This course will introduce the theoretical foundations, algorithms,
methodologies, and applications of neural networks and deep learning. It will help to design and develop an application-specific deep learning models and also provide the practical knowledge handling and analyzing real world applications. To impart knowledge on the basic principles and concepts in digital image and video processing. To explore and demonstrate real time image and video analytics in solving practical problems of commercial and scientific interests.
Module 1: INTRODUCTION TO DEEP LEARNING & ARCHITECTURES
What is deep learning? Machine learning vs. Deep learning, Types of neural network? Its application History of perceptron Linearity and non-linearity of data Artificial Neural Networks (ANN) - Functions in ANN – Activation function, Loss function, classification / Regression Problems – Applications
Module 2: LEARNING IN DEEP NETWORKS
Optimizers-Gradient descent, Stochastic Gradient Descent, Mini Bath Gradient Descent, Gradient Descent with Momentum, Adaptive Gradient, Adadelta-RMSprop, Adaptive Moment Estimation. Forward and Backward propagation Chain rule, Weight Initialization methods. Module 3: LEARNING IN DEEP NETWORKS- 2 Activation function- Sigmoid, tanh, Relu, softmax, prelu, softmax, etc. Regression Loss functions- o Mean Squared Error, o Mean Absolute Error, o Huber Loss, o Logarithmic Error. Classification Loss functions- o Binary Crossentropy, o Categorical Crossentropy, o sparse categorical Crossentropy, o Hinge loss Vanishing Gradient/ Exploding Gradient Dying Relu Problem
Module 4: BUILDING & IMPROVING DEEP NEURAL NETWORKS
Hyper-parameter tuning, Regularization – Dropouts. Data Augmentation, Stratification Generalization Gap – Under-fitting Vs Over-fitting. Optimization – Momentum, Learning rate schedules, AdaGrad, RMSProp and Adam optimization. Initialization – weights, Bias.
Module 5: CONVOLUTION NEURAL NETWORKS
CNN architecture, Pixel, RGB concept, Image structure. Convolution layer Data augmentation, convolution, stride, Padding & types, Pooling and types Edge detection Transfer Learning Architecture Advanced architectures: AlexNet, ResNet, VGG16 and others. Popular CNN Model Architectures Module 6: COMPUTER VISION FOUNDATIONS Basic steps of Image processing system – Pixel relationship- Image Transforms-. Image Enhancement- Histogram Processing, Spatial filtering, Frequency Domain filtering Image Processing - Colour. Linear Algebra Primer - Pixels and Filters - Edge Detection - Features and Fitting. Feature Descriptors - Image Resizing – Segmentation
Module 7: IMAGE SEGMENTATION, COMPRESSION AND COLOUR
Module 8: OBJECT RECOGNITION AND IMAGE/VIDEO RETRIEVAL
Video Analysis Action Recognition Video based rendering Context and scene understanding Case Study: Surveillance
Module 9: UNSUPERVISED DEEP LEARNING
Autoencoders (Variational Autoencoders (VAEs), Denoising Autoencoders) Generative Adversarial Networks (GANs) DCGAN, WGAN, CycleGAN & Pix2Pix They have gained a lot of attention because they can create fake content. This technology is considered a revolution in the field of Artificial Intelligence for producing high quality results, remaining one of the most popular and relevant topics. Self-Organizing Maps (SOMs)
Module 10: RECENT TRENDS & REAL TIME PROJECTS ON EVERY
Digital Image Processing EC801B Contacts: 3L Credits: 3 Objective: The Course Provides Grounding in Digital Filter and Transforms Techniques For Image Processing
Digital Image Processing EC801B Contacts: 3L Credits: 3 Objective: The Course Provides Grounding in Digital Filter and Transforms Techniques For Image Processing
