X Speaker Profile

I am an Associate Director with more than 9 years’ experience in data and analytics. I am part of Deloitte Analytics, a cross-functional team with a focus on embedding data and analytics and AI in organisations across Africa. I am also responsible for the Digital Transformation team in Risk Advisory with a focus on exponential technologies, AI, Digitisation and rapid prototyping. I am passionate to help people on their journey to enable their company to become a data driven organisation, and have a passion for moving Africa into the fourth industrial revolution and enabling the true potential.

X Topic Abstract

Artificial Intelligence and Machine Learning provide faster, more accurate assessment with enormous data processing powers, and consider a wide variety of factors which leads to better-informed and data-backed decisions. AI based outcomes are based on complex algorithms and sophisticated rules to help clients streamline and optimize processes ranging from digital adoption to personalized banking, fraud management, intelligent automation and NLP based customer satisfaction analysis.

Speaker Profile

Manisha Banthia has over 23 years of experience in Analytics in Financial Services and currently heads the Analytics CoE at Fiserv Global Services.

She has earlier worked in Infosys consulting team heading their analytics team and was instrumental in developing building analytics products, including a patent for ‘Customer Analytics for Enterprises’.During her stints at Oracle Financial Services Software (earlier iflex solutions) she built analytics solutions for Citibank (Asia Pacific), and US clients like American Stock Exchange and Country Financial.

Manisha has also worked on Marketing Analytics, Financial Performance and Fraud and Risk Analytics during her tenure at MetricStream and National Stock Exchange of India.

Manisha is based out of and heads analytics teams in India, Costa Rica and US for Fiserv.

X Topic Abstract

Natural Language Processing has been the new chip in the block of AI. There has been a significant progress in the domain of language understanding using Deep learning technology. This has given rise to plethora of applications which uses language , written or spoken form, as one of the key medium of communication. Language understanding, especially understanding the context or the big picture, was one of the key challenges.

Thanks to the progress of Deep Learning computing, more specifically evolution of RNN network, this is now falling in line and has been a tremendous success. Automated translation, now almost in real time, Google Duplex AI voice assistant placing a call for seat reservation in Restaurant is a reality. Amazon Alexa has almost become a household butler.

This summer, Tokyo Olympic 2020 will see lot of innovative voice enabled solution to bridge the language barrier. NLP has a tremendous potential as key AI technology and this technology will impact almost all types of business and value added services. It will unblock a key bottleneck in the area of process automation.

In this talk we will explore the evolution of NLP technology, core building blocks and computing backend with some example use cases. We will also talk about few uses cases we have implemented in Tata Elxsi - for example , Virtual sales assistant and Virtual Car Assistant.

Speaker Profile

Biswajit Biswas is Chief Data Scientist @ Tata Elxsi heading the AI Center of Excellence responsible for driving innovation in Digital systems for Communication, Media and Transportation business verticals . He has over 22+ years of Industry experience.

He is very well versed in AI and Deep learning , especially, in the area of computer vision, language and text processing . Before heading Data Science team, he was responsible for Wireless Communication, Multimedia Systems, Digital signal processing , Image, Video and Audio coding, Machine Vision System. He has hands on with developing algorithms in C/C++, Python, TensorFlow, R and devotes significant time in coding, coaching team in various development activities. He also holds a patent for innovation in V2X related technology and has filed Patents for few other innovations in the area of Predictive Machine learning.

He is a Certified Data Scientist from Massachusetts Institute of Technology (MIT, US) Media Lab , Masters from BITS and BE from Jadavpur University Calcutta.

X Topic Abstract

In e-commerce , customer experience can be divided into broadly 3 parts - pre purchase, purchase and post purchase. Drivers of good customer experience vary across these 3 parts of the customer journey. At Myntra , we heavily use data based insights and advanced ML models to identify what product features, merchandise selection and service propositions optimise customer experience while helping the business meet it's financial goals. Some examples include search personalisation, catalogue optimisation using computer vision, automated return approval for "trusted" customers etc. The presentation would cover some examples of projects that have been undertaken at Myntra and also talk about what metrics should be tracked to ensure your company is meeting the customer experience goals.

