Opening

Hands-on Session

Continued Building Classification Models in the presence of class distribution skewness

Introduction to Neural Networks

Perceptron Model, Neural Architectures, Back Propagation algorithm, Self-Organizing Maps

Deep Learning Algorithms

Convolution Neural Networks (CNN), Recurrent Neural Networks(RNN), Long short-term memory(LSTM)

Application of Deep Learning in the Automotive and Manufacturing Industry

Deep learning as applied to NLP and computer vision; generative adversarial networks (GANs) when labeled data is scarce

Creating a real-life AI Data Product Blueprint

We live in a VUCA world (Volatile, Uncertain, Complex and Ambiguous) world where the velocity of disruptions is dramatically increasing – Uber disrupting the taxi network, Amazon disrupting the neighborhood store, MOOCs disrupting the traditional educational model, how can AI world respond to these changes? This is a path breaking process which originated in Silicon Valley. It’s an actionable playbook for problem solvers. This is a hand on/application oriented practical session where at the end of 90 mins would have an AI product blueprint which you can then take back to your respective organizations for execution. This actionable AI product blue print can be a conversation starter and get your respective organizations mobilized to intercept the opportunities which present itself. This activity is multidimensional in nature. The blueprint which is the end deliverable would touch upon the following 5 areas: 1. The business dimension 2. The data dimension 3. The ML/AI/DL dimension 4. “Under the hood� architecture of AI data product 5. “Over the hood� architecture of AI data product

Why Should You Believe My Model Predictions? Indeed, Why Should You?

How do we reliably *explain* the classification or prediction done by a black-box ensemble of algorithms? How do we demonstrate to the end-user what the underlying relationships between the inputs and outputs are, for traditionally black-box systems? [This applies to explaining predictions from black box tools, & very partially to describing factors explaining a known outcome] How could we influence decision-makers enough to place trust in predictions made by a model?

NVIDIA Hands-On Session: GPU Accelerated Deep Learning - The Path for Tomorrow

Deep learning is the fastest-growing field in Artificial intelligence, helping computers make sense of infinite amounts of data in the form of images, sound, and text. Preventing diseases, building smart cities, revolutionizing analytics - these are just a few things happening today with AI and specifically, deep learning. Today's deep learning solutions rely almost exclusively on GPU-accelerated computing to train and speed up challenging applications such as image, handwriting, and voice identification. This workshop will focus on how GPU-accelerated Deep Learning has evolved in the recent times with focus on application domains and tools.

NVIDIA Hands-On Session

Applications of deep learning with Caffe, Theano, and Torch; image classification with DIGITS

Recommender Systems

• - Content-based filtering, collaborative filtering, matrix-factorization-based approaches
• - Contextual recommendation, conversational recommendation, group recommendation
• - Recommendation diversity, evaluation of recommender systems; people, fashion and news recommendation
• - Adaptive learning

Time Series Analysis

Active, Proactive, Transfer and Multi-Task Learning

Active and proactive learning; transfer and multitask learning, including their applications (case studies in finance, manufacturing, and customer service)

Opening

Applications of Data Science to Manufacturing

Decision making in supply chains is based on optimization models and the data from past sales. Several software packages have been developed and used in the Industry with the aim of delivering quality products to the customers at the right cost. Currently, there are several new trends that are happening in the supply chain arena. Globalization has created dispersed supply chains which are vulnerable and dependent on entities and factors that are exogenous to the supply chain. Also, technologies such as Big data, Cloud computing, Blogs, Social Media, Internet of Things and Mobility have become sources of large volumes and several varieties of data.
In this lecture, I would first present some recent big data start-ups that are revolutionizing or disrupting the traditional manufacturing networks. We then discuss how the new developments in tagging, sensing and embedding effects the four important supply chain processes: procurement, manufacturing, maintenance & repair and retail. Next, we present the big data ecosystem model: bigdata service chain, institutions (governments and social groups) and their influence on data availability, resources (natural, human, financial, and industry inputs) and delivery service infrastructure (communication and decision). This leads us to the question: what data should I collect, and what algorithms should I use to make decisions that would result in better business outcomes.
Data based decision making, particularly with unorganized and non-numerical data is a relatively unexplored area of research with abundant opportunities. The takeaways from this lecture are opportunities for both research and start-ups in this evolving area.

