AI with Deep Learning

Artificial Neural Networks (ANN) are inspired by biological neurons and utilize a collection of artificial neurons stacked and connected in layers to model complex data. This week dives deeper into the underlying functionality of neural networks and how to use common, open-source Deep Learning (DL) libraries Keras and Tensorflow to build neural networks to solve business problems.

• Deep Learning and history
  
• Multi-layer perceptron
  
• Types of Activation functions
  
• Training a neural network
  
• Backpropagation
  

Different optimization algorithms are used in practical setups to improve the performance of neural networks, and deploying regularization techniques like dropout and batch normalization helps ensure neural networks gain a better understanding of the patterns in complex data. In this week, we dive deeper into the concepts of optimization and regularization to illustrate the best practices for building generalized and robust neural networks to solve business problems.

• Optimizers and their types 
• Weight initialization and its techniques 
• Regularizations and its techniques 
• Types of neural networks

Analyze the data provided by a wind energy provider regarding equipment health, build various neural network models, and find the best one that will help identify failures so that the equipment can be repaired before failing/breaking and the overall maintenance cost can be brought down. 

Tools & Concepts- Exploratory Data Analysis, Data Preprocessing, Tensorflow, Keras, Artificial Neural Networks and Regularization

Natural Language Processing (NLP) is a branch of AI that focuses on processing and understanding human language to facilitate the interaction of machines with it. This week, participants get introduced to the world of NLP, different text processing and vectorization techniques, and steps to combine text preprocessing with Machine Learning techniques for solving NLP problems

• Overview of NLP 
• History of NLP 
• Different applications of NLP 
• Text Cleaning 
• Text Vectorization 
• Sentiment Analysis

Word embeddings allow us to numerically represent complex textual data, thereby enabling us to perform a variety of operations on them. This week’s topics introduces participants to different word embedding techniques and the steps involved in designing and implementing hands-on solutions combining word embedding methods with Machine Learning techniques for solving NLP problems.

• Introduction to word embeddings 
• Word2Vec, CBOW and Skip-gram 
• Negative Sampling 
• GloVe 
• Applications of Word Embeddings

Transformers are neural network architectures that develop a context-aware understanding of data and have revolutionized the field of NLP by exhibiting exceptional performance across a wide variety of tasks. This week, we dive into the underlying workings of different transformer architectures and how to use them to solve complex NLP tasks.

• Introduction to Transformers 
• Components of a Transformer 
• Different Transformer Architectures 
• Applications of Transformers
  

Large Language Models (LLMs) are ML models that are pre-trained on a large corpora of data and possess the ability to generate coherent and contextually relevant content. Retrieval augmented generation (RAG) combines the power of encoder and generative models to produce more informative and accurate outputs from a knowledge base. This module introduces LLMs, explains their working, and provides a thorough coverage of leveraging sentence transformers to encode data, vector databases to store and efficiently retrieve information from the encoded data, and LLMs to use the information to enhance the quality and relevance of the generated output.

• Introduction to LLMs 
• Working of LLMs 
• Applications of LLMs 
• Introduction to Embeddings and Tokenization 
• Byte-Pair Encoding (BPE) Tokenization 
• Computation and application of Sentence Embeddings 
• Retrieval-Augmented Generation (RAG)

Analyze the data of stock news and prices, develop an AI-driven sentiment analysis system that will automatically process and analyze news articles to gauge market sentiment, and summarize the news at a weekly level to help financial analysts optimize investment strategies and improve client outcomes.

Tools & Concepts - Exploratory Data Analysis, Word Embeddings, Transformers, Sentence Transformers, Large Language Models, Prompt Engineering and Sentiment Analysis

Computer Vision is a branch of AI that focuses on understanding and extracting meaningful insights from image data. This week, we cover an overview of the world of Computer Vision and the techniques to process images and extract meaningful patterns from them.

• Overview of Computer Vision 
• Color pixel and image representation 
• Edge Detection 
• Kernels 
• Padding 
• Strides and Pooling 
• Flattening to a 1D Array

Given the complex nature of image data, Convolutional Neural Networks (CNNs) are utilized to capture relevant spatial information in images. Transfer learning is a method to leverage the underlying knowledge needed to solve one problem to other problems. In this week, we cover the fundamentals of CNNs, how to build them from scratch, and how to leverage common CNN architectures via transfer learning to solve different image classification problems.

• ANN Vs CNN 
• CNN Architecture 
• Introduction to Transfer Learning 
• Common CNN Architectures

Build a robust image classifier using CNNs to efficiently classify different plant seedlings and weeds to maximize crop yields and minimize human involvement. 

Tools & Concepts - Image Processing, Image Classification, Keras, Tensorflow, Convolutional Neural Networks and Transfer Learning

Model deployment is the process of making a trained Machine Learning model accessible to a wider audience by operationalizing it. This module introduces participants to model deployment, provides an overview of its need in generating business value from ML models, and serializing and deploying ML models using Python libraries like Streamlit.

• Introduction to Model Deployment 
• Serialization 
• Deployment using Streamlit

Containerization is the process of packaging applications and their dependencies into self-contained units called containers to ensure consistent execution across different environments. This module dives into packaging ML models and their dependencies into containers using Docker and simplifying deployment of the ML models using Python libraries like Flask.

• Introduction to Containerization 
• Docker 
• Deployment using Flask

Analyze historical customer data, build a predictive model that predicts whether or not a customer will discontinue using a bank's credit card services, and identify the key factors affecting the customer’s decision. 

Tools & Concepts - Exploratory Data Analysis, Random Forest, Hyperparameter Tuning, Scikit-Learn, Model Deployment, Docker and Flask

This course will help you explore how to solve business problems by generating code using Generative AI tools, examine the capabilities of text-to-image and image-to-text GenAI tools like DallE through business use cases, and explore the speech recognition capabilities of audio-to-text GenAI tools like Whisper through business use cases.

This course will help you get introduced to recommendation systems and learn how to build recommendation systems that use past product purchase and satisfaction data to make high-quality personalized recommendations.

• Introduction to Recommendation Systems 
• Market Basket Analysis 
• Popularity-based and Content-based Recommendation Systems 
• Collaborative Filtering 
• Hybrid Recommendation Systems