Projects

• Predicted imagined movements from raw electroencephalogram signals collected from 22 electrodes.
• Evaluated CNNs, LSTMs, spatiotemporal CNNs, GANs, and Transformers.
• A spatiotemporal CNN with an LSTM achieved the best accuracy of 74.7%.

Skills: Python · PyTorch · Deep Learning

• Crawled tweets about the Super Bowl game between the Patriots and Seahawks and performed lemmatization, feature extraction, dimensionality reduction, and word embedding generation.
• Built L1, L2, random forest, and perceptron models to classify tweet fan bases and predict retweet counts.
• Evaluated results using accuracy, F1 score, and AUC-ROC curves, achieving 90% inference accuracy.

Skills: Python · NLP · Machine Learning · Scikit-Learn · Pandas

• Built four image-colorization models using ResNet, U-Net, and generative adversarial networks.
• Implemented a data loader with suitable transforms and an inference script for grayscale images.

Skills: Python · PyTorch · Deep Learning · OpenCV

• Created a dataset from existing videos and built CNN, LSTM, HMM, and GAN models for visual speech recognition.
• Used generative adversarial networks for data augmentation, improving viseme-recognition accuracy by 3.7% over the baseline.

Skills: Python · PyTorch · Deep Learning