Gradient Episodic Memory

Gradient Episodic Memory (GEM) is a continual learning technique aiming to mitigate catastrophic forgetting—the tendency of neural networks to forget previously learned tasks when learning new ones. Current research focuses on improving GEM's efficiency and effectiveness, exploring variations like Averaged GEM (A-GEM) and incorporating techniques such as data augmentation and sharpness-aware optimization to enhance performance and generalization. These advancements are significant for addressing the limitations of traditional deep learning in scenarios requiring sequential learning from non-stationary data, with applications ranging from robotics and reinforcement learning to natural language processing and speech recognition.

Papers

October 1, 2024

Fine-Grained Gradient Restriction: A Simple Approach for Mitigating Catastrophic Forgetting
Bo Liu, Mao Ye, Peter Stone, Qiang Liu
Continual LEArning Catastrophic Forgetting Simple Approach Memorization Capacity Core Challenge Gradient Episodic Memory

August 24, 2024

Data Augmentation for Continual RL via Adversarial Gradient Episodic Memory
Sihao Wu, Xingyu Zhao, Xiaowei Huang
Reinforcement Learning Data Augmentation Continual Reinforcement Learning Gradient Episodic Memory

June 25, 2024

Grass: Compute Efficient Low-Memory LLM Training with Structured Sparse Gradients
Aashiq Muhamed, Oscar Li, David Woodruff, Mona Diab, Virginia Smith
Large Language Model Gradient Sparsification Projection Matrix Gradient Episodic Memory Grass Growth Sparse Update

May 15, 2024

MGSER-SAM: Memory-Guided Soft Experience Replay with Sharpness-Aware Optimization for Enhanced Continual Learning
Xingyu Li, Bo Tang
Continual LEArning Catastrophic Forgetting Continual Learning Experience Replay Replay Based Gradient Episodic Memory Sharpness Aware Optimization

September 21, 2023

Clustering-based Domain-Incremental Learning
Christiaan Lamers, Rene Vidal, Nabil Belbachir, Niki van Stein, Thomas Baeck, Paris Giampouras
Catastrophic Forgetting Continual Learning Domain Incremental Learning Gradient Episodic Memory

May 31, 2023

Adam Accumulation to Reduce Memory Footprints of both Activations and Gradients for Large-scale DNN Training
Yijia Zhang, Yibo Han, Shijie Cao, Guohao Dai, Youshan Miao, Ting Cao, Fan Yang, Ningyi Xu
Memory Trace Targeted Activation Penalty Neural Network Training Gradient Episodic Memory Gradient Accumulation

April 25, 2023

A Multi-Task Approach to Robust Deep Reinforcement Learning for Resource Allocation
Steffen Gracla, Carsten Bockelmann, Armin Dekorsy
Multi Task Learning Multi Task Resource Allocation Actor Critic Algorithm Robust Deep Reinforcement Learning Gradient Episodic Memory

February 14, 2023

EPISODE: Episodic Gradient Clipping with Periodic Resampled Corrections for Federated Learning with Heterogeneous Data
Michael Crawshaw, Yajie Bao, Mingrui Liu
Stochastic Gradient Heterogeneous Data Gradient Norm Uncertain Episode Length Gradient Clipping Gradient Episodic Memory Efficient Gradient

July 11, 2022

Online Continual Learning of End-to-End Speech Recognition Models
Muqiao Yang, Ian Lane, Shinji Watanabe
Automatic Speech Recognition Continual LEArning End to End Online Continual Learning Speech Recognition Model Gradient Episodic Memory

April 11, 2022

CFA: Constraint-based Finetuning Approach for Generalized Few-Shot Object Detection
Karim Guirguis, Ahmed Hendawy, George Eskandar, Mohamed Abdelsamad, Matthias Kayser, Juergen Beyerer
Constructive Approach Continual Learning Shot Object Detection Gradient Episodic Memory

March 22, 2022

Provable Constrained Stochastic Convex Optimization with XOR-Projected Gradient Descent
Fan Ding, Yijie Wang, Jianzhu Ma, Yexiang Xue
Gradient Descent Stochastic Convex Optimization Gradient Episodic Memory

March 12, 2022

Sparsity and Heterogeneous Dropout for Continual Learning in the Null Space of Neural Activations
Ali Abbasi, Parsa Nooralinejad, Vladimir Braverman, Hamed Pirsiavash, Soheil Kolouri
Continual LEArning Catastrophic Forgetting Sparsity Increase Structured Dropout Continual Learning Benchmark Continual Learner Null Space Neuron Activation Gradient Episodic Memory

Gradient Episodic Memory

Papers

Fine-Grained Gradient Restriction: A Simple Approach for Mitigating Catastrophic Forgetting

Data Augmentation for Continual RL via Adversarial Gradient Episodic Memory

Grass: Compute Efficient Low-Memory LLM Training with Structured Sparse Gradients

MGSER-SAM: Memory-Guided Soft Experience Replay with Sharpness-Aware Optimization for Enhanced Continual Learning

Clustering-based Domain-Incremental Learning

Adam Accumulation to Reduce Memory Footprints of both Activations and Gradients for Large-scale DNN Training

A Multi-Task Approach to Robust Deep Reinforcement Learning for Resource Allocation

EPISODE: Episodic Gradient Clipping with Periodic Resampled Corrections for Federated Learning with Heterogeneous Data

Online Continual Learning of End-to-End Speech Recognition Models

CFA: Constraint-based Finetuning Approach for Generalized Few-Shot Object Detection

Provable Constrained Stochastic Convex Optimization with XOR-Projected Gradient Descent

Sparsity and Heterogeneous Dropout for Continual Learning in the Null Space of Neural Activations