https://github.com/andreacossu/continuallearning-sequentialprocessing
Continual Learning with Gated Incremental Memories for Sequential Data Processing. IJCNN 2020. Continual Learning with Recurrent Neural Networks (RNNs) inspired by Progressive network architecture.
https://github.com/andreacossu/continuallearning-sequentialprocessing
audioset continual-learning machine-learning pytorch recurrent-neural-networks rnn
Last synced: 6 months ago
JSON representation
Continual Learning with Gated Incremental Memories for Sequential Data Processing. IJCNN 2020. Continual Learning with Recurrent Neural Networks (RNNs) inspired by Progressive network architecture.
- Host: GitHub
- URL: https://github.com/andreacossu/continuallearning-sequentialprocessing
- Owner: AndreaCossu
- License: gpl-3.0
- Created: 2020-01-23T08:47:30.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2021-10-13T07:35:56.000Z (almost 4 years ago)
- Last Synced: 2025-03-28T03:32:35.521Z (6 months ago)
- Topics: audioset, continual-learning, machine-learning, pytorch, recurrent-neural-networks, rnn
- Language: Python
- Homepage:
- Size: 193 KB
- Stars: 15
- Watchers: 2
- Forks: 4
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# ContinualLearning with Gated Incremental Memories for sequential data processing
[Paper](https://arxiv.org/abs/2004.04077) accepted at IJCNN 2020.
## MNIST task
Run the script `mnist.py` with your hyperparameters of choice.
## Audioset task
Download `bal.h5`, `eval.h5` and `unbal_train.h5` from [here](https://drive.google.com/drive/folders/1IlsVeAD9iAhK1Keu958RR8hXd2rcRnq5?usp=sharing) and put them in `tasks/audioset/data/`.
Then, run `audioset_task.py` with your hyperparameters of choice.
## Devanagari task
Download Devanagari dataset from [here](https://drive.google.com/file/d/1dcP0m02bRyKGebZxwq_jMifuTZsXk5RJ/view?usp=sharing) and put `Train` and `Test` folder inside `tasks/mnist/data/Devanagari_CL/`.
Then, run `mnist.py --devanagari` with your hyperparameters of choice.