Projects
Neural basis for auditory processing
Perelman school of Medicine, UPenn
Septmeber 2020 - Present
Advisors: Maria Geffen, Yale Cohen and Konrad Kording
-
Design experiments and establish mechanisms underlying human perceptual decision-making using Bayesian models and generalized linear models.
-
Investigate role of inhibitory neurons in auditory processing in the brain through multi-level analysis of neural network data.
Research Internship: Natural Language Processing
IBM Watson
Summer 2020
Advisor: Young-suk Lee
Developed and trained multilingual AMR system using weak supervised learning within a large PyTorch codebase. Achieved highly competitive performance that surpassed SOTA in German, Spanish, Italian and Chinese.
Thesis Research: Auditory system processing and neuroscience
Co-affiliated with the Audition lab and the Bhaumik Institute, UCLA
September 2014 - August 2020
Advisors: Dolores Bozovic and Alex Levine
Fluctuation analysis of nonequilibrium limit cycle oscillators : Application to hair cells
Developed a general framework to study effects of stochasticity on non-equilibrium active limit cycle oscillators, specifically inner ear hair bundles, using principles drawn from dynamical systems and statistical mechanics. This framework advances our understanding of fluctuation theorems and how they may be used to probe complex biological systems.
Development of a BCI interface for translation of neural signals to text
Visiting Graduate Student, Brain Computer Interface Lab, UCLA
July 2018 - Jan 2020
Advisor: William Speier
Analyzed multi-electrode ECoG data using supervised learning to detect novel features encapsulating speech production information. Additionally implemented a deep RNN and a temporal language model to identify underlying phonemes and generate text.
Comparison of different deep neural network architectures in decoding speech-producing neural signals
UCLA Course: ECE 293AS
Advisor: Jonathan Kao
Compared and critiqued performances (in Pytorch) using CNNs, bi-directional LSTMs and Autoencoders to map multi-electrode neural signals onto speech phonemes.
Speech recognition and speaker verification
UCLA Courses: EE 241A and EE 241B
Advisor: Abeer Alwan
-
Robustness of features and models for text-dependent speaker verification
Investigated speaker-dependent features and classifiers such as SVM and GMM, for speaker verification under constraints of noisy environments and short utterances. -
Digit recognition under noisy and gender mismatch conditions
Developed algorithm to address mismatch conditions in an automatic speech recognition task using hidden markov models. Experience using HTK toolkit.
Study of spin-injection into electrolytes
Physics of Nanostructured Materials Lab, EPFL
Summer 2013
Developed and characterized techniques to study dynamic nuclear polarization using passage of spin currents through ferro-magnetic electrodes.
Micromagnetic simulation of magnetic reversal in magnetic nanodisks
Spintronics and Thin Film Magnetism Lab, IISc
Summer 2012
Investigated through numerical simulations, the optimization of bit-patterned media for storage of information in hard disk magnetic material.