Transforming a business problem requirement to a machine learning problem statement is an extremely important step in both Industry and academia. The concepts of vector representation of knowledge, mapping between vectors, loss functions and the data set forms fundamental steps in any machine learning problem formulation. For instance consider a problem of predicting likes and dislikes of a customer. How do one represent a customer object as a vector? What are the attributes of the same? Are they homogeneous or nonhomogeneous? How are the recommended objects vectorized? The dive deep leaves one with a number of questions and possibilities. In this summer school some of the sample problems are discussed at this level. The last lecture of the school talks about interesting facts and myths in machine learning in terms of inductive versus deductive reasoning abilities.

The mechanism of working of machine learning requires one to understand the regression and classification settings, loss function, minimization approaches such as gradient descent. The standard metrics for performance assessment of a machine learning model, issues during a production deployment, data issues such as class imbalance, missing values, label corruption and noise are discussed. In addition the critical importance of automatic differentiation in enabling large deep neural networks, the back-propagation and convolutional neural networks are discussed. The workshop also introduces reinforcement learning methodology and unsupervised learning of structure from data. There are code demonstrations and hands-on sessions on machine learning and deep learning topics.

The summer school concludes with demonstrations and sharing of experiences from Industry, faculty research in this area, and discussing scope and limitation of the approaches.

The 1st and the 2nd summer schools happened in 2019 and 2021 years in the month of July. This is a continuation of the same however with mild revisions. We have added a reinforcement learning topic here. The logistics have also been modified to make 90 minute slots instead of 50 minute slots. The previous two schools were successful with good participation and positive feedback.

The primary goal of the summer school is to onboard students/faculty/practitioners onto the topics of ML through both theory and hands-on sessions involving the following three unique features:

(i) Up skill for the need of the hour - Business requirement to problem statement formulation
(ii) Impart essential knowledge of ML, DL, RL and Unsupervised learning methodologies
(iii) Code demonstrations and hands-on sessions