​

RESEARCH

​

Age Dependent Causality Changes in Cortical Hubs over Lifespan

​

Gradual cognitive decline in healthy aging might be due to deteriorating capabilities of hub regions as 'function integrators'. To systematically capture age dependent differences, we apply graph theoretic centrality measures and identify hubs in young and old populations.We hypothesize that,community structure of older brains will have low modularity,but higher number of modules due to disconnection.

Additionally, for both the groups, we explore differences in causal relation of hubs participating in Default Mode Network(PCC-mPFC-IPC), Salience Network(ACC-insula) and Central Executive Network(DLPFC-PPC) using Multivariate Granger Causality Model. We hypothesized that, some important regions will have disrupted causality dynamics,suggesting their vanishing role as hubs in older age group. (Poster)

​

Exploring structural and functional connectivity in Autism Spectrum Disorder

​

Since Autism Spectrum Disorder (ASD) is a connectivity disorder, brain connectivity studies have been particularly useful in understanding the differences between autistic individuals compared to normally developed ones. This project attempts to gain insights of ASD using whole brain network analysis of resting state connectivity and structural connectivity. Differences were found prominently in the default mode networks which is supposed to play a crucial role in self-referential thinking and theory of mind. As expected, structural underconnectivity was observed and functional connectivity (FC) exhibited both underconnectedness and hyperconnectedness. Such abnormalities in connections could possibly lead to explain the varied behavior exhibited by autistic individuals and help in their diagnosis. (Report)

​

Operationalization of Multilingualism in Language Contact Situation

​

Aim of this study was to devise a method to measure multilingualism in the Indian context, where most people are bi- or trilinguals irrespective of their education or socioeconomic status. We applied clustering analysis to the language-scores obtained on a Language Usage Questionnaire, LUQ (Vasanta et al, 2010). The study validated that  self-rating of language proficiency / use frequency is sufficient to serve as an accurate estimate of the results derived based on the questions of LUQ. This was one of the first attempts at characterizing multilingual status of Indian participants based on responses to a questionnaire. It assured that objective measure of multilingualism is still possible where formal fluency and proficiency tests are either not available or cannot be employed for deciding the multilingual status. (Questionnaire) (PPT)

​

​

COURSE

​

Restricted Boltzmann Machine for OCR on Nengo Framework

​

Nengo is a Python based software package for simulating large-scale neural systems. It uses Neural Engineering Framework (NEF) at its core.It includes customizable models of spike generation, synaptic plasticity, self-organized and error driven learning. With use of concept
called ’Semantic Pointers’ introduced in Nengo, Eliasmith et al. built biologically plausible model- Spaun that can perform many predefined
tasks including higher level cognitive functions such as action selection and perception. In this project, we reviewed the Visual System Model used by Spaun and explored possibilities of similar models with help of Nengo. Performance of preexisting and proposed models were evaluated on OCR datasets. Weights were learnt by Restricted Boltzman Machine for MNIST and CHAR74K datasets and inputted into Nengo. Performance was tested for different neural coding schemes. (PDF)

​

​

Understanding Visual Attention

​

Feature integration theory (FIT) suggests that features are registered early, automatically and in parallel, whereas actual objects are separated only at later stage, requiring focused attention. On the other hand, stimulus having conjunction of more than one separable features require serial attention. In this project, we replicated  and validated  ’Visual search’ part of seminal paper by Treisman and Gelade on FIT using Psychtoolbox. (PDF)

​

Bird Species Classification based on Audio Recordings

​

In this study we proposed a method for automatic classification of bird species from their audio recordings.Each syllable corresponding to a piece of vocalization is first segmented for each recording.These syllables serve as basic recognition unit. For each syllable, two dimensional Mel frequency cepstral coefficients are calculated as vocalization features to represent temporal variations within a syllable.Principal Component Analysis (PCA) is used to increase the classification accuracy at a lower dimensional feature vector space.In our experiments the
best classification accuracy is 87.23% by Support Vector Machine (SVM) for the classification of 35 bird species. (PDF)

​

​

​

​

​