Hello!

My name is

Jacob

Download my CV

What I do

I strongly believe that the path to true artificial general intelligence lies in how we handle data with many modalities as well as the data where certain modes are missing. Humans are incredibly flexible. Yes, we rely heavily on sight, sound, and language to understand our environment, but we don’t always have the luxury of all three. Even so, we can make powerful and accurate inferences with less data. Take away our vision and we can still paint a decent picture of our surroundings. Given a language barrier we are still able to infer general ideas based on actions and tone. This flexibility is imperative in our lives, and should be in our algorithms as well. Data isn’t always perfect; this is something that I have become intimately familiar with. Nonetheless our artificial intelligence should be able to adapt and perform strongly even in the absence of some key information.

Research

I am honored to have researched alongside multiple professors at West Virginia University including Drs. Gianfranco Doretto and Xin Li, Prashnna Gyawali. Here, I massively improved my expertise in PyTorch, gained exposure to Transformer networks and multimodal learning, and performed survival analysis on Alzheimer’s Disease.

Educate

Outside of AI research, I am very passionate about teaching what I know to others. From beginner AI workshops to advanced topics such as Transformer encoding and decoding, I strive to make artificial intelligence accessible to anyone with the desire to learn. I have run workshops and brought in guest speakers from companies such as Tesla and Noblis to speak with members of the Artificial Intelligence Club at West Virginia University. I have also brought AI down to the high school level through "AI in a Day" workshops, where I taught everything from the fundamentals of Python to simple neural networks for image classification.

Learn

I am always eager to further my AI knowledge. I love experimenting with new models and processes to find what works best for me. The interesting thing about AI is that there is always something new to learn. Until recently, I was only interested in computer vision, but my work with Dr. Li has opened my eyes to the importance of multimodal learning. Now, I aim to combine techniques from computer vision with natural language processing.

My work

Multimodal Federated Learning in Healthcare: a review

Sep 2023

Submitted to Cell Patterns [awaiting review decision]. Federated Learning (FL) provides a decentralized mechanism where data need not be consolidated, thereby enhancing the privacy and security of sensitive healthcare data. The integration of FL in the healthcare domain supports the ongoing progress of multimodal learning in healthcare while ensuring the security and privacy of patient records within local data-holding agencies.

img

Explainable Survival Analysis for Dementia Prediction

June - July 2023

Extended abstract submitted to IEEE-BHI EMBS 2023. This study explores different machine learning-based survival analysis approaches to predict the probability of Alzheimer’s Disease (AD) dementia progression. We utilize the Alzheimer’s Disease Neuroimaging Initiative (ADNI) data and analyze different features to explain their importance in disease progression. The study's findings can help us understand mechanism of AD Dementia, predict the patients' AD shift efficiently and recommend personalized treatment to mitigate or postpone the effects of AD.

img

Deep Convolutional GAN

Jan - Dec 2022

Developed using PyTorch as my Capstone project for my B.S Mathematics at WVU. Implements a Deep Convolutional Generative Adversarial network (DCGAN) to generate faces based on the CelebA dataset. Click to download the full paper.

img

Opioid Trafficking Detection on TikTok

Jan - Dec 2022

Developed as part of my B.S Computer Science capstone project. This project introduces a TikTokScraper Python API that can easily search, download, and label videos from the platform. Furthermore, Vision Transformers are applied to the data in a binary classification task to determine the presence of potentially illegal substances in these videos. It should be noted that this project acts as a stepping stone to further exploration of the data in the multimodal space by including audio and text information in future classification tasks. Click to download the full paper.

img

Automated Image Captioning usng Transformer Achitecture

Nov 2022

Lead programmer of a small team for the week long Purdue Krannert School of Management Case Competition. We were tasked to develop a model to automatically caption images in Kyrgyz, Hausa, and Thai. Click for the full code.

img

AI Learns to play Contra

Apr - May 2021

Used Convolutional Neural Networks and Reinforcement Learning to create an agent to learn how to play the first level of the NES game, Contra. Built as a group project for an Artificial Intelligence class at West Virginia University.

img
img

About me

Doctoral Student at WVU

I am originally from Charles Town, WV, where I grew up building computers, playing lacrosse, and marching in my high school band. I completed my Bachelors degrees in Computer Science and Mathematics at West Virginia University in December 2022. In August, returned to WVU to pursue my Ph.D in Computer Science, researching with Drs. Xin Li and Prashnna Gyawali. My main research focus is on multimodal machine learning that is robust to missing modality.

Feel free to email me :)

jthrasher0100@gmail.com