X Topic Abstract

Most of them think that AI/ML is a complete black box and only accuracies matter, which is not the case in industry level scenarios or any Machine Learning problem statements.

Interpretation of ML models is highly important and how do we interpret that and tweak it to get industry level stats is the thing which will be covered in this session

Speaker Profile

X Topic Abstract

Since we have only 30 minutes for the presentation, I am planning to take some of the below use cases from Banking.

(1) Real Time Offer Recommendation System. Recommend real time merchant offers for customers based on his current geo location on Bank app. Give customer offers based on cluster of Merchants in the area he is visiting. These recommendations will be based on their propensity and interest segments.

(2) Leads Optimization using Machine Learning. Recommendation of a right product for a customer through right channel based on propensity, profitability, risk and utility of a product.

(3) Credit Card Attrition model. Machine Learning models to identify the customers who are likely to move out from bank beforehand so the campaign team can devise proper marketing strategies.

(4) Anti Money Laundering. Machine Learning models to predict the customers who are engaging in Anti Money Laundering.

(5) ATM cash Optimization. Machine Learning Models to predict the ATM replenishment, frequency and the cash.

(6) Face recognition for Branch Customers. Identification of preferred & AML watchlist customers who walk into branch. Use of deep learning models to identify customers (Preferred & AML watchlist) walking in into the bank.

X Topic Abstract

Data science and Machine learning (DS & ML) platforms are now quite popular and there are more than a dozen good platforms available in the market from reputed vendors. What are DS & ML platforms and what are the advantages of having one in your enterprise. How do you decide whether you need one? Do you need it now or should you wait? What are the capabilities that you should be looking for in your Data science platform?. Learn in this session

Speaker Profile

Vinod Khader has around 20 years experience in Software Development and is an Associate Director at IBM Software Labs in the Data and AI division. In the current role, he leads the development of Watson Machine Learning Platform on IBM's public and private cloud platforms with teams across the globe. Watson Machine Learning platform help Data Scientists, Data Engineers and App Developers in managing the end to end Machine Learning and Model management life cycle starting from Training, Evaluation, Deployment and Scoring (prediction).

X Topic Abstract

Service Desks requires good strategy and innovation to improve customer service and to support business goals. Cisco TAC is constantly looking at ways to provide customers with the information they need via a self-service capability in an attempt to reduce the number of support requests that are opened. This is frequently referred to as case avoidance. A significant amount of information is lost while exchanging mail, chats or over case notes.

The impediment to progress is the inability to derive actionable insights from this unstructured text since most of the prioritize corrective action is buried in these notes. Using advanced ML techniques along with text mining solution, the objective is to extract actionable insights for TAC that can be used to drive down Service Request volume. The proposed solution comprises of text parser and automated case deflection Conversational NLP solution which can provide an immediate solution using historic data analysis"

Speaker Profile

Vivek is currently a data scientist @Cisco. He has 7+ years of work experience across E-commerce, mortgage, Retails and CPG domains. He is an extensive NLP researcher, he has a pragmatic approach in text mining, NLP, Machine learning and social media analytics. Prior to Cisco, he has also worked across organizations like Xerox, BRIDGEi2i and Altisource.

X Topic Abstract

In digital advertising, Behavioural and Contextual are two prominent ad targeting strategies used for serving relevant ads in front of target audiences. Behavioural targeting focuses on user’s online behaviour and typically utilizes cookies for tracking their browsing behaviour and subsequently placing relevant ads. Contextual targeting, on the other hand, solely focuses on the content of the page that a user is browsing which represents the context and accordingly places ads.

