Mtl Implem

Implementation of some multitask learning algorithms. More coming.
Alternatives To Mtl Implem
Project NameStarsDownloadsRepos Using ThisPackages Using ThisMost Recent CommitTotal ReleasesLatest ReleaseOpen IssuesLicenseLanguage
a month ago6apache-2.0Python
pix2code: Generating Code from a Graphical User Interface Screenshot
a year ago27otherPython
StyleGAN2 - Official TensorFlow Implementation
2 months ago7mpl-2.0Jupyter Notebook
:robot: :speech_balloon: Deep learning for Text to Speech (Discussion forum:
Transformers Tutorials3,823
23 days ago142mitJupyter Notebook
This repository contains demos I made with the Transformers library by HuggingFace.
2 years ago31Python
[CVPR2020] Adversarial Latent Autoencoders
5 years ago42Python
CNNs for sentence classification
Surface Defect Detection1,614
2 months ago11mitPython
📈 目前最大的工业缺陷检测数据库及论文集 Constantly summarizing open source dataset and critical papers in the field of surface defect research which are of great importance.
Tacotron 21,572
3 years ago230mitPython
DeepMind's Tacotron-2 Tensorflow implementation
19 days ago25apache-2.0Python
Code for the paper: Detecting Photoshopped Faces by Scripting Photoshop
3d Pointcloud1,374
5 days ago2Python
Papers and Datasets about Point Cloud.
Alternatives To Mtl Implem
Select To Compare

Alternative Project Comparisons


This is a matlab implementation of the paper bellow: Learning Incoherent Sparse and Low-Rank Patterns from Multiple Tasks

The script to run is called main. All the dataset, but Ar face dataset are available in the folder 'datasets'. Ar face dataset is too big to put in the archive. And also it is not public.

However one could ask for the dataset in here

Im the main script, the variable dataname allows one to choose the dataset to use. There is 3 choices: 'face' 'scene' 'school'

If 'face' is choosen, one should know that since we treat images, the algorithm is very time consuming. To avoid spending a day on training,'max_iter' is restrained to 1000.

The results are quite similar to the paper. However the parameters were not crossvalidated like in the paper, An heuristic choice of the parameters has been made.

The results for the feature extraction can be seen in the folder, feature extraction. Each picture shows the P (left) and Q (right) matrix learned. We can see that the sparse matrix captures details of the face. The low rank matrix on the other hand captures the general shape of the face. Example bellow:

alt tag

For the classification results of the scene dataset, a screendhot is furnished bellow, however those result can be reproduced by runing main with scene dataset. We can see that those results are quite similar to the ones displayed in the paper.

alt tag

Popular Dataset Projects
Popular Paper Projects
Popular Data Processing Categories

Get A Weekly Email With Trending Projects For These Categories
No Spam. Unsubscribe easily at any time.