NeurIPS 2024 accepted papers on Continual learning

Online and replay-based · 16 papers

• Saliency-driven Experience Replay for Continual Learning (9 citations)
• Make Continual Learning Stronger via C-Flat (7 citations)
• A Topology-aware Graph Coarsening Framework for Continual Graph Learning (5 citations)
• Label Delay in Online Continual Learning (5 citations)
• How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? (2 citations)
• Parseval Regularization for Continual Reinforcement Learning (2 citations)
• Predictive Attractor Models (2 citations)
• RanDumb: Random Representations Outperform Online Continually Learned Representations (2 citations)
• Disentangling and mitigating the impact of task similarity for continual learning (1 citation)
• Learning Successor Features the Simple Way (1 citation)
• Continual learning with the neural tangent ensemble
• DASH: Warm-Starting Neural Network Training in Stationary Settings without Loss of Plasticity

and 4 more in the interactive map.

Continual fine-tuning of LLMs · 14 papers

• Mixture of Experts Meets Prompt-Based Continual Learning (8 citations)
• CLAP4CLIP: Continual Learning with Probabilistic Finetuning for Vision-Language Models (7 citations)
• Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference (4 citations)
• SAFE: Slow and Fast Parameter-Efficient Tuning for Continual Learning with Pre-Trained Models (4 citations)
• A Practitioner's Guide to Real-World Continual Multimodal Pretraining (3 citations)
• Learn more, but bother less: parameter efficient continual learning (3 citations)
• CorDA: Context-Oriented Decomposition Adaptation of Large Language Models for Task-Aware Parameter-Efficient Fine-tuning (2 citations)
• Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (2 citations)
• Continual Learning with Global Alignment (1 citation)
• Incremental Learning of Retrievable Skills For Efficient Continual Task Adaptation (1 citation)
• Leveraging Catastrophic Forgetting to Develop Safe Diffusion Models against Malicious Finetuning (1 citation)
• Model Sensitivity Aware Continual Learning (1 citation)

and 2 more in the interactive map.

Class-incremental · 10 papers

• GACL: Exemplar-Free Generalized Analytic Continual Learning (8 citations)
• Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (7 citations)
• Replay-and-Forget-Free Graph Class-Incremental Learning: A Task Profiling and Prompting Approach (6 citations)
• Continual Audio-Visual Sound Separation (4 citations)
• Task-recency bias strikes back: Adapting covariances in Exemplar-Free Class Incremental Learning (4 citations)
• Prospective Representation Learning for Non-Exemplar Class-Incremental Learning (3 citations)
• An Efficient Memory Module for Graph Few-Shot Class-Incremental Learning (1 citation)
• Black-Box Forgetting (1 citation)
• Task Confusion and Catastrophic Forgetting in Class-Incremental Learning: A Mathematical Framework for Discriminative and Generative Modelings (1 citation)
• F-OAL: Forward-only Online Analytic Learning with Fast Training and Low Memory Footprint in Class Incremental Learning

Domain and prompt-based · 8 papers

• Visual Prompt Tuning in Null Space for Continual Learning (6 citations)
• Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (5 citations)
• Continual Learning in the Frequency Domain (5 citations)
• PCoTTA: Continual Test-Time Adaptation for Multi-Task Point Cloud Understanding (4 citations)
• DoFIT: Domain-aware Federated Instruction Tuning with Alleviated Catastrophic Forgetting (3 citations)
• Not Just Object, But State: Compositional Incremental Learning without Forgetting (2 citations)
• CONTRAST: Continual Multi-source Adaptation to Dynamic Distributions (1 citation)
• Vector Quantization Prompting for Continual Learning