Due to rising concerns around privacy, which has resulted in regulations like the GDPR (General Data Protection Regulation) in Europe, it is getting increasingly difficult (and expected to get even more so in the future) to target users based on tracking their browsing behaviour. In view of these challenges, we have developed a novel contextual targeting approach which uses contextual features like URL, keywords, postcode, browser, operating system, etc. for optimizing digital marketing campaigns. The developed framework optimizes various business KPIs such as CVR (Conversion Rate) and CPA (Cost Per Acquisition) using a control theoretic proportional feedback mechanism that dynamically learns & optimizes campaign performance.

Comparison of our feedback loop-based optimization approach with more traditional manual/heuristic based approaches has resulted in savings of close to 40% in the overall campaign budget in over 60% of cases that we have analysed so far. These results highlight the benefits of our approach in optimizing ad spend in a more restricted “Privacy-First” world.

Speaker Profile

Manish Pathak is currently working as a Data Scientist at MIQ Digital where his work primarily includes building Machine learning and Advanced Analytics products for the digital advertising domain.

He has experience working in the areas of Supervised and Unsupervised Learning techniques and Deep Neural Network models. Graduated with a dual degree in Electrical and Electronics Engineering and Masters in Physics from BITS Pilani, Manish is an avid learner and a Data Science blogger too.

He believes in democratizing AI/Machine learning knowledge and help organizations build impactful and revenue generating data products.

X Speaker Profile

I am an Associate Director with more than 9 years’ experience in data and analytics. I am part of Deloitte Analytics, a cross-functional team with a focus on embedding data and analytics and AI in organisations across Africa. I am also responsible for the Digital Transformation team in Risk Advisory with a focus on exponential technologies, AI, Digitisation and rapid prototyping. I am passionate to help people on their journey to enable their company to become a data driven organisation, and have a passion for moving Africa into the fourth industrial revolution and enabling the true potential.

X Topic Abstract

In business areas where there are decisions involved that have social implications and we have seen biases in human decisions or data skewed undesirably towards a particular category, it is important that our ML/AI algorithms are tuned to eliminate any such bias that are bound to creep in automatically due to historical data. How to we identify those scenarios and eliminate biases effectively from our Machine Learning algorithms.

Speaker Profile

Manas heads the Wealth Technology Business Unit for ANZ Bank in Bengaluru, a man with diverse interests apart from his role in Technology Delivery – a passionate champion of Innovation, he has helped institute an Innovation Framework in ANZ Wealth Globally including an Innovation Contest that is saving hundreds of thousands of dollars. Manas is also the Chair of the ICT-Academia Expert Committee of Chamber of Industry and Commerce (BCIC), where he is presently leading the organisation in a number of Industry-Academia initiatives.

X Topic Abstract

AutoML, as the name suggests, is the process of automating the process of applying machine learning to real-world scenario.

When applying machine learning models, people generally do data pre-processing, feature engineering, feature extraction and, feature selection. After all this, one can select the best algorithm and fine tune the parameters in order to obtain the best/optimized results. AutoML is a series of concepts and techniques that used to automate these processes.

There are many service providers for AutoML like Amazon SageMaker, Azure Machine Learning AutoML, Google Cloud AutoML etc. There exists many AutoML framework like TPOT, Auto Sklearn etc. AutoML helped a lot in terms of reducing human efforts.

Speaker Profile

A dynamic result oriented professional with blend of Data Science, Analytics Consulting, Business Analytics and project & people management experience comprising of 15 + years from project scoping to entire execution, in several successful shared services organizations in India in technology, CPG & retail sector.

o Has global experience in-terms of developing and implementing marketing, communication, Marketing Mix Modeling, 360-degree unified view of customers, different types of predictive modeling (Churn, next most logical product etc.), and business solutions across different domains.

Also has rich experience in formulating and implementing light listening, deep listening, loyalty analysis, competitive landscape analysis & sentiment prediction models using Social media data for different brands/domains. I also have worked in Marketing Media spend as well as channel optimization.

