Browsing by Author "Goodman, Ilan N."

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Analyzing statistical dependencies in neural populations
    (2005) Goodman, Ilan N.; Johnson, Don H.
    Neurobiologists recently developed tools to record from large populations of neurons, and early results suggest that neurons interact to encode information jointly. However, traditional statistical analysis techniques are inadequate to elucidate these interactions. This thesis develops two multivariate statistical dependence measures that, unlike traditional measures, encompass all high-order and non-linear interactions. These measures decompose the contributions of distinct subpopulations to the total dependence. Applying the dependence analysis to recordings from the crayfish visual system, I show that neural populations exhibit complex dependencies that vary with the stimulus. Using Fisher information to analyze the effectiveness of population codes, I show that optimal rate coding requires negatively dependent responses. Since positive dependence through overlapping stimulus attributes is an inherent characteristic of many neural systems, such neurons can only achieve the optimal code by cooperating.
  • Thumbnail Image
    Item
    Performance limits of brain machine interfaces
    (2010) Goodman, Ilan N.; Johnson, Don H.
    Understanding the constraints governing information transfer between electrodes and neurons is crucial to the effective design of neural prostheses. In sensory prostheses such as cochlear implants, information is transferred to the brain by stimulating neurons to produce sensation. In motor prostheses such as cortically controlled bionic limbs, neural recordings are processed to extract information needed to control a computer or mechanical device. In each case, performance of the prosthesis hinges on how effectively information can be conveyed to or from the device at the interface between brain and machine. In this thesis, we investigate the performance capabilities and constraints of brain machine interfaces (BMIs) using an information theoretic approach. Modeling the BMI as a vector Poisson process channel, we compute the information capacity of several different types of BMI channels. Since capacity defines the ultimate fidelity limits of information transmission by any system, this approach gives us an objective way of evaluating and comparing different types of BMIs by determining the best possible performance of each system given its unique constraints. For stimulation BMIs, we examine how the capacity of the system scales with the number of inputs, the constraints on the inputs, and inter-neuronal dependencies. For control BMIs, we quantify the loss in performance that results from using extracellular recordings, where signals from multiple neurons are received on a single electrode. This performance loss can be mitigated through spike sorting, and we show how the properties of the spike sorting algorithm have direct consequences for the resulting BMI capacity. We also provide extensions to the basic models to account for signal attenuation, cross-talk, and measurement noise. Finally, we discuss the real-world significance of BMI capacity in the context of Rate-Distortion Theory, and interpret the capacity results using performance criteria that are relevant to BMIs. This framework provides a direct way to compare competing systems, and allows us to make predictions about the specific conditions necessary for a BMI to achieve a desired performance level.