Understanding Visual Pathway

The volume and quality of data recorded from the brain are constantly increasing, giving us a better view of mental processes. We collaborate with neuroscience labs, primarily the Gallant lab, to develop methodology for analyzing such data. We focus on understanding human vision by studying the representation of images and videos in the early visual areas. These experiments are great examples for modern statistical work: both the treatment (a video, or sequence of images) and the response (continuous brain-scans, or multiple electrodes) are high-dimensional structured objects. We develop principled methods to relate the stimuli and responses for both prediction and interpretation purposes. These include, among others, methods for supervised feature-extraction; high-dimensional (and semi-parametric) regression models relating the features to neural activity; and methods to aggregate information across multiple responses. In particular, multiple methods based on sparse coding and deep convolutional neural networks for feature extraciton in natural images are analyzed. My collaborators in this project are Yuansi Chen, Adam Bloniarz, Jack Gallant and Bin Yu.


Compression of convolutional neural networks

Recent success of convolutional neural networks with a huge number of parameters in computer vision tasks makes it necessary to study the compression schemes for these networks. Our goal is to identify the redundancy of CNNs by determining similarities between convolutional filters. Additionally, in order to decrease computational cost and memory usage, we aim to introduce compressing approaches for these large-scale networks while maintaining their performance. We introduce compression schemes for convolutional neural networks that have the minimum effect on the performance of the network. We are also interested in identifying the redundant information of a deep netwrok. My collaborator in this project is Bin Yu.


Do retinal ganglion cells project natural scenes to their principal subspace and whiten them?

Several theories of early sensory processing suggest that it whitens sensory stimuli. In this project, we study key predictions of the whitening theory using recordings from ganglion cells in salamander retina responding to natural movies. We have been able to show that while the power spectrum of ganglion cells decays less steeply than that of natural scenes, it is not completely flattened. Moreover, retinal ganglion cells reduce the dimensionality of the photoreceptor signal, as evidenced by their significantly lower numbers. Whitening theory predicts that only the top principal components of the visual stimulus should be transmitted. We study evidences that supports this prediction. My collaborators in this project are Cengiz Pehlevan, Bin Yu and Dmitri Chklovskii.


Registration of medical images usinf bi-dimensional empirical mode decomposition

Efficient feature selection from images leads to a lower operational cost in image processing procedures. We study the problem of biomedical image registration based on features extracted by Bidimensional Empirical Mode Decomposition (BEMD). We propose a hierarchically registration method based on BEMD features. This method improves the registration accuracy. This work is in collaboration with Emad Fatemizadeh.


Estimation of Muscle Force with EMG Signals Using Hammerstein-Wiener Model

Estimation of muscle force is essential for monitoring or control purposes in many studies and applications that include direct human involvement such as control of prosthetic arms and human-robot interaction. We introduce new models to estimate the force of muscle from the EMG signals. Estimation is based on Hammerstein-Wiener Model. The introduced model is trained on a data sets recorded from several subjects and tested on a hold out data set. The results show low error rate between measured force and estimated force. This work is in collaboration with Rahman Khorsandi and Edmond Zahedi.



© Reza Abbasi Asl 2015