Panel: Internet of Things 1, Things 2, and the AI Hat!

Natural Language Processing

Text categorization, sentiment categorization, named entity extraction, parsing, machine translation and dialogs (chatbots); classical constraint/rule systems, word embeddings, CRFs, and neural methods

Real-Time Cross-Lingual Global Media Monitoring

Detailed explanation of the research and technology challenges implemented in a pipeline of modules in “Event Registry� (http://eventregistry.org/) for:

• • collecting global news stories in real-time (250,000-500,000 articles per day from over 100,000 global sources) [topics like crawling the web data, cleaning the textual data, etc],
• • linguistic and semantic enrichment of the collected information (using Wikifier.Org system, annotating a text into WikiData knowledge graph) [topics like language normalization, lexical, syntactical and light weight semantic annotation of texts including semantic disambiguation; classification, clustering, scalability issues, etc.],
• • connecting linguistic and semantic information across languages (using XLing.ijs.si cross-lingual matching tool for 100 languages) [topics like large scale mapping between languages for information retrieval and classification],
• • organization of enriched news stories into “eventsâ€? and “storylinesâ€? (i.e., related “eventâ€? sequences) [topics like how to make sense out of large amounts of textual data],
• • providing individual and aggregated information through API and graphical user interface for further use [textual data visualization and semantic search].
• Advanced topics include:

• • Modeling news bias (how different news sources are reporting about the same events),
• • Modeling causality in global society (how history is repeating itself),
• • Probabilistic prediction of future events (how much can we tell about the future, knowing the past).

Hands-On Session: Recommending News Articles

The Task is to recommend news articles to a user given a set of articles previously read by the user. This hands on session will expose you to text analysis techniques such as tokenizing, stemming, lemmatization, named entity extraction, token weighting as well as semantic representations such as topic models.

Hands-On Session (Continued): Recommending News Articles

Opening

A short Plenary panel about the importance of AI / ML / Analytics in the BFSI sector and some of your didactic experiences / learnings .

Data Stream Mining I

• • Introduction to data streams
• • Learning from data streams: classification and regression models, concept drift, clustering, frequent pattern mining
• • Evolving networks: social network analysis, community detection, modeling evolution of communities

Data Stream Mining II

• • Introduction to data streams
• • Learning from data streams: classification and regression models, concept drift, clustering, frequent pattern mining
• • Evolving networks: social network analysis, community detection, modeling evolution of communities

Panel discussion : 5th Age of Banking. How Analytics is transforming the BFSI industry

Data Science for Digital Commerce

The talk focuses on the use of Data Science in digital commerce with applications in Recommendation based on cross selling and upselling, Dynamic Pricing and SEO optimization.

Reinforcement Learning

Reinforcement learning is a paradigm that aims to model the trial-and-error learning process that is needed in many problem situations where explicit instructive signals are not available. It has roots in operations research, behavioral psychology, and AI. Recently Deep Reinforcement Learning methods have achieved significant successes by marrying the representation learning power of deep networks and the control lear ning abilities of RL. This has resulted in some of the most significant recent breakthroughs in AI such as the Atari game player and the Alpha Go engine from Deepmind. This success has opened up new lines of research and revived old ones in the RL community. In this talk, I will give introduce the reinforcement learning paradigm and briefly review the progress made in deep RL. We will discuss the problem of recommendations and that of Atari game playing in detail.

Building a World-Class AI/Data Science Team and Closing Remarks

What skills sets are required in machine learning, analytics, and data bases? We will also discuss the importance of domain knowledge as well as leadership and support roles.

Hands-On Session

· Exercise One: A simple bandit algorithm, corresponding to a recommendation application. The set up will be completely synthetic.

· Exercise Two: A simple spatial navigation problem. The domain will again be a simulated domain. This corresponds to an arcade learning environment, but the setting is simplified so that learning can be completed in the allotted time.

Talk Using India Stack Adhar in BFSI

Hands On Session