Josh DeOliveira

Josh DeOliveira

PhD Candidate @ WPI Data Science DAISY LAB | (He/Him) jcdeoliveira@wpi.edu

About me

I am currently looking for internships in summer 2025!

Research Interests

My research is on Deep Learning Optimization, with a focus on discovering ways to understand the dynamics of how neural networks learn during training.

Bio

I’m a PhD candidate at Worcester Polytechnic Institute (WPI) in the DAISY Lab, advised by Professor Elke Rundensteiner. Prior to my PhD, I earned B.S. degrees in computer science and data science at WPI.

Recent News

  • Dec 24: First author paper on the applicability of theory relating to infinitely-large neural nets for improving real model training accepted @ AAAI 2025.

  • Oct 24: First author paper on stabilizing GANS accepted at @ IEEE Big Data 2024.

  • July 24: Coauthored paper on robust anomaly detection accepted at @ SIGMOD 2025

  • Sept 23: Won first place award at the GSA’s Applied AI Challenge in LLMS with Topologe for detecting long-form generated text from blackbox LLMs.

Select Publications

Paper Figure
The Surprising Effectiveness of Infinite-Width NTKs for Characterizing and Improving Model Training
Joshua DeOliveira , Walter Gerych, Elke Rundensteiner.
AAAI, 2025.
Paper Figure
GAN Stabilization Under Practical Training Assumptions.
Joshua DeOliveira , Walter Gerych, Elke Rundensteiner.
IEEE Big Data, 2024.
Paper Figure
HAR-CTGAN: A Mobile Sensor Data Generation Tool for Human Activity Recognition.
Joshua DeOliveira , Walter Gerych, Aruzhan Koshkarova, Elke Rundensteiner, Emmanuel Agu.
IEEE Big Data 4th Special Session on HealthCare Data, 2022.

Misc

In my free time I enjoy cooking, thrifting, weightlifting, and learning new languages.