让多任务学习自动识别冲突并智能分配梯度，提升训练效果
Task Weighting through Gradient Projection for Multitask Learning
- 仅在任务冲突时动态调整梯度投影概率，否则自由更新
- 在nuScenes、CIFAR-100、CelebA上多数任务性能显著提升
- 适合需要自适应任务优先级的多任务场景
多任务学习中，任务梯度冲突常导致模型性能下降。现有方法如PCGrad通过梯度投影缓解此问题，通常能加快收敛并提升指标。本文提出一种新方法，将梯度投影算法扩展为同时实现任务优先级调节。与传统按损失缩放权重不同，该方法仅在任务存在冲突时启用加权机制，否则保持原训练流程。我们用概率分布代替固定权重，决定冲突情况下的梯度投影选择。在nuScenes、CIFAR-100和CelebA数据集上的实验表明，该方法是有效的任务加权策略。结合多种权重方案，相比均匀投影概率的PCGrad，多数任务的性能指标均有显著提升。
原文摘要 · Abstract (English)
In multitask learning, conflicts between task gradients are a frequent issue degrading a model's training performance. This is commonly addressed by using the Gradient Projection algorithm PCGrad that often leads to faster convergence and improved performance metrics. In this work, we present a method to adapt this algorithm to simultaneously also perform task prioritization. Our approach differs from traditional task weighting performed by scaling task losses in that our weighting scheme applies only in cases where tasks are in conflict, but lets the training proceed unhindered otherwise. We replace task weighting factors by a probability distribution that determines which task gradients get projected in conflict cases. Our experiments on the nuScenes, CIFAR-100, and CelebA datasets confirm that our approach is a practical method for task weighting. Paired with multiple different task weighting schemes, we observe a significant improvement in the performance metrics of most tasks compared to Gradient Projection with uniform projection probabilities.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。