arXiv subjects

Richa Tripathi

Publications and source records attributed to Richa Tripathi.

6 recordsLinked to original sources

arXiv2026

An explainable hierarchical self attention-based approach for tremor detection in the time domain

Tremor is a common movement disorder associated with conditions like Parkinson's disease and Essential tremor, traditionally diagnosed through expert clinician assessment. Current automated detection methods rely on frequency-domain features informed by clinical expertise. In this work, we present an explainable, two-stage hierarchical framework for tremor detection in the time domain that learns tremor patterns directly from 3D kinematic marker time-series data across entire tremor-provoking trials. Our framework combined a deep convolutional and long short-term memory network to learn tremor representations from short, discrete, non-overlapping time segments of kinematic time series data from trials, which are then processed by a vision transformer that models their long-term temporal dynamics of time segment features for trial (session) level classification. Evaluated across nine body parts, the framework achieved F1-scores of 0.594 - 0.947 depending on body parts (average: 0.765), falling short of the frequency-domain state-of-the-art performance (0.909) while requiring minimal preprocessing. Attention weights and gradient-based class activation maps (Grad-CAM) identified time-domain features of tremor across body parts. This proof of concept demonstrated the feasibility of data-driven time-domain modeling for tremor detection across anatomically diverse body parts, while reducing reliance on expert-engineered spectral features and providing posthoc interpretability of temporal and anatomical patterns of tremor.

cs.CV↗

arXiv2020

The nonlinearity of interactions drives networks of neural oscillators to decoherence at strong coupling

While phase oscillators are often used to model neuronal populations, in contrast to the Kuramoto paradigm, strong interactions between brain areas can be associated with loss of synchrony. Using networks of coupled oscillators described by neural mass models, we find that a transition to decoherence at increased coupling strength results from the fundamental nonlinearity, e.g., arising from refractoriness, of the interactions between the nodes. The nonlinearity-driven transition also depends on the connection topology, underlining the role of network structure in shaping brain activity.

q-bio.NC↗

arXiv2020

Prediction of the disease controllability in a complex network using machine learning algorithms

The application of machine learning (ML) techniques span a vast spectrum ranging from speech, face and character recognition, medical diagnosis, anomaly detection in data to the general classification, prediction, and regression problems. In the present work, we solve the problem of predicting R_0 for disease spreading on complex networks using the regression-based state-of-art ML techniques. R_0 is a metric that determines whether the disease-free epidemic or an endemic state is asymptotically stable and hence indicates the controllability of the disease spread. We predict R_0 , based on training the ML models with structural properties of complex networks, irrespective of the network type. The prediction is possible because: (a) The structure of complex networks plays an essential role in the spreading processes on networks (b) The regression techniques such as Support Vector Regression and Artificial Neural Network Model can be very efficiently used for prediction problems, even for non-linear data. We obtained good accuracy in the prediction of R_0 for the simulated networks as well as real-world networks using these techniques. Moreover, the ML model training is a one-time investment cost in terms of training time and memory, and the trained model can be used for predicting R_0 on unseen/new examples of networks.

cs.SI↗

arXiv2019

A class of randomized Subset Selection Methods for large complex networks

Most of the real world complex networks such as the Internet, World Wide Web and collaboration networks are huge; and to infer their structure and dynamics one requires handling large connectivity (adjacency) matrices. Also, to find out the spectra of these networks, one needs to perform the EigenValue Decomposition(or Singular Value Decomposition for bipartite networks) of these large adjacency matrices or their Laplacian matrices. In the present work, we proposed randomized versions of the existing heuristics to infer the norm and the spectrum of the adjacency matrices. In an earlier work [1], we used Subset Selection (SS) procedure to obtain the critical network structure which is smaller in size and retains the properties of original networks in terms of its Principal Singular Vector and eigenvalue spectra. We now present a few randomized versions of SS (RSS) and their time and space complexity calculation on various benchmark and real-world networks. We find that the RSS based on using QR decomposition instead of SVD in deterministic SS is the fastest. We evaluate the correctness and the performance speed after running these randomized SS heuristics on test networks and comparing the results with deterministic counterpart reported earlier. We find the proposed methods can be used effectively in large and sparse networks; they can be extended to analyse important network structure in dynamically evolving networks owing to their reduced time complexity.

physics.data-an↗

arXiv2019

A subset selection based approach to finding important structure of complex networks

Most of the real world networks such as the internet network, collaboration networks, brain networks, citation networks, powerline and airline networks are very large and to study their structure, and dynamics one often requires working with large connectivity (adjacency) matrices. However, it is almost always true that a few or sometimes most of the nodes and their connections are not very crucial for network functioning or that the network is robust to a failure of certain nodes and their connections to the rest of the network. In the present work, we aim to extract the size reduced representation of complex networks such that new representation has the most relevant network nodes and connections and retains its spectral properties. To achieve this, we use the Subset Selection (SS) procedure. The SS method, in general, is used to retrieve maximum information from a matrix in terms of its most informative columns. The retrieved matrix, typically known as subset has columns of an original matrix that have the least linear dependency. We present the application of SS procedure to many adjacency matrices of real-world networks and model network types to extract their subset. The subset owing to its small size can play a crucial role in analyzing spectral properties of large complex networks where space and time complexity of analyzing full adjacency matrices are too expensive. The adjacency matrix constructed from the obtained subset has a smaller size and represents the most important network structure. We observed that the subset network which is almost half the size of the original network has better information flow efficiency than the original network.

physics.soc-ph↗

arXiv2018

Characterizing functional brain networks and emotional centers based on Rasa theory of Indian aesthetics

In Indian history of arts, Rasas are the aesthetics associated with any auditory, visual, literary or musical piece of art that evokes highly orchestrated emotional states. In this work, we study the functional response of the brain to movie clippings meant to evoke the Rasas through network analysis. We extract functional brain networks using coherence measures on EEG recordings of film clips from popular Indian Bollywood movies representing nine Rasas in the Indian Natyasastra. Structural and functional network measures were computed for these brain networks, averaging over a range of significant edge weights, in different brainwave frequency bands. We identify segregation of neuronal wiring in the brain into modules using a community detection algorithm. Further, using mutual information measure, we compare and contrast the modular organizations of brain network corresponding to different Rasas. Hubs identified using centrality measure reveal the network nodes that are central to information propagation across all Rasas. We also observe that the functional connectivity is suppressed when high-frequency waves such as beta and gamma are dominant in the brain. The significant links causing differences between the Rasa pairs are extracted statistically.

q-bio.NC↗