1. Predict the labels of a YouTube video using extracted frame-level and video-level features.
2. Large 1.5 Terabyte dataset containing 6 million examples.
3. Implemented existing state of the art deep learning architectures such as InceptionNet, Squeeze-Excitation Resnet, and SqueezeNet in Tensorflow on Google Cloud platform.
4. Applied deep learning techniques such as hyperparameter tuning, Regularization and Optimization.
5. Developed a hybrid architecture which combined SE-Resnet with LSTMs and mixture of experts to achieve a Global Average Precision (GAP) score of 0.83, a competitive score compared to the top score of 0.89.