Comments (10)
这些都是要case by case讨论的,MTL research在optimization的研究一般都是基于最简单的HPS结构,而实际场景中的网络会更加复杂,可能导致这些optimization的方法不work
from libmtl.
大佬,那在多任务里,除了梯度策略、网络架构,还有其它对提点有帮助的技术吗
from libmtl.
根据实际场景的问题去设计特定的网络结构应该是最有效的方法,但这并不通用
from libmtl.
我理解多任务学习要解决的最主要的问题就是任务间的冲突。
根本解决的办法是设计一种网络,共享部分是多任务的耦合部分,每个任务独享的网络部分是任务独有的特征(任务间的冲突特征),那就会对模型预测效果有最明显的提升,是吗?
from libmtl.
根据实际场景的问题去设计特定的网络结构应该是最有效的方法,但这并不通用
在特定问题下可以利用问题的先验信息去设计网络结构,但这并不是通用方法,比如MTAN引入了attention可以解决很多CV的感知任务但很难被用于其他领域比如NLP。相比之下,optimization的研究是比较通用的
from libmtl.
根据实际场景的问题去设计特定的网络结构应该是最有效的方法,但这并不通用
在特定问题下可以利用问题的先验信息去设计网络结构,但这并不是通用方法,比如MTAN引入了attention可以解决很多CV的感知任务但很难被用于其他领域比如NLP。相比之下,optimization的研究是比较通用的
@Baijiong-Lin
大佬,在CV的感知任务这一块,除了MTAN的attention模块,其它还有推荐的吗?比如MMOE这种门电路的结构可以用吗,或者其它的。
from libmtl.
MMoE理论上肯定能用,但效果未知。你可以看看这篇survey:Multi-Task Learning for Dense Prediction Tasks: A Survey
from libmtl.
MMoE理论上肯定能用,但效果未知。你可以看看这篇survey:Multi-Task Learning for Dense Prediction Tasks: A Survey
看了,我做cv的感知,想试试MTAN和PAD-Net。这篇综述里也说了,优化策略还不如超参数的网格搜索。我把你开源的optimization都试了个遍,结果发现最有效的居然是GLS。您对类似yolop,yolopv2这一类硬参数共享的文章在不大幅增加FLOPS的基础上,有啥改进建议嘛?
yolop:https://arxiv.org/pdf/2108.11250.pdf
yolopv2:https://arxiv.org/pdf/2208.11434.pdf
from libmtl.
在我们的实验里,GLS在有些dataset上表现不错但在另一些dataset上表现不佳。我对yolop这类的工作不太熟悉。
from libmtl.
谢谢
from libmtl.
Related Issues (20)
- for GNN HOT 12
- How to transfer to other projects.for example,yolov5 HOT 2
- gpu_id is setted as 1, but it still runs on gpu 0. Why is the gpu_id setting invalid? HOT 1
- saving model HOT 1
- Question about Cross stitch implementation HOT 1
- Issue with val Epoch and `self.base_result` HOT 2
- 更换损失函数时的两个问题 HOT 6
- Performance issue for NYUv2 HOT 4
- Support for Visualization HOT 2
- v1.1.6文档 HOT 4
- MSELoss missing
- PLE模型中如何使用MGDA-UB的weighting HOT 4
- Gradvac的梯度更新 HOT 4
- 我想问问分支和主线的区别,另外我想问问是否计划实现ubernet HOT 4
- 可视化工具 HOT 2
- nashmtl修改更新次数后报错 HOT 24
- question about uw HOT 2
- 怎么样求得网络公共部分 HOT 1
- 大佬考虑过支持mscoco数据集么,可以把检测加进来这种,另外最近新出了一篇论文,mod-squad,感觉想法很好 HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from libmtl.