多任务学习

最新推荐文章于 2023-03-30 19:04:02 发布

zhurui_xiaozhuzaizai

最新推荐文章于 2023-03-30 19:04:02 发布

阅读量391

点赞数 1

CC 4.0 BY-SA版权

分类专栏：自然语言处理文章标签：多任务学习

本文链接：https://blog.youkuaiyun.com/weixin_36378508/article/details/115694966

自然语言处理专栏收录该内容

106 篇文章

订阅专栏

综述
An Overview of Multi-Task Learning in Deep Neural Networks.pdf
https://zhuanlan.zhihu.com/p/269492239

动态loss:
多任务学习之动态权重
MMOE
SNR
Multi-task learning using uncertainty to weigh losses for scene geometry and semantics
简介：https://zhuanlan.zhihu.com/p/65137250utm_source=wechat_session
https://blog.youkuaiyun.com/cdknight_happy/article/details/102618883
https://zhuanlan.zhihu.com/p/65137250utm_source=wechat_session

Multi-Task Learning as Multi-Objective OptimizationAuxiliary Tasks in Multi-task Learning
TF实现:https://github.com/ranandalon/mtl
PYTORCH实现:https://github.com/oscarkey/multitask-learning/blob/c4503c044ca7a29bebd4e70e9e030524654e5d00/multitask-learning/cityscapestask/losses.py#L9
A PyTorch implementation of Liebel L, Körner M. Auxiliary tasks in multi-task learning[J]. arXiv preprint arXiv:1805.06334, 2018.
https://github.com/Mikoto10032/AutomaticWeightedLossother:

dialog1:
https://git.100tal.com/jituan_AILab_Alg_DM_NLP/auto_fill_blanks_english/tree/tbc_2transformers
就这两块做了修改吧，这loss感觉就像是你对一个文本打了两个标签，最后计算两个标签的loss的和。

如何使用不确定性来解释你的模型
Using Uncertainty to Interpret your Model
https://www.sohu.com/a/245880658_114877
What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision? https://zhuanlan.zhihu.com/p/100998668https://blog.youkuaiyun.com/weixin_39779106/article/details/78968982