Caleb Jones Shibu

Neural Dynamics Team.

Seattle,

calebshibu@arizona.edu

GitHub

Allen Institute, Seattle, WA, Software Engineer 2 - March, 2024 - Present
- Working on development of post processing pipeline for images produced by SLAP2 microscopes to measure & underst and neuron activity.
Biotronics, Ames, IA, Machine Learning Engineer - Jan, 2024 - Feb, 2024
- Worked on cleaning up the image preprocessing codebase.
- Focused on explainablility of ResNet model which predicted Intra-Muscular Fat (IMF) values from ultrasoud images from livestocks.
University of Arizona, Tucson, AZ, Graduate Research Assistant - Aug, 2021 - Dec, 2023
- Developed a network-based finger tapping and imaging rating application using PyGame [Git].
- Acquired fNIRS, EEG and Gaze data from multiple subjects.
- Developed a real-time physio visualization tool using PyQT5 [Git].
- Developed a script for data conversion and labeling, improving the quality and usability of data [Git] [Git] [Git].
- Conducted classification experiments using LSTM to classify valence and arousal from fNIRS and EEG signals, and authored a paper that was accepted for the 2023 NeurIPS conference [nips.cc].
Sree Chitra Tirunal Institute for Medical Sciences & Technology, Kerala, India, Project Scientist - Jan, 2021 - Jul, 2021
- Developed a neurofeedback game application using PyGame that performs real-time filtering of fNIRS signals. This application incorporates a deep learning model for real-time prediction of brain states. The predictions generated by the model are dynamically integrated into the PyGame interface, thereby facilitating an advanced interaction between the model output and user input.
- Developed an explainable AI (xAI) model for fNIRS signal classification using DeepSHAP, which interprets model predictions in terms of channel names. For instance, if the model predicts activation of the left motor cortex, the xAI module will illustrate which channels have positively and negatively contributed to that prediction.
St. Jude Children’s Research Hospital, Memphis, TN, Research Intern - Sep, 2020 - Jan, 2021
- Developed a deep learning-based classification system aimed at differentiating between active and passive brain states. These states are associated with single-trial lower limb motor preparations in stroke patients. Even though the motor cortex controls the knee and hip regions closely, the model was successful in classifying the brain activations caused by knee and hip movements in both active and passive states.
Sree Chitra Tirunal Institute for Medical Sciences & Technology, Kerala, India, Research Intern - Jun, 2019 - Sep, 2020
- Developed handcrafted features for fNIRS signals using PCA and ICA. These features were then utilized to improve the classification accuracy of machine learning classifiers, such as SVM and KNN, in the classification of fNIRS signals as left-, right- brain activation or rest.
- Created a sliding window-based CNN and LSTM Deep Learning model for fNIRS signals, increasing the classification accuracy from 55% to 97%.

2023

The ToMCAT Dataset

Adarsh Pyarelal, Eric Duong, Caleb Jones Shibu, Paulo Soares, Savannah Boyd, Payal Khosla, Valeria Pfeifer, Diheng Zhang, Eric S Andrews, Rick Champlin, Vincent Paul Raymond, Meghavarshini Krishnaswamy, Clayton Morrison, Emily Butler, Kobus Barnard

Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2023

PDF BibTeX

2022

Explainable artificial intelligence model to predict brain states from fNIRS signals

Caleb Jones Shibu, Sujesh Sreedharan, Arun KM, Chandrasekharan Kesavadas, Ranganatha Sitaram

Frontiers in Human Neuroscience Brain-Computer Interfaces (2022).

PDF BibTeX

2020

Comparison of classification performance of handpicked, handcrafted, and automated-features for fNIRS-BCI system

Caleb Jones Shibu, Sujesh Sreedharan, Arun KM, Chandrasekharan Kesavadas

2020 5th International Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS). IEEE, 2020, pp. 152–157

PDF BibTeX