Gradient Manipulation

Gradient manipulation is a burgeoning field exploring how altering gradients during model training or inference can improve performance, enhance robustness, or reveal model vulnerabilities. Current research focuses on applications such as improving the efficiency and safety of reinforcement learning, enhancing the transferability of adversarial attacks, and developing more effective methods for detecting backdoors and out-of-distribution samples. These techniques are impacting various machine learning domains, from speech and image recognition to safe AI development, by addressing critical challenges in model training, optimization, and security. The resulting advancements contribute to more robust, efficient, and trustworthy AI systems.

Papers

November 27, 2024

Proactive Gradient Conflict Mitigation in Multi-Task Learning: A Sparse Training Perspective
Zhi Zhang, Jiayi Shen, Congfeng Cao, Gaole Dai, Shiji Zhou, Qizhe Zhang, Shanghang Zhang, Ekaterina Shutova
Training Data Multi Task Learning Many Sparse Simultaneous Learning Gradient Conflict Task Gradient Gradient Manipulation

August 29, 2024

Revisit Micro-batch Clipping: Adaptive Data Pruning via Gradient Manipulation
Lun Wang
Speech Model Mini Batch Adaptive Clipping Gradient Manipulation

August 22, 2024

Enhancing Transferability of Adversarial Attacks with GE-AdvGAN+: A Comprehensive Framework for Gradient Editing
Zhibo Jin, Jiayu Zhang, Zhiyu Zhu, Chenyu Zhang, Jiahao Huang, Jianlong Zhou, Fang Chen
Adversarial Attack Generative Adversarial Network Transferable Attack Transferable Adversarial Attack Improving Transferability Gradient Manipulation

May 31, 2024

Flexible and Efficient Surrogate Gradient Modeling with Forward Gradient Injection
Sebastian Otte
Deep Learning Framework Extension Study Automatic Differentiation Surrogate Gradient Forward Gradient Gradient Manipulation Gradient Amplification

May 2, 2024

Balance Reward and Safety Optimization for Safe Reinforcement Learning: A Perspective of Gradient Manipulation
Shangding Gu, Bilgehan Sel, Yuhao Ding, Lu Wang, Qingwei Lin, Ming Jin, Alois Knoll
Reinforcement Learning Visual Perspective Safe Reinforcement Learning Reward Structure Probabilistic Safety Safe Reinforcement Learning Algorithm Gradient Manipulation

December 27, 2023

Generating gradients in the energy landscape using rectified linear type cost functions for efficiently solving 0/1 matrix factorization in Simulated Annealing
Makiko Konoshima, Hirotaka Tamura, Yoshiyuki Kabashima
Matrix Factorization Value Function Simulated Annealing Energy Minimization Matrix Vector Product Gradient Manipulation

December 23, 2023

Gradient Shaping for Multi-Constraint Safe Reinforcement Learning
Yihang Yao, Zuxin Liu, Zhepeng Cen, Peide Huang, Tingnan Zhang, Wenhao Yu, Ding Zhao
Multi Objective Optimization Safe Reinforcement Learning Safe Reinforcement Learning Algorithm Gradient Manipulation

October 26, 2023

Network Design through Graph Neural Networks: Identifying Challenges and Improving Performance
Donald Loveland, Rajmonda Caceres
Graph Neural Network Technical Challenge System Performance Non Rigid Editing Network Design Graph Edit Distance Gradient Manipulation

September 21, 2023

Multi-Task Cooperative Learning via Searching for Flat Minima
Fuping Wu, Le Zhang, Yang Sun, Yuanhan Mo, Thomas Nichols, Bartlomiej W. Papiez
Multi Task Learning Multi Task Medical Image Analysis Flat Minimum Gradient Manipulation

January 29, 2023

Gradient Shaping: Enhancing Backdoor Attack Against Reverse Engineering
Rui Zhu, Di Tang, Siyuan Tang, Guanhong Tao, Shiqing Ma, Xiaofeng Wang, Haixu Tang
Backdoor Attack Backdoor Defense Reverse Engineering Gradient Manipulation Trigger Inversion

June 16, 2022

Gradient-Based Adversarial and Out-of-Distribution Detection
Jinsol Lee, Mohit Prabhushankar, Ghassan AlRegib
Distribution Detection Natural Gradient Gradient Based Adversarial Text Expressivity Transfer Gradient Manipulation

Gradient Manipulation

Papers

Proactive Gradient Conflict Mitigation in Multi-Task Learning: A Sparse Training Perspective

Revisit Micro-batch Clipping: Adaptive Data Pruning via Gradient Manipulation

Enhancing Transferability of Adversarial Attacks with GE-AdvGAN+: A Comprehensive Framework for Gradient Editing

Flexible and Efficient Surrogate Gradient Modeling with Forward Gradient Injection

Balance Reward and Safety Optimization for Safe Reinforcement Learning: A Perspective of Gradient Manipulation

Generating gradients in the energy landscape using rectified linear type cost functions for efficiently solving 0/1 matrix factorization in Simulated Annealing

Gradient Shaping for Multi-Constraint Safe Reinforcement Learning

Network Design through Graph Neural Networks: Identifying Challenges and Improving Performance

Multi-Task Cooperative Learning via Searching for Flat Minima

Gradient Shaping: Enhancing Backdoor Attack Against Reverse Engineering

Gradient-Based Adversarial and Out-of-Distribution Detection