Deep Learning and Neural Network Training & Certification

• Weekdays – 7 to 8 weeks
• Weekend – 9 to 10 weeks
• FastTrack – 6 to 7 weeks

• Weekdays – 7 to 8 weeks
• Weekend – 9 to 10 weeks
• FastTrack – 6 to 7 weeks

• Role of Machine Learning in field of Artificial Intelligence
• Deep Learning v/s Machine Learning
• Brief History of AI Recap: SL, UL and RL
• Classification and regression in supervised learning
• Clustering and association in unsupervised learning
• Algorithms that are used in these categories
• Introduction to AI and neural networks
• Deep Learning: Successes Last Decade
• What is AI and Deep Learning
• Discussion: Self-Driving Car Object Detection
• Applications of Deep Learning
• Challenges of Deep Learning
• Discussion: Sentiment Analysis Using LSTM
• Fundamentals of statistics
• Hypothesis testing
• Probability distributions
• Hidden Markov models

• Multi-layer network introduction
• Regularization
• Deep neural networks
• Multi-layer perceptron
• Overfitting and capacity
• Neural network hyperparameters
• Logic gates
• Different activation functions used in neural networks: - ReLu, Softmax, Sigmoid, and hyperbolic functions
• Back propagation, forward propagation, convergence, hyper parameters

• Various methods that are used to train artificial neural networks
• Perceptron learning rule
• Gradient descent rule
• Tuning the learning rate
• Regularization techniques
• Optimization techniques
• Stochastic process
• Vanishing gradients
• Transfer learning
• Dropout layer
• Regression techniques: - including Lasso L1 and Ridge L2
• Unsupervised pre-training
• Xavier initialization and more

• Understanding how Deep Learning works
• Activation functions
• Illustrating perceptron
• Perceptron training
• Multi-layer perceptron
• Key parameters of perceptron
• TensorFlow
• Tensorflow and Its Ecosystem
• Python libraries in TensorFlow : - code basics, variables, constants, placeholders, graph visualization
• Use-case implementation
• Keras

• Keras high-level neural network for working on top of TensorFlow
• Defining complex multi-output models
• Composing models using Keras
• Sequential and functional composition
• Batch normalization
• Deploying Keras with TensorBoard
• Neural network training process customization
• Hands On: Build a Deep Learning Model Using Keras
• Hands On: Build a Deep Learning Model Using Tensorflow

• Using TFLearn API to implement neural networks
• Defining and composing models
• Deploying TensorBoard

• Introduction to deep neural networks (DNNs)
• Several building blocks of artificial neural networks (ANNs)
• The architecture of DNN and its building blocks
• Reinforcement learning in DNN concepts
• various parameters, layers, and optimization algorithms in DNN
• activation functions

• Optimization Algorithms
• SGD, Momentum, NAG, Adagrad, Adadelta , RMSprop, Adam
• Hands on Exercise: MNIST Dataset
• Batch Normalization
• Understanding Exploding and Vanishing Gradients
• Hyperparameter Tuning
• Interpretability
• Width vs Depth

• What is a convolutional neural network?
• Understanding the architecture and use-cases of CNN
• Pooling layers
• Visualize CNN
• How to fine-tune a convolutional neural network
• Transfer learning
• Understanding recurrent neural networks
• Kernel filter
• Feature maps, and pooling
• Deploying convolutional neural networks in TensorFlow

• Introduction to the RNN model
• Use cases of RNN
• Modeling sequences
• Training RNNs with back propagation
• Long short-term memory (LSTM)
• Recursive Neural Tensor Network theory
• The basic RNN cell
• Unfolded RNN
• Dynamic RNN
• Time-series predictions

• GPU in Deep Learning v/s CPU in Deep Learning
• The significance of GPUs in training Deep Learning networks
• Forward pass and backward pass training techniques

• Introduction to RBM and autoencoders
• Deploying RBM for deep neural networks
• Using RBM for collaborative filtering
• Autoencoders’ features
• Applications of autoencoders
• Hands on: Autoencoder Model for MNIST Data

• What is Time Series
• Techniques and applications
• Components of Time Series
• Moving average
• Smoothing techniques
• Exponential smoothing
• Univariate time series models
• Multivariate time series analysis
• Arima model
• Time Series in Python
• Sentiment analysis in Python (Twitter sentiment analysis)
• Text analysis
• Hands-on Exercise – Forecasting feature value in a series after analysing time-series data and sequence of measurements to understand nature of phenomenon

• Time series techniques and applications
• Time series components
• Moving average, smoothing techniques, exponential smoothing
• Univariate time series models
• Multivariate time series analysis
• ARIMA model
• Time series in Python
• Sentiment analysis in Python (Twitter sentiment analysis)
• Text analysis
• Hands-on Exercise – Analysing and Forecasting using time series

• Image processing
• Natural Language Processing (NLP)
• speech recognition
• Video analytics

• Automated conversation bots leveraging any of the following descriptive techniques: IBM Watson, Microsoft’s Luis, Google API.AI, Amazon Lex, Open–Closed domain bots, Generative model
• Sequence to sequence model (LSTM)

Industry: - Stock market trading

Problem Statement: - Make a prediction model to predict price of stock

This project is to predict the volatility and price value of a stock by analysing the change with time and comparing multiple stocks with time. By implementing recurrent neural network, LSTM and time series you can make a predictive model which can generates the output close enough to the real stock prices in real time.

Industry: - Miscellaneous

Problem Statement: - Make a prediction model to predict handwritten characters or numbers.

Here you will get hands on experience with how to use features of images for building up predictive model. We will develop a CNN model using features of hand written images of characters or numbers to make a prediction over future input values by the user in the form of random images.

Industry: - Ecommerce

Problem Statement: - Build an Artificially intelligent chatbot

There is an Ecommerce platform wants to provide best in class services to the user through most interactive AI based chatbot. Here you will use NLP and neural network based model to understand the customer need and respond accordingly. It will be great hands on experience with Tensorflow components, natural language processing and querry handling.

Industry: - Search Engine

Problem Statement: - Build a model to search most relatable image over internet using the image given by user

This project will be build up using Tensorflow and CNN to best analyse an image after the training given to the model. You need to train a model, make the losses to least possible value and distribution of activation and gradients. Complete feature engineering over unstructured data set will be understood and practised on completion of this hands-on experience.