RESEARCH
With improved neural interface devices, neuroscientists can now gather gather data in previously unimagined quantities. This demands and allows for the development of analysis methods which are crucial to understand dynamic systems function of brain networks.My goal as researcher is twofold. My first goal is to address the challenges in the analysis of large data-sets of human/rodents neural recordings obtained by great effort with the neural devices developed in my research lab. This involves developing algorithms and computational tools for data preprocessing, sampling, event detection and hypothesis testing in MATLAB and Python.My second goal is to implement these techniques to understand inter-cortical communication in the brain. With a better sense of the mechanisms that cause large-scale network dysfunction, we can get closer in the diagnoses and treatment of neuropyschiatric diseases such as epilepsy.
Columbia University
Graduate Research Assistant
Translational NeuroElectronics Lab
Prof. Dion Khodagholy
Sept 2018 - Present
• Identify biomarkers of memory process
• Analyze large-scale intracranial EEG datasets from rodents performing behavior
• Develop neural computational tools using signal processing, statistics and machine learning
First author, peer-reviewed manuscripts from my PhD
Disruption of large-scale intercortical communication in epilepsy
Published: Brain 2019
In this study, I investigated how intercortical communication is disrupted in the neural networks of patients with focal epilepsy.
Using large-scale human intracranial electroencephalography (iEEG) recordings, we determined that interictal epileptiform discharges (IEDs) in patients with focal epilepsy can couple with spindles, creating pathologic functional connectivity that enables epileptic activity to exert influence beyond the epileptic network.
These findings suggest that these IED-driven, spatiotemporally specific patterns of abnormally coordinated brain activity could provide a mechanism for large-scale disruption of neural network function in focal epilepsy. Consequently, this coupling may represent a therapeutic target for closed-loop therapy to address cognitive comorbidities and disease progression in these patients.
Hippocampal-cortical dialogue for memory consolidation
Published: PNAS 2023
Reactivation of long-term memories enables experience-dependent strengthening, weakening, or updating of memory traces. Although coupling of hippocampal and cortical activity patterns facilitates initial memory consolidation, whether and how these patterns are involved in post reactivation memory processes are not known.
We monitored the hippocampal–cortical network as rats repetitively learned and retrieved spatial and nonspatial memories. We show that interactions between hippocampal sharp wave–ripples, cortical spindles, and cortical ripples are jointly modulated in the absence of memory demand but independently recruited depending on the stage of memory and task type.
Our findings suggest that specific, time-limited patterns of oscillatory coupling can support the distinct memory processes required to flexibly manage long-term memories in a dynamic environment.
Projects
Projects completed for graduate courses
Speech Emotion Classification using RNN
Course: Speech and Audio Processing
Since a lot of information in human speech is conveyed through emotional cues, there has been a growing interest in speech emotion recognition. However, automatic speech emotion recognition (SER) is a challenging task as it heavily depends on the effectiveness of the features used for classification.
In this project, I designed an attention-based bi-directional LSTM neural network to classify speech based on 8 emotions. This RNN performs better than traditional machine learning classifiers as it automatically discovering emotionally relevant features from speech.
Sign Language Detection using CNN
Course: Image Processing
American Sign Language (ASL) facilitates exchange of information between deaf members, yet it is still a challenge for them to communicate with the general public. Written form of communication with the deaf can not only be umbersome
but also impractical in cases of emergencies. Hence, a real-time sign detection system can be a bridge to bring these two communities together.
In this project, I designed a convolutional neural network (CNN) that can learn features from images of user’s ASL signs and translate those images into text. A test accuracy of 86% was achieved.
Virtual-reality mediated teleoperation
Published : IMECE 2019
In this project, we used virtual reality (VR) headsets and glove-like interfaces as a method for teleoperation of a robot. For effective teleoperation, it is imperative that the operator possess adequate feedback regarding the remote device state. This work utilizes commodity VR technology to replicate critical remote task features in a purely software/virtual environment from sensor data. The system is designed and executed on a real robotic platform, and preliminary operation is encouraging. Moreover, high dexterity VR gloves provide an intuitive and natural interaction for the operator, as the user may present fluid and motion commands without the use of an unnatural game-controller. A user study was conducted to compare performance along several relevant metrics between use of glove-like and game controller interfaces for VR teleoperation. These metrics include time to completion, path length, and jerk as a measure of path smoothness. The results of said study suggest strongly that teleoperator performance improves with the adoption of glove-like interfaces.
Education
Columbia University
Doctor of Philosophy (Ph.D.)
2018-Present
Electrical Engineering
Columbia University
Masters of Science (MS)
2018-2019
Electrical Engineering
Trinity College
Bachelors of Science (Summa Cum Laude)
2014-2018
Engineering and Physics
Contact
Email
LinkedIn
GitHub
