Graduate Research Assistant

WEIQI SUN

919-***-**** ******@****.***

SUMMARY

Ph.D. student with 6 years of research and 2 years of professional experience in deep learning, machine learning, and computer vision. Diverse advanced industrial experiences including adversarial robustness, compression and acceleration of neural networks, and optimization in deep learning. EDUCATION

North Carolina State University (NC State), Raleigh, NC 2018 - Present Ph.D. in Electrical Engineering

Beijing University of Posts and Telecommunications (BUPT), Beijing, China 2015 - 2018 M.Eng. in Electronics and Communication Engineering Hebei University (HBU), Hebei, China 2011 - 2015

B.Eng. in Communication Engineering

TECHNICAL SKILLS

Programming Python, C++, Java, C#, JavaScript, HTML, SQL, MATLAB, Scripting Advanced PyTorch, TensorFlow, Keras, CUDA

Mathematics Convex Optimization, Probability, Stochastic Process, Queuing Theory, Probabilistic Graphical Models PROFESSIONAL EXPERIENCE

Knowledge Distillation Based Top-k Adversarial Perturbations Python, PyTorch Graduate Research Assistant, NC State Fall 2020 - Present

Designed powerful ordered top-k adversarial attacks towards image classi cations by learning via knowledge distillation and exploring semantic knowledge of labels;

Achieved 1.52x improvement of attack success rate and 0.42x reduction of perturbation energy on Top-5 attacks in ImageNet-1000 dataset compared with Carlini-Wagner (C&W) method;

Learned the imperceptible benign perturbation added to the sketches using knowledge distillation and improved the classi cation accuracy from 25% to 99% in ImageNet-Sketch dataset;

Reduced the perturbation energy in adversarial attacks by exploiting the spatial saliency maps and frequency compo- nents of images.

Exploiting Computational Reuse in Deep Neural Networks Python, C++, CUDA, TensorFlow Graduate Research Assistant, NC State Fall 2019 - Spring 2020

Explored the redundancy of similar pixels in images and designed computational reuse method for neural networks using k-means and locality-sensitive hashing (LSH) to achieve compression and acceleration in deep learning;

Theoretically analyzed the misclassi cation probability and the convergence rate of applying computational reuse to convolutional neural network (CNN) inference and training;

Implemented a custom TensorFlow operator of long short-term memory (LSTM) with computational reuse using CUDA and cuBLAS.

Variational Autoencoders for Gait Recognition and Audio Classi cation Python, Keras, TensorFlow Course Project for Probabilistic Graphical Models, NC State Fall 2020

Detected the out-of-distribution (OOD) noises, e.g. respiratory and speech audios, mixed in normal and abnormal PCG using variational autoencoder (VAE) and achieved AUC of 0.69;

Predicted the environmental context for lower limb prostheses using LSTM-VAE and hidden Markov models (HMMs) and achieved F1 score of 0.63.

Full-Stack Web Development C#, JavaScript, jQuery, HTML Software Engineer Intern, Hebei Tianyi High-Tech Co., Ltd, China Summer 2015

Developed an o ce automation system for an insurance company to issue notice and manage personnel;

Developed an E-commerce website for an insurance company to manage orders and membership and help customers to inquiry product information.

SERVICE & AWARDS

Presenter "Impact of User Mobility on Transmit Power Control in Ultra Dense Networks," IEEE ICC Reviewer IEEE Transactions on Vehicular Technology, IEEE Access, IEEE ICC, IEEE WCNC Awards Excellent Graduate Student at BUPT (Top 2%), First Prize of Scholarship at HBU (Top 5%)

Contact this candidate