My interest primarily lies in building quantitative models that provide algorithmic and/or mathematical framework for cortical mechanisms underlying everyday cognitive tasks, primarily visual processing. Before joining University of Tuebingen, I spent some time as a Research Assitant at Computational Visual Neuroscience Lab (UMN) , where I worked with Professor Kendrick Kay on developing a fully generative the model that factors in both bottom-up, sensory-driven responses and top-down, goal-driven attentional state, connections to accurately characterize cortical responses for arbitrary combinations of stimuli and attentional loci. In the "distant" past, I focussed on applying signal processing and machine learning to extract inherent patterns in Magnetic Resonance (MR) images, in both its native image space and frequency space (also called k-space). I undertook my bachelor thesis at Medical Image Processing Lab (EPFL) where I studied structural and functional correlates of personality under the supervision of Professor Dimitri Van de Ville and Professor Sven Haller. Prior to that I was a visiting student researcher at Tomography Lab (Indian Institute of Science) where I worked on developing variable density sampling (VDS) schemes for fast and efficient reconstruction of MR Images under the supervision of Professor Kasi Rajgopal.
Abstract: The revised NEO Personality Inventory (NEO-PI-R), popularly known as the five-factor model, defines five personality factors: Neuroticism, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness. The structural correlates of these personality factors are still a matter of debate. In this work, we examine the impact of subtle cognitive deficits on structural substrates of personality in the elderly using DTI derived white matter (WM) integrity measure, Fractional Anisotropy (FA). We employed canonical correlation analysis (CCA) to study the relationship between each personality factor of the NEO-PI-R and FA measures. Agreeableness was the only personality factor to be associated with FA patterns in both population groups. In MCI cases, Openness was significantly related to FA data whereas the inverse was true for Conscientiousness. Furthermore, we generated saliency maps using bootstrapping strategy which revealed a larger number of positive correlations in healthy brain aging in contrast to the MCI status. The MCI group was found to be associated with a predominance of negative correlations indicating that higher Agreeableness and Openness scores were mostly related to lower FA values in interhemispheric and cortico-spinal tracts and a limited number of higher FA values in cortico-cortical and cortico-subcortical connection. Altogether these findings support the idea that WM microstructure may represent a valid correlate of personality dimensions, and also indicate that the presence of early cognitive deficits led to substantial changes in the associations between WM integrity and personality factors.
Abstract: Swarm optimization algorithms which are motivated by the hierarchical working, efficient self-organizing skills and highly developed foraging ability of the bee population are being increasingly used in wide array fields. In this paper, we propose a novel data-driven algorithm, using swarm intelligence of artificial bee colony (ABC), called k-ABC, to generate an adaptive variable density sampling (VDS) scheme for compressive sampling (CS) based data acquisition for fast MR imaging. The algorithm exploits the behaviour of the three types of bees - scout-bees, employed-bees and onlooker bees, with certain modifications based on the characteristics of variable density sampling. We introduce the concept of searching for the high quality food sources in annular regions of varying widths, called as bins, to optimise the process of foraging. The k-ABC algorithm uses magnitude k-space distribution of a reference single-slice MR image as the underlying fitness value distribution to generate an adaptive sampling scheme. We have also addressed the problem of designing a tailor-made template sampling scheme for 3D MR volume imaging for the very first time. Retrospective simulations show that the proposed k-ABC adaptive VDS scheme gives significant improvement over other sampling schemes for both single slice and multi-slice MR imaging. Further, for the task of implementation, a modified projection algorithm that takes into account the location of each sample in k-space has been introduced, which provides a significant improvement in the reconstructed image quality with minimum trade-off in terms of scan-time.
Abstract: There exists an inherent characteristic distribution in the magnitude spectrum of the Fourier Transform of Magnetic Resonance (MR) images, also referred to as k-space, the high energy samples being concentrated near the center of the magnitude k-space and low energy samples at larger distances. However, there is no method that directly takes into account both the spatial location and the energy contained by a sample, while generating a variable density sampling (VDS) distribution. In this paper, we propose two methods that exploit the underlying structure in MR images and produce adaptive, organ-specific, VDS schemes. First,a weighted K-Means algorithm, an unsupervised learning method, that groups or bins samples with similar energy level weighted by its distance from the center of k-space. Second, a simple binning-based approach that maximizes the overall energy content of the sampling scheme and minimizes the average distance between neighboring samples. Extensive retrospective CS-MRI simulations have shown that the proposed methods perform significantly better than the existing adaptive methodologies.
Abstract: Categorisation of huge amount of data on the multimedia platform is a crucial task. In this work, we propose a novel approach to address the subtle problem of selfie detection for image database segregation on the web, given rapid rise in the number of selfies being clicked. A Convolutional Neural Network (CNN) is modeled to learn a synergy feature in the common subspace of head and shoulder orientation, derived from Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG) features respectively. This synergy was captured by projecting the aforementioned features using Canonical Correlation Analysis (CCA). We show that the resulting networks convolutional activations in the neighbourhood of spatial keypoints captured by SIFT are discriminative for selfie-detection. In general, proposed approach aids in capturing intricacies present in the image data and has the potential for usage in other subtle image analysis scenarios apart from just selfie detection. We investigate and analyse the performance of the popular CNN architectures (GoogleNet, Alexnet), used for other image classification tasks, when subjected to the task of detecting the selfies on the multimedia platform. The results of the proposed approach are compared with these popular architectures on a dataset of ninety thousand images comprising of roughly equal number of selfies and non-selfies. Experimental results on this dataset shows the effectiveness of the proposed approach.
Abstract: Detection of the start and the end time of words in a continuous speech plays a crucial role in enhancing the accuracy of Automatic Speech Recognition (ASR). Hence, addressing the problem of efficiently demarcating word boundaries is of prime importance. In this paper, we introduce two new acoustic features based on higher order statistics called Density of Voicing (DoV) and Variability of Voicing (VoV) obtained from the bispectral distribution, which when used along with the popular prosodic cues helps in drastically reducing the recognition error rate involved. An ensemble of three different models has been designed to minimize the false alarms, during word boundary detection, by maximizing the uncorrelatedness in prediction from each model. Finally, the majority-voting rule was used to decide if the given frame encompasses a word boundary. The contribution of the work lies in: (i) Converting word boundary detection into a supervised learning problem (ii) Introduction of two new higher order statistical features (iii) Using ensemble methods to find the best model for prediction. Experimental results for NTIMIT Database shows the efficacy of the proposed method and its robustness to noise added during telephonic transmission.
Abstract: The self-sustained dynamics of the bee population in nature is a result of their hierarchical working culture, efficient organizing skills and unique highly developed foraging ability, which enables them to interact effectively among each other as well as with their environment. In this paper, a novel algorithm utilizing the bees swarm intelligence, and its heuristics based on quality and quantity of food sources (nectars) is proposed to generate a variable density sampling (VDS) scheme for compressive sampling (CS) based fast MRI data acquisition. The algorithm uses the scout-bees for global random selection process which is further fine-tuned by employed and onlooker-bees who forage locally in the neighborhood giving prime importance to points possessing high fitness values (or high energy) usually located around the center of k-space. The algorithm introduces the concept of searching for the high quality food sources in annular regions, called as bins, of varying widths. Retrospective CS-MRI simulations show that the proposed k-ABC based VDS scheme performs significantly better than other sampling schemes.