Introduction to Deep Learning
● Built a fully functional autograd driven DL framework with implementation for functional modules like Conv1D, Conv2D, LSTM, GRU and optimization algorithms including Adam, SGD and RMSprop.
● Trained MobileNet, ConvNext Net and ResNet models from scratch for classification of person’s ID based on dataset of face images and performed face verification using cosine similarity metric.
● Designed an end to end system for speech to text transcription using a combination of Recurrent Neural Networks (RNNs) and Attention models, such that the system was able to transcribe a given speech utterance to its corresponding transcript. The architecture for the project was inspired from Listen Attend and Spell Paper.
Learning Outcomes:
- Neural Networks As Universal Approximators
- Convolutional Neural Networks
- Time Series, Recurrent Networks and LSTMs
- Language Models and Sequence To Sequence Prediction
- Attention and Transformer
- GANs, Autoencoders and VAEs
Programming Language: Pytorch, Python