About

Sijia Liu

Affiliated Professor

Google Scholar Research

Categories

Sijia Liu is an assistant professor at the CSE department of Michigan State University, an affiliated professor at IBM Research, and an MIT-IBM Watson AI Lab affiliated PI. He received the PhD (with All-University Doctoral Prize) in electrical and computer engineering from Syracuse University. He was a postdoctoral research fellow at the University of Michigan from 2016-2017, and a research staff member at the MIT-IBM Watson AI Lab in 2018-2020. His research interests include scalable and trustworthy AI, e.g. scalable optimization for deep models, machine unlearning for vision and language models, AI robustness, and data-model efficiency.

Publications with the MIT-IBM Watson AI Lab

Understanding and Improving Visual Prompting: A Label-Mapping Perspective

CVPR

What Is Missing in IRM Training and Evaluation? Challenges and Solutions

ICLR

TextGrad: Advancing Robustness Evaluation in NLP by Gradient-Driven Optimization

ICLR

Joint Edge-Model Sparse Learning is Provably Efficient for Graph Neural Networks

ICLR

A Theoretical Understanding of Shallow Vision Transformers: Learning, Generalization, and Sample Complexity

ICLR

Fairness Reprogramming

NeurIPS AI Fairness

Advancing Model Pruning via Bi-level Optimization

NeurIPS Optimization

Learning to Generate Image Source-Agnostic Universal Adversarial Perturbations

IJCAI

Revisiting Contrastive Learning through the Lens of Neighborhood Component Analysis: an Integrated Framework

ICML

Data-Efficient Double-Win Lottery Tickets from Robust Pre-training

ICML

How unlabeled data improve generalization in self-training? A one-hidden-layer theoretical analysis

ICLR

Optimizer Amalgamation

ICLR

How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective

ICLR

Reverse Engineering of Imperceptible Adversarial Image Perturbations

ICLR

Why Lottery Ticket Wins? A Theoretical Perspective of Sample Complexity on Sparse Neural Networks

NeurIPS

When does Contrastive Learning Preserve Adversarial Robustness from Pretraining to Finetuning?

NeurIPS

Adversarial Attack Generation Empowered by Min-Max Optimization

NeurIPS

Decentralized Learning for Overparameterized Problems: A Multi-Agent Kernel Approximation Approach

ICLR

Is There a Trade-Off Between Fairness and Accuracy? A Perspective Using Mismatched Hypothesis Testing

ICML Machine Learning

Fast Learning of Graph Neural Networks with Guaranteed Generalizability: One-hidden-layer Case

ICML Machine Learning

Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases

ECCV

An Image Enhancing Pattern-based Sparsity for Real-time Inference on Mobile Devices

ECCV Computer Vision

Long Live the Lottery: The Existence of Winning Tickets in Lifelong Learning

ICLR

RT3D: Achieving Real-Time Execution of 3D Convolutional Neural Networks on Mobile Devices

AAAI Machine Learning

Robust Overfitting may be mitigated by properly learned smoothening

ICLR

On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning

ICLR Machine Learning

Generating Adversarial Computer Programs using Optimized Obfuscations

ICLR Machine Learning

The Lottery Tickets Hypothesis for Supervised and Self-supervised Pre-training in Computer Vision Models

CVPR Computer Vision

Self-Progressing Robust Training

AAAI Machine Learning

Hidden Cost of Randomized Smoothing

AISTATS Machine Learning

Lottery Ticket Preserves Weight Correlation: Is It Desirable or Not?

ICML

Fast Training of Provably Robust Neural Networks by SingleProp

AAAI

Rate-improved inexact augmented Lagrangian method for constrained nonconvex optimization

AISTATS

Higher-Order Certification For Randomized Smoothing

Optimization

Training Stronger Baselines for Learning to Optimize

Optimization

The Lottery Ticket Hypothesis for the Pre-trained BERT Networks

Efficient AI

Adversarial T-shirt! Evading Person Detectors in A Physical World

ECCV Computer Vision

Min-Max Optimization without Gradients: Convergence and Applications to Black-Box Evasion and Poisoning Attacks

ICML

Proper Network Interpretability Helps Adversarial Robustness in Classification

ICML

Towards Verifying Robustness of Neural Networks against Semantic Perturbations

CVPR

Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning

CVPR

Sign-OPT: Defending the hard-label black-box cyber attack

Cybersecurity Adversarial Machine Learning

Adversarial Robustness vs Model Compression, or Both?

Robustness

On the Design of Black-box Adversarial Examples by Leveraging Gradient-free Optimization and Operator Splitting Method

Optimization Robustness

On the Convergence of A Class of Adam-Type Algorithms for Non-Convex Optimization

Optimization

Towards Certificated Model Robustness Against Weight Perturbations

AAAI

Automating machine learning with a joint selection framework

AAAI AutoML

ZO-AdaMM: Zeroth-Order Adaptive Momentum Method for Black-Box Optimization

NeurIPS

ZO-AdaMM: Derivative-free optimization for black-box problems

Optimization NeurIPS

Topology Attack and Defense for Graph Neural Networks: An Optimization Perspective

IJCAI

Fast Incremental von Neumann Graph Entropy Computation: Theory, Algorithm, and Applications

ICML

CNN-Cert: An Efficient Framework for Certifying Robustness of Convolutional Neural Networks

Robustness

signSGD via Zeroth-Order Oracle

Optimization Robustness

Structured Adversarial Attack: Towards General Implementation and Better Interpretability

Robustness Explainability

AutoZOOM: Autoencoder-based Zeroth Order Optimization Method for Attacking Black-box Neural Networks

Robustness Optimization

Zeroth-Order Stochastic Variance Reduction for Nonconvex Optimization

Optimization Deep Learning