Objective
Code Kata is defined as an exercise in programming which helps hone our skill through practice and repetition. In machine learning programming, Code Kata for implementing ML algorithms is very important, becuase we can realize the details ( such as Data Sampling, Weight initialization, various training strategy ...) while implementing the algorithm.
I implement various algorithms using in Deep Learning and organize them into scripts.. I'll update one script each week.
If you have a good topic, feel free to leave it on the issue! I will try to implement it as much as possible!
How to do the Code Kada together? (set-up environment)
Do not worry! I provide the environment written as a docker image.
# Run it From the root project directory
docker-compose up -d
Deep-Learning Katas List
Embedding Network for Handling Categorical Variables
Goals
- Implement a simple Deep Learning Model Handling variables by Embedding Layer
- Acheive Accuracy similar to that of a Random Forest Classifier
Dataset
Maxout Network using Tensorflow
Goals
- Implement Maxout Activation Layer using Keras custom Layer
- Check the performance by learning fashion-MNIST
Dataset
- Fashion-MNIST
papers
two ways for visualizing CNN, guided back propagation and GradCAM
Goals
-
Implement two methods to interpret the output of CNN
-
Algorithm implementation based on the operation of Resnet50v2
papers
CRNN(Text Recognition) Using Tensorflow <--TODO
Goals
- Implement CRNN, Deep Learing Model for reading text in images.
Datasets
- synthetic dataset using MNIST
Papers
Shake-Shake Regularization using Tensorflow <-- TODO
Goals
- Implement Shake-Shake Regularization which directly applies noise inside the Deep Learning Model
Datasets
- CIFAR-10
Papers
CopyRight
This repository is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.