Skip to content

A multiple instance learning technique for training CNNs using coarse/weak labels.

Notifications You must be signed in to change notification settings

shaziaakbar/cluster-mil

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

cluster-mil


Description:

cluster-mil is a weakly supervised learning technique which uses concepts from the Multiple Instance Learning (MIL) framework to train a convolutional deep neural network. As data is only available in batches, we pretrain a variational autoencoder (unsupervised) and then estimate class labels from weak labels provided during training.

This work was presented at ML4H @ NeurIPS 2018 and a full description of the method is available on ArXiv: Cluster-Based Learning from Weakly Labeled Bags in Digital Pathology

MNIST-BAG

Alt text

The implementation of the MNIST-BAG experiment is given in mnist-bag.py and utilizes our novel loss function (loss.py). Pretrained autoencoder models are also provided in /latent_models.

About

A multiple instance learning technique for training CNNs using coarse/weak labels.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages