Academic Projects

Reproduction of “A Simple Framework for Contrastive Learning of Visual Representations” Dec 2023

• Supervisor: Dr. Mehmet Kerem Turkcan
• Implemented a SimCLR model using TensorFlow 2 and Keras for contrastive learning on CIFAR-100.
• Developed custom data augmentation techniques and NT-Xent loss function for self-supervised learning.
• Utilized transfer learning with ResNet architecture, trained on Google Cloud Platform in a team of three.

Predicting Apple Inc.’s Stock Closing Price with LSTM Apr 2023

• Supervisor: Dr. Gabriel Young
• Developed an LSTM model using PyTorch to forecast Apple’s stock prices based on 5 years of historical data.
• Implemented data preprocessing, including MinMaxScaler normalization and time series decomposition.
• Achieved accurate predictions (RMSE: 7.48) using rolling forecast validation on an 80-20 train-test split.

Weakly Supervised Learning: Label Noise and Correction May 2023

• Supervisor: Ying Liu
• Developed an image classification model for CIFAR-10 dataset with noisy labels using PyTorch.
• Implemented a label correction approach to address label noise, improving model accuracy.
• Achieved improved performance over baseline, balancing computational efficiency and prediction accuracy.