In this course, we introduce Machine Learning, AI, and Data Science through a number of presentations and interactive sessions. The goal is to create a taxonomy and a way of thinking in terms of data and algorithms. We follow a case-based (learning by example and analogies) approach and avoid heavy mathematical notation, focusing more on illustrations and diagrams. The course is based on well established books by recognized experts, including  “The Master Algorithm”, “Algorithms to Live by”, “Predictive Analytics”, “Quest for Artificial Intelligence”, “Prediction Machines” and  “Super Crunchers”. Participants are provided a solid reference point for further investigation and for making statements based on solid and undisputable sources.

In this course, we introduce the basic concepts of data science. Participants work mainly with real datasets and learn about data representation. The focus is exploring and plotting data, discovering noise, and cleaning datasets. We emphasize the most critical parts of ML and AI, which are data preparation and feature engineering. With very simple models and rules, we teach participants how to address and solve some problems inspired by the well-known Kaggle competition website. In this course we avoid platforms that require programming skills, allowing the focus to remain on learning the concepts using already-familiar tools for business managers, Excel and Google Sheets.

In this course, we introduce the basic concepts of data science. Participants work mainly with real datasets and learn about data representation. The focus is exploring and plotting data, discovering noise, and cleaning datasets. We emphasize the most critical parts of ML and AI, which are data preparation and feature engineering. With very simple models and rules, we teach participants how to address and solve some problems inspired by the well-known Kaggle competition website. We follow the most successful strategy in the industry which is: always start with the simplest baseline. Working with a simple baseline you can very easily quantify the added value of ML and AI in your organization. It is the basic ingredient of a good cost function.

In this course, we study the 3 pillars of Machine Learning: Optimization, Linear Algebra, and Probabilities. The Linear Algebra part is based on the very successful books “Coding the Matrix: Linear Algebra through Applications to Computer Science” and “Linear Algebra Done Right”. The optimization part is covered from the books “Convex Optimization”, “Non-convex Optimization for Machine Learning”  and “A Gentle Introduction to Optimization”. For the probabilistic part, we will use “Doing Bayesian Data Analysis”, “Bayesian Methods for Hackers: Probabilistic Programming and Bayesian Inference” and “Statistical Rethinking: A Bayesian Course with Examples in R and Stan”.  During this course, participants are exposed to these math fundamentals through practical coding exercises and lectures.

In this course, we introduce a more rigorous mathematical and algorithmic approach for studying machine learning and AI. Participants learn about:

• Different algorithms and how to use them in practice;
• Strengths and the weaknesses of each of the algorithms;
• How to debug models that use various algorithms;
• Determining when an algorithm works (and when it doesn’t); and
• A methodology for testing, evaluating, and interpreting the results of AI and ML models.

In this course, participants move beyond “out-of-the-box” algorithm implementations and are introduced to modern ML frameworks including TensorFlow, Keras, and PyTorch. Participants will learn about the architecture and the philosophy of the platforms, and how to actually code algorithms covered in the introductory courses. This course relies mainly on ipython notebooks and coding exercises.

This course centers around a dataset and a problem selected by the company; we work with the team (product/project manager and engineers) to solve it. After providing initial guidance on the approach and techniques to be used, we have several follow-ups with the team to provide ongoing assistance, as well as feedback to properly evaluate and test the model. This course is ideal for teams ready to evaluate their ability to apply the concepts that have been taught in earlier sessions.

In this advanced course, participants are exposed to the latest advances in AI/ML, and start developing a highly valued skill: implementing and adapting research from top universities. Our trainers participate in major conferences like NIPS, ICLR,  UAI, and ICML and have expertise on the most recent research papers. We often work directly with the authors of the papers to verify and reproduce the code.

In this course, participants learn the fundamentals of deep learning, examine different architectures, and learn best practices on when to use (and not use) as well as how to best train deep learning networks. We touch on the necessary math and dive into coding examples from the books “Deep Learning with Python”, “Hands-On Machine Learning with Scikit-Learn and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems”. Participants learn about the theory behind the famous GANs (Generative Adversarial Networks), Recurrent Neural Networks, Convolutional Nets and others. We use the Keras framework as our tool to understand how networks work. During the training, participants learn how to deal with deep learning failures and how to debug models by using common sense.

This course builds on “Introduction to Deep Learning”, exposing participants to the mathematical principles of deep learning as described in the bestseller “Deep Learning”. In this course, we follow a more rigorous path, including all the mathematical details necessary to understand how deep learning works. We also review some recently published papers that expose the limitations of deep learning, along with some comparisons between competing architectures. This course will help participants understand the latest trends and advances in research.

This course starts with the basic natural language processing (“NLP”) methodologies, such as bag of words and topic modeling. We introduce Bayesian Methods like LDA and then move on to more recent deep learning algorithms such as Word2Vec and its variances (i.e., Doc2Vec). We also dive into more complicated architectures like LSTMs/ConvNets and Sequence to Sequence Models. The course concludes with work on building some NLP applications, such as question answering and text summarization. As a class project, we build a simple chatbot.

In this course, participants are introduced to the convolutional network architecture. We show participants how to train small datasets like CIFAR-10 and how to use big pre-trained models on datasets like CIFAR-100 and Imagenet. Participants learn how to build models for Imagenet challenges, such as image recognition and object localization (YOLO algorithm). Emphasis is placed on Generative models like GANs (Generative Adversarial Networks). We go through recent academic papers and show participants how to make full use of GANs for interesting applications, such as image captioning and style transfer.

TensorFlow, PyTorch, CNTK, MXnet, Keras, etc. share some common design principles that are very different from the typical imperative programming languages. They introduce the concept of the execution graph, a novel declarative programming paradigm. In this course, we discuss the design principles and, through hands-on exercises, demonstrate the strengths and weaknesses of these platforms. We introduce debugging strategies and coding practices that will help participants avoid common pitfalls. Towards the end of the course, participants will code an algorithm from an academic paper.

Probabilistic programming is one of the most promising areas of machine learning. It is the easiest framework for incorporating domain-specific knowledge into AI/ML models. In this course, we will introduce participants to the concepts of Probabilistic Programming through two state of the art frameworks: Pyro (Developed by Uber AI) and Probabilistic Soft Logic (PSL, developed by the University of California, Santa Cruz). Participants will learn:

• The theory behind these languages (Inference model, algorithms);
• How to use knowledge graphs like freebase in models;
• How to incorporate user domain knowledge in models;
• How to solve relational problems like entity resolution;
• How to build generative models; and
• How to solve problems with small amounts of data.