Bolor-Erdene Zolbayar

Software Engineer

Insight Data Science•  May 2020 - Present

• Built a fashion outfit recommendation system, trained on decision behavior of human stylists, that enables users to find clothing pieces that match what is in their online shopping cart or wardrobes • Applied inception_v3 and resnet CNN based transfer learning for image embedding and nltk for word embedding, then learned fashion compatibility by Bi-LSTM, trained on AWS using PyTorch and Docker • Achieved more than 70% success rate on overlapping between model results and stylist recommendations • Deployed a website and delivered API service for fashion websites using FLASK and StreamLit

Software Engineer

Pennsylvania State University•  May 2016 - May 2020

• Exposed security vulnerabilities of ML models in self-driving cars and intrusion detection systems to help ML experts to build reliable and trustworthy systems against attacks • Designed GAN-based attack algorithms that fool ML models in the image and network domains using Tensorflow and PyTorch • Built CNN, MLP models that imitate the decision boundaries of target ML models trained with large datasets such as Imagenet, SVHN, and CICIDS-2017; evaluated the attacks against robust classic ML and DNN models and achieved greater than 90% success rates • Wrote a chapter in the book "Game Theory and Machine Learning for Cyber Security" being published by U.S. Army Research Lab in July 2020 and submitted a paper to the top CS security conference in August 2020

Pennsylvania State University

Computer Science, MS•  May 2018 - August 2020

Focusing on the security of machine learning in object and network intrusion detection domain.