties of experimental IT neurons might primarily be shaped by bottom-up stimulus-space statistics, with little influence of top-down task-specific information. The population of experimental PFC neurons, on the other hand, shows tuning properties that cannot be explained just by stimulus tuning. These analyses are compatible with a model of object recognition in cortex (Riesenhuber 2000) in which a population of shape-tuned neurons provides a general basis for neurons tuned to different recognition tasks.

Charting out the complete brain microstructure of a mammalian species is a grand challenge. Recent advances in serial sectioning microscopy such as the Knife- Edge Scanning Microscopy (KESM), a high-throughput and high-resolution physical sectioning technique, have the potential to finally address this challenge. Nevertheless, there still are several obstacles remaining to be overcome. First, many of these serial sectioning microscopy methods are still experimental and are not fully automated. Second, even when the full raw data have been obtained, morphological reconstruction, visualization/editing, statistics gathering, connectivity inference, and network analysis remain tough problems due to the unprecedented amounts of data. I designed a general data acquisition and analysis framework to overcome these challenges with a focus on data from the C57BL/6 mouse brain. Since there has been no such complete microstructure data from any mammalian species, the sheer amount of data can overwhelm researchers. To address the problems, I constructed a general software framework for automated data acquisition and computational analysis of the KESM data, and conducted two scientific case studies to discuss how the mouse brain microstructure from the KESM can be utilized. I expect the data, tools, and studies resulting from this dissertation research to greatly contribute to computational neuroanatomy and computational neuroscience.

Modern neuron models are large, complex entities. The ability to better simulate these complex models has been iven by the development of ever more powerful and cheap computers. The capacity to manage and understand the models has lagged behind improvements in simulation ability almost from the inception of neuron modeling. Despite the computing power currently available, more powerful simulation platforms and strategies are needed to cope with current and next generation neuron models. This thesis develops methodologies aimed at better characterizing motoneuron models. The hypothesis presented is that relationships between model outputs in addition to the relationships between model inputs (parameters) and outputs (behaviors) provide a characteristic description of the model that describes the model in a more useful way than just model behaviors. This description can be used to compare a model to different implementations of the same motoneuron and to experiment data. Data mining and data reduction techniques were used to process the data. Principal component analysis was used to indicate a significant, consistent reduction in dimensionality in an intermediate, mechanistic layer between model inputs and outputs. This layer represents the non-linear relationships between input and output, implying that if the non-linear relationships of a model were better understood and accessible, a model could be manipulated by varying the mechanism layer members, or rather the model parameters that primarily affect a mechanism layer member. Hierarchical cluster analysis showed similarity between sensitivity analyses data from models with random parameter sets. A main cluster represented the main region of model behavior with outlying clusters representing non-physiological behavior. This indicates that sensitivity analysis data is a good candidate for a model signature. The results demonstrate the usefulness of cluster analysis in identifying the similarities between data used as a model characterization metric or model signature. Its application is also valuable in identifying the main region of useful activity of a model, thus helping to identify a potential 'average' parameter set. Furthermore, factor analysis also proves functional in identifying members of the mechanism layer as well as the degree to which model outputs are affected by these members.

Includes bibliographical references (p. 153-167).

The goal of this dissertation is to advance our understanding of perceptual decisions. A perceptual decision is a decision that is based on sensory evidence. For example, a monkey must choose whether to eat a food item based on sensory information such as its color, texture or odor. Previous research has identified regions of the brain involved in the encoding of sensory information as well as areas involved in transforming encoded representations of stimuli into signals useful for forming decisions about those stimuli. Researchers carried out much of this work by painstakingly observing the firing of single neurons or small groups of neurons while a subject performs a task, and used this information to propose and evaluate models of the decision process. However, previous studies have also shown that sensory stimuli are encoded in a distributed fashion across populations of neurons rather than in individual or small groups of neurons. Thus it is likely that populations of neurons, rather than individual neurons, are responsible for the formation of a decision. Here I directly address the question of how decisions are formed through the collective activity of populations of cortical neurons. I used voltage-sensitive dye imaging, a technique that allowed me to simultaneously monitor millions of neurons in sensory cortex, while primates performed a simple yet challenging binary decision task. I also used psychophysical techniques and computational modeling to address fundamental questions about the nature of perceptual decisions. Here I provide new evidence that choice-related neural activity is distributed across a broad population of neurons, and that most of the decision-related neural activity occurs as early as primary sensory cortex. I propose a physiological and computational mechanism for the subject’s decision process in our task, and demonstrate that this process is likely sub-optimal due to intrinsic uncertainty about sensory stimuli. Overall, I conclude that in our task, perceptual decisions are likely to be limited primarily by the quality of evidence that resides in populations of neurons in sensory cortex, secondarily by sub-optimal decoding of these sensory signals, and to a much lesser extent by additional downstream neural variability. / text

We consider three problems, motivated by questions in computational neuroscience, related to recovering continuous quantities from binary or discrete data or measurements in the context of sparse structure. First, we show that it is possible to recover the norms of sparse vectors given one-bit compressive measurements, and provide associated guarantees. Second, we present a novel algorithm for spike-sorting in neural data, which involves recovering continuous times and amplitudes of events using discrete bases. This method, Continuous Orthogonal Matching Pursuit, builds on algorithms used in compressive sensing. It exploits the sparsity of the signal and proceeds greedily, achieving gains in speed and accuracy over previous methods. Lastly, we present a Bayesian method making use of hierarchical priors for entropy rate estimation from binary sequences. / text

Real-world decision making is challenging due, in part, to changes in the underlying reward structure: the best option last week may be less rewarding today. Determining the best response is even more challenging when feedback validity is low. Presented here are the results of two experiments designed to determine the degree to which midbrain reward processing is responsible for detecting reward contingency changes when feedback validity is low. These results suggest that while midbrain reward systems may be involved in detecting unexpected uncertainty in non-stationary environments, other systems are likely involved when feedback validity is low – namely, the locus-coeruleus-norepinephrine system. Finally, a computational model that combines these systems is described and tested. Taken together, these results downplay the role of the midbrain reward system when feedback validity is low, and highlight the importance of the locus-coeruleus-norepinephrine system in detecting reward contingency changes.

Includes bibliographical references (p. 47-49).