PROJECTS
MusicComp: Original Music Composer
-
Developed a program which can compose original music pieces including several instruments such as piano, violin, and saxophone. ·
-
A generative adversarial neural network, whose discriminator and generator parts are recurrent neural networks (RNN-GAN), is utilized. ·
-
Developed program can also produce music according to the selected emotional theme. ·
-
Instruments which appear in music can also be selected by the user.
ImStyle: Painter Detection and Restyling
-
By using deep learning, developed a program that restyles the given image of the user according to the style of a painter selected (Van Gogh’s, Monet’s, etc.). ·
-
For the restyling process, deep convolutional generative adversarial networks (DC-GAN) are utilized. ·
-
Additionally, developed a program that predicts the painter of the given artistic piece by using convolutional neural networks (CNN).
Cryptocurrency Stock Prediction.
-
Developed a program that could predict the future stock prices of selected cryptocurrencies. ·
-
For prediction, I have trained a deep recurrent neural network with past values of cryptocurrencies obtained from coinmarketcap.com. ·
-
The accuracy of the program is 85% for the test data and 83% for the training data.
Chess Player
-
Developed a chess-playing program by using reinforcement learning which learned to play chess with trial and error. ·
-
I have used TensorFlow libraries in Python for coding deep reinforcement learning code. ·
-
As deep reinforcement learning method, deep recurrent Q-Learning (DRQN) was used.