Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (2 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2018)

Despite progress in understanding the organization and function of neural sensory systems, fundamental questions remain about how organisms convert visual, auditory, and other sensory input into useful representations to understand the world and guide behavior. An important and fruitful line of work models the brain as an unsupervised statistical learner, examining how a sensory system may optimize for efficient representation of the natural environment or for explicit representation of useful structure in that environment. This dissertation explores efficient coding and sparse coding models of the visual and auditory systems, the data these systems process, and how these models are affected by the constraints imposed by implementation in biological neural systems. First, I show that both natural images and natural sounds have statistical structure amenable to a sparse coding model but that the sparse structure of these two types of natural data also differ in interesting ways that may be relevant to extending the success of sparse coding in describing primary visual cortex (V1) to analogous regions of the auditory system. I also discuss how a related model may shed light on how the neurons in these sensory systems are organized in space based on coding for related stimulus properties. Second, I show that a sparse coding model with biological constraints requires its inputs to be whitened in order to learn sparse features using synapse-local learning rules. This observation provides a novel explanation for the separation of sparse coding and spatial decorrelation into, respectively, V1 simple cells and preceding areas including retina. Third and finally, I turn back to the auditory system and extend existing work on efficient coding in the cochlea to account for the requirement of causality, i.e., determining a code without knowledge of the future of a signal.

    Cover page: Statistical learning models of sensory processing and implications of biological constraints
    • Article
    • Peer Reviewed
    Berkeley Research Impact Initiative (BRII) (2019)
    Sparse coding models of natural images and sounds have been able to predict several response properties of neurons in the visual and auditory systems. While the success of these models suggests that the structure they capture is universal across domains to some degree, it is not yet clear which aspects of this structure are universal and which vary across sensory modalities. To address this, we fit complete and highly overcomplete sparse coding models to natural images and spectrograms of speech and report on differences in the statistics learned by these models. We find several types of sparse features in natural images, which all appear in similar, approximately Laplace distributions, whereas the many types of sparse features in speech exhibit a broad range of sparse distributions, many of which are highly asymmetric. Moreover, individual sparse coding units tend to exhibit higher lifetime sparseness for overcomplete models trained on images compared to those trained on speech. Conversely, population sparseness tends to be greater for these networks trained on speech compared with sparse coding models of natural images. To illustrate the relevance of these findings to neural coding, we studied how they impact a biologically plausible sparse coding network's representations in each sensory modality. In particular, a sparse coding network with synaptically local plasticity rules learns different sparse features from speech data than are found by more conventional sparse coding algorithms, but the learned features are qualitatively the same for these models when trained on natural images.
      Top