mlsanityData to 7 Oct 2026

Gradient Surgery for Multi-Task Learning

arXiv 2001.06782 ↗First posted 19 Jan 2020cs.LG cs.CV cs.ROcited by 1,210

Not on the map. Why?

Cited by 45 in Diffusion models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

While deep learning and deep reinforcement learning (RL) systems have demonstrated impressive results in domains such as image classification, game playing, and robotic control, data efficiency remains a major challenge. Multi-task learning has emerged as a promising approach for sharing structure across multiple tasks to enable more efficient learning. However, the multi-task setting presents a number of optimization challenges, making it difficult to realize large efficiency gains compared to learning tasks independently. The reasons why multi-task learning is so challenging compared to single-task learning are not fully understood. In this work, we identify a set of three conditions of the multi-task optimization landscape that cause detrimental gradient interference, and develop a simple yet general approach for avoiding such interference between task gradients. We propose a form of gradient surgery that projects a task's gradient onto the normal plane of the gradient of any other task that has a conflicting gradient. On a series of challenging multi-task supervised and multi-task RL problems, this approach leads to substantial gains in efficiency and performance. Further, it is model-agnostic and can be combined with previously-proposed multi-task architectures for enhanced performance.

Key facts. Cited by 1,210 papers in the map (data through 7 Oct 2026). First posted on arXiv 19 Jan 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 89; 2022: 114; 2023: 157; 2024: 191; 2025: 263; 2026: 396.

1,210 papers.

Sort:

By area:

  1. 2021ImageNet-21K Pretraining for the Massescited by 607
  2. 2024LESS: Selecting Influential Data for Targeted Instruction Tuningcited by 514
  3. 2022MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledgecited by 469
  4. 2021Conflict-Averse Gradient Descent for Multi-task Learningcited by 441
  5. 2022Prompt-aligned Gradient for Prompt Tuningcited by 339
  6. 2021An Empirical Study of Training End-to-End Vision-and-Language Transformerscited by 271
  7. 2021Bridge Data: Boosting Generalization of Robotic Skills with Cross-Domain Datasetscited by 248
  8. 2023AdaMerging: Adaptive Model Merging for Multi-Task Learningcited by 248
  9. 2024Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunitiescited by 247
  10. 2023Efficient Diffusion Training via Min-SNR Weighting Strategycited by 242
  11. 2022Contrastive Learning as Goal-Conditioned Reinforcement Learningcited by 241
  12. 2023Rewarded soups: towards Pareto-optimal alignment by interpolating weights fine-tuned on diverse rewardscited by 232
38 more
  1. 2021MT-Opt: Continuous Multi-Task Robotic Reinforcement Learning at Scalecited by 226
  2. 2021Efficiently Identifying Task Groupings for Multi-Task Learningcited by 218
  3. 2022Multi-Task Learning as a Bargaining Gamecited by 209
  4. 2022BEVerse: Unified Perception and Prediction in Birds-Eye-View for Vision-Centric Autonomous Drivingcited by 191
  5. 2021Fishr: Invariant Gradient Variances for Out-of-Distribution Generalizationcited by 186
  6. 2021Multi-Task Reinforcement Learning with Context-based Representationscited by 169
  7. 2021Distribution Matching for Heterogeneous Multi-Task Learning: a Large-scale Face Studycited by 168
  8. 2023RoboAgent: Generalization and Efficiency in Robot Manipulation via Semantic Augmentations and Action Chunkingcited by 167
  9. 2021ExT5: Towards Extreme Multi-Task Scaling for Transfer Learningcited by 161
  10. 2024Personalizing Reinforcement Learning from Human Feedback with Variational Preference Learningcited by 147
  11. 2024VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Taskscited by 136
  12. 2023Diffusion Model is an Effective Planner and Data Synthesizer for Multi-Task Reinforcement Learningcited by 114
  13. 2023Task-Specific Skill Localization in Fine-tuned Language Modelscited by 104
  14. 2021Stabilizing Deep Q-Learning with ConvNets and Vision Transformers under Data Augmentationcited by 101
  15. 2021Reasonable Effectiveness of Random Weighting: A Litmus Test for Multi-Task Learningcited by 100
  16. 2022M3ViT: Mixture-of-Experts Vision Transformer for Efficient Multi-task Learning with Model-Accelerator Co-designcited by 97
  17. 2024Representation Surgery for Multi-Task Model Mergingcited by 95
  18. 2021Reset-Free Reinforcement Learning via Multi-Task Learning: Learning Dexterous Manipulation Behaviors without Human Interventioncited by 90
  19. 2023Multiple Physics Pretraining for Physical Surrogate Modelscited by 90
  20. 2021Continual Learning for Text Classification with Information Disentanglement Based Regularizationcited by 86
  21. 2021Multi-Task Self-Training for Learning General Representationscited by 85
  22. 2023Diffusion Models for Reinforcement Learning: A Surveycited by 84
  23. 2021RotoGrad: Gradient Homogenization in Multitask Learningcited by 82
  24. 2022Learning and Retrieval from Prior Data for Skill-based Imitation Learningcited by 77
  25. 2024Data Shapley in One Training Runcited by 76
  26. 2023FAMO: Fast Adaptive Multitask Optimizationcited by 74
  27. 2022Auto-Lambda: Disentangling Dynamic Task Relationshipscited by 73
  28. 2022Offline Q-Learning on Diverse Multi-Task Data Both Scales And Generalizescited by 72
  29. 2024MMPareto: Boosting Multimodal Learning with Innocent Unimodal Assistancecited by 72
  30. 2022NeRF-SOS: Any-View Self-supervised Object Segmentation on Complex Scenescited by 71
  31. 2023Independent Component Alignment for Multi-Task Learningcited by 69
  32. 2024Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradientcited by 69
  33. 2022Improving Test-Time Adaptation via Shift-agnostic Weight Regularization and Nearest Source Prototypescited by 67
  34. 2022In Defense of the Unitary Scalarization for Deep Multi-Task Learningcited by 65
  35. 2021Conservative Data Sharing for Multi-Task Offline Reinforcement Learningcited by 64
  36. 2023Massive Editing for Large Language Models via Meta Learningcited by 62
  37. 2021Multi-Task Learning in Natural Language Processing: An Overviewcited by 61
  38. 2021Domain Generalization via Gradient Surgerycited by 60

50 most cited of 1,210 All 1,210 (JSON)