Deep Learning for AI: Foundations, Architecture & Applications

Deep Learning for AI: Foundations, Architecture & Applications
(In Association with iHUB Divyasampark IIT Roorkee)

About this Course: 

Course Overview:

Module 1: Introduction to Deep Learning & Neural Networks

• Introduction to Deep Learning & Neural Networks.
• Artificial Intelligence, Machine Learning & Deep Learning.
• Evolution & Applications of Deep Learning.
• Traditional ML vs Deep Learning.
• Need of Representation Learning & Deep Learning.
• Fundamentals of Artificial Neural Networks-biological vs artificial Neurons; weight and Bias.
• Overview of types of Neural Networks.
• Overview of Deep Learning workflow: Data->Model-> Training->Evaluation->Prediction.
• Hands-on: Implement a Simple Artificial Neuron.
  
  ·

Module 2: Machine Learning Foundations for Deep Learning

• Supervised vs Unsupervised Learning; Training, validation & Test Datasets.
• Fundamentals of Learning: Features, Targets, Model Parameters & Loss Functions.
• Gradient Descent: Learning Rate, Batch & Mini-batch Learning.
• Regression & Classification: Linear Regression (Single & Multivariable), Polynomial Regression, Logistic Regression, Sigmoid Function & Decision Boundary.
• Model Generalization: Overfitting, Underfitting & Bias-Variance Trade-off.
• Model Evaluation: Mean Absolute Error (MAE); Root Mean Squared Error (RMSE); Accuracy; Precision; Recall, F1-Score; Confusion Matrix.
• Introduction to Regularization & Its Role in Deep Learning.
• Hands-on: Build and evaluate a regression/classification model using Python.

Module 3: Artificial Neural Networks (ANN) & Deep Learning Fundamentals

• Perceptron Models: Weight & Bias; Linear & Non-linear Models; 
• Neural Network Architecture: Layers, Neurons & Connections.
• Activation Functions- Sigmoid, Tanh, ReLU, Leaky ReLU, Softmax
• Loss Functions: MSE, Binary Cross-Entropy, Categorical Cross-Entropy
• Forward Propagation: Network prediction & Loss Calculation.
• Back propagation: Chain Rule; Gradient Descent; Error Propagation; Weight & Bias Updates.
• Weight Initialization: Random Initialization; Xavier/Glorot and He Initialization.
• Training Neural Networks: Epoch; Batch; Mini-batch; Iteration; Learning rate; Optimizers.
• Optimization: SGD, Momentum, RMSProp, Adam
• Regularization & Training Techniques: Overfitting & Underfitting of Deep Learning Models; Dropout & Weight Regularization, Batch Normalization.
• Hands-on: Build & Train a Neural Network from scratch.
   

Module 4:  PyTorch Fundamentals & Convolutional Neural Networks (CNN)

• Introduction to PyTorch
• Tensors & Tensor Operations; Optimizers & Loss Functions
• Building Neural Networks with PyTorch; Training & Validation Loops.
• Basics of Convolutional Neural Networks (CNN)
• Image Fundamentals: Digital images; Pixels; RGB and grayscale images; Image dimensions; Image tensors.
• CNN Architecture: Why CNNs; Local connectivity; Convolution operation; Kernels/filters; Feature maps; Stride; Padding.
• Pooling: Max Pooling, Average Pooling, Downsampling.
• Important CNN Architectures: LeNet, AlexNet etc.
• CNN applications in Computer Vision: Image Classification; Feature Extraction & Object Recognition.
• Hands-on: Image Classification project: CIFAR-10 Image Classification using CNN.
   

Module 5:  Deep Learning for Sequential & Temporal Data

• Sequential Data; Time-Series & Text Data; Difference between static and sequential data.
• Recurrent Neural networks (RNN): Motivation for RNN; Recurrent neurons; Hidden state; Unrolling an RNN; Forward propagation through time. 
• RNN Architectures: Sequence-to-Sequence; Sequence-to-Vector; Vector-to-Sequence; Many-to-Many Architecture.
• Problems with Vanilla RNN: Vanishing gradients; Exploding gradients; Long-term dependencies; Need for LSTM & GRU.
• AutoEncoders: Encoder; Latent representation; Decoder; Reconstruction loss; Dimensionality reduction; Denoising autoencoders; Applications
• Generative Adversarial Networks (GAN): Generator; Discriminator; Adversarial training; GAN architecture; Training challenges; Mode collapse
• Hands-on: Sentiment Classification using RNN.
   

Module 6: Transformers & Attention Mechanism

• Attention Mechanism: Limitations of RNN/LSTM, Query-Key-Value (QKV).
• Attention scores and weights, scaled dot-product attention, self-attention.
• Transformer-based Natural Language Processing (NLP): Tokenization, embeddings, contextual representations, sequence classification and text generation..
• Transformer Architecture: Encoder–Decoder architecture, multi-head attention, positional encoding

Module 7: Pre-Trained Transformers & Deep Learning Applications

• Introduction to Pre-trained Transformer Models and Transfer Learning.
• Understanding GPT, Open AI API, Generating Text, Customizing GPT output.
• Key word text summarization, coding a simple chatbot using LangChain in Python
• Hugging Face: Transformer Pipeline, Pre-trained tokenizers, Special tokens
• Q&A models: BERT architecture, Tokenizer, Embeddings, Calculating response
• Creating QA bot, BERT, RoBERTa, DistilBERT, GPT vs BERT vs XLNET, XLNET Embeddings & Fine Tuning.
• Hands-on: Build a Transformer-based text classification application using a pre-trained model.
   

Prerequisites and eligibility:

• Basic knowledge of Python programming is necessary for this course.
• This course can be taken up by any undergraduate/postgraduate student of Basic & Applied Sciences, Engineering, Management and Computer Applications and also by Research Scholars/Faculties/Working Professionals who want to upskill themselves.
• Participants need to have a laptop/PC (with a minimum of 4 GB RAM, 100 GB HDD, Intel i3 processor) and proper internet/Wi-Fi connection.

Contact Person: Dr. Subrat Kotoky

Email: [email protected] / [email protected]

Phone: 9085317465 / 8473874389

Expert Profile: 

Mr. Shreyas Shukla
            Professional Corporate Trainer & Microsoft Azure Certified Data Engineer
            MTech-IIT Kharagpur & BE- The Aeronautical Society of India, New Delhi
             4+ years of experience in leading online professional courses for different
             leading organizations

 Has successfully conducted 30+ courses and trained 3000+ learners in the fields of   Python Programming, Data Analytics, Machine Learning, Deep Learning, Computer Vision etc., Generative AI till now. 

           Certifications:   
1. DP-203: Microsoft Certified: Azure Data Engineer Associate
2. DP-900: Microsoft Certified: Azure Data Fundamentals
3. AZ-900: Microsoft Certified: Azure Fundamentals

Register Now