The Objective: The Electronics and Communication Sciences Unit, Indian Statistical Institute, Kolkata is organizing the Winter School on Deep Learning: From Perceptrons to Transformers. This winter school will focus heavily on imparting a hands-on experience towards developing a wide range of classical and advanced deep learning models, in addition to making the associated theory easy to understand. Participants will learn from the basics of machine learning to the advanced deep learning-based approaches with application to Computer Vision and Natural Language Processing. Theoretical lectures will be delivered by renowned professors and scientists (from ISI and other esteemed organizations) who have made significant contributions in their areas of research. The lectures will be supplemented by extremely detailed hands-on sessions instructed by post-docs and research scholars.

Course coverage: The winter school will have the following course structure (theory and associated hands-on)

• Basics of Python

• Basics of the Deep Learning Library: PyTorch

• Essentials of Vector Calculus and Linear Algebra for Machine Learning

• Conceptual Fundamentals of Machine Learning, Image Processing, Computer Vision, Natural Language Processing

• Perceptrons and Backpropagation

• Ingredients of Deep Learning: Gradient Descent, Batch Normalization, Regularization, Dropout

• Convolutional Neural Networks (CNN), Convolutional Autoencoders

• CNN for Object Classification, Detection, and Segmentation

• Recurrent Neural Network, LSTM, Word Embedding

• Attention Models and Transformer (BERT and Visual Transformer)

• Deep Generative Models (GAN and VAE)

• Weakly Supervised Deep Learning, Self-Supervised Learning

• Meta-Learning and Few-Shot Learning

• Deep Reinforcement Learning

• Explainable Artificial Intelligence

• Geometric Deep Learning

Mode of tutorials: Lectures and Hands-on sessions will be conducted in online mode only. All sessions will be on Fridays and Saturdays, and the recordings will be shared with all the participants.