- Implemented CNN using PyTorch and Keras frameworks on Dogs vs Cats dataset.
- Showed validation curves and tuned hyperparams for Keras CNN.
- Used Transfer Learning idea to get 96% accuracy (used Xception and then fine tuned with low learning rate).
- Trained U-Net and FPN models for segmentation problem on Flood Segmentation dataset.
- Used pretrained encoder (efficientnetb2 NN) on ImageNet dataset.
- Showed Loss and IoU score curves and predicted mask on test example for both models.
- Data preprocessing (lemmatizing, removing urls, html tags, stopwords and using bigrams).
- Trained Bidirectional LSTM, showed validation curves and tested on self-made reviews.
- Trained Bidirectional LSTM with Word2Vec embeddings (embeddings trained on IMDB dataset), showed validation curves and tested on self-made reviews.
- Used t-SNE to visualize Word2Vec embeddings.
- Trained Bidirectional LSTM with pre-trained FastText embeddings from gensim, showed validation curves and tested on self-made reviews.
- Trained BERT model using PyTorch framework.