Usman Roshan
Associate Professor, Department of Computer Science, New Jersey Institute of Technology
NJIT Machine Learning Club
Ph.D. Computer Science, The University of Texas at Austin, Advisor: Tandy Warnow

Office: GITC 4214B
Mailing address: Department of Computer Science, NJIT, GITC 4100, University Heights, Newark, NJ 07102
Phone: 973-596-2872 (office)
Fax: 973-596-5777 (fax)
Best way to reach me: usman@njit.edu

---

Research: I am looking for students with strong programming skills and mathematical maturity to work on medical AI, deep learning, and robust machine learning projects. These projects involve algorithms, large datasets, Python, numpy, deep learning libraries such as Keras and Pytorch, and parallel programming on CPUs and GPUs. These projects would lead to a Masters or PhD thesis.

• Adversarial attack challenge - we invite you to perform white box attacks on our CIFAR10 binary classification sign activation 0-1 loss neural network models. In order to succeed in this challenge you must produce adversaries within L-infinity distance of 16/255 of the original image using any method of your choice. Your adversaries must bring down the model accuracy to almost 0%, which is on-par accuracy with the same model structure with relu activations and cross-entropy loss. (Models and scripts to call the models available shortly below.)
  • Placeholder
  • Placeholder
• Robust machine learning
  • Defending against black-box adversarial attacks with gradient-free trained sign activation neural networks (2020) - See GitHub site for source code and data
  • Towards adversarial robustness with 01 loss neural networks (2020) - See GitHub site for source code and data
  • On the transferability of adversarial examples between convex and 01 loss models (2020)
  • Stochastic coordinate descent for 01 loss and its sensitivity to adversarial attacks (2019)
• Medical AI (with Rutgers, Robert Wood Johnson Hospital, Kessler Foundation, and Medical College Wisconsin)
  • Accurate and adversarially robust classification of medical images and ECG time-series with gradient-free trained sign activation neural networks (2020)
  • Two-stage and dual-decoder convolutional U-Net ensembles for reliable vessel and plaque segmentation in carotid ultrasound images (2020)
  • Classifying histopathology images with random depthwise convolutional neural networks (2020)
  • An Ensemble-Based Confidence Score to Increase Accuracy of Carotid Ultrasound Vessel Lumen Segmentation by Convolutional Neural Networks (2020)
  • Deep Learning Neural Networks To Automate The Identification Of Vessel Lumen In B- Mode Carotid Ultrasound Images (2020)
  • CPM-RadPath2019: Brain tumor classification with tumor segmentations and a dual path residual convolutional neural network from MRI and pathology images (2019)
  • BraTS2019: A multi-path decoder network for brain tumor segmentation (2019)
  • Vessel lumen segmentation in internal carotid artery ultrasounds with deep convolutional neural networks (2019)
  • 2.5D convolutional neural network system for identifying stroke lesions in brain MRI images (2019)
  • A fully 3D multi-path convolutional neural network with feature fusion and feature weighting for automatic lesion identification in brain MRI images (2019)
• Prior work

Publications

Teaching (Previous courses)

Students (complete list of student thesis available here)