She Have proven record of accomplishments of setting up new processes & procedures & nurturing the fresh talent into specialist. She

X Topic Abstract

Description: NLP (Natural Language Processing) is the new buzzword that has set the AI / ML world into fire. In actuals, NLP is evolving into a new branch of AI altogether. So here, let's discuss and have hand on the core of NLP - which is nothing but - TEXT CLASSIFICATION.

Let's set the stage:

Working on a ML problem is like cooking .

Yeah while cooking, one has to decide the dish and recipe, source the ingredients, cook them and present them beautifully. On the same lines, the agenda will be:

· Prep work:
o A refresher on ML concepts and NLP
o Understand about TEXT Classification and its importance which can brings in billions of revenue
· Decide the dish & Recipe
· Here we will be setting up the problem
Setting the environment
Explanation of the same
· Source the ingredients:
Learn the data set
Load the dataset
· Let's start cooking:
· Coding and solving the issue / problem in hand
· Includes modeling, evaluating the model etc
· Serve the DISH
· Do the prediction and analyze the results

Take away:
· The participants will get to know on:
· Why TEXT classification is so important (Tech giants like Google are making billions out of it)
· Learn about important concepts like & various techniques like:

Bag of words
Vectorization
TF-IDF (used by Google for displaying search results)
Pipelines
How to choose the best classifier
What is performance tuning

See in action algorithms like:
§ SVM
§ Naïve Bayes
§ Performance tuning with Grid search

Compare the performance with the super-efficient NLTK library

Speaker Profile

Bharathi is a Principal engineer with AT&T Communication services India ltd. He works in the 4G and 5G space who is a pioneer in RPA and also in AI field. He is also a PYTHOnator who like to transform the industry. Under his mentorship, hundreds of people (from students community and also from the industry) got into the bandwagon on Python and automation. He holds more than 78 awards within his company and also holder of THREE nanodegrees from Udacity (all in AI field). He loves to develop Alexa based skills as his hobby and ready to embrace new challenges every day.

X Topic Abstract

TV advertisement has always been the most preferable medium for marketers to reach a mass audience. Since the inception of smart/connected TVs, now we have access to second by second viewing data of households about their TV watching behaviour with the consent. At MIQ Digital India Pvt. Ltd. we collect this data in realtime with a scale of handling millions of requests per second and wrangling collected data as per the marketers need by retargeting TV viewers based on their demographics on digital medium across different channels. Automation Process looks at the intent of marketers targeting relevant audiences encapsulated by configuration . In order to achieve seamless processing, this particular data science solution is supported by Golang and Pyspark to achieve concurrency and accuracy for TV viewer segmentation.

Speaker Profile

Parth is currently a Senior Data Scientist at MIQ Digital India Pvt Ltd. He designs algorithms for finding similar users based on their Connected TV viewing pattern, predicting TV show tune-in, etc. With past experience of working on a varied set of data science problems in the field of Customer support, Revenue prediction, Customer churn prediction for subscription based business. He has completed his Masters in Technology with a Data science major from IIIT .

X Topic Abstract

In this talk, we will cover the basic of Deep Learning and how to implement it hands on for text, images and videos. We will take 3 problems in Machine Learning for text, images and video and build classifier using Deep Learning. We will talk about different architecture like LSTM, CNN for treating images and video. The key take away from this talk is to get the basic architecture of Deep Learning and how to implement them in pytorch to do basic text or image classification.

Speaker Profile

I am a deep learning researcher, after graduating from Indian Statistical Institute with Master's in Computer Science, with specialization in NLP and IR, I have worked with Bing Ads, Microsoft R&D and HSBC Bank as decision scientist to build risk models for bank. I am currently working in Amazon as a data scientist to build models to cater to amazon internal network.

X Topic Abstract

Introducing latest development in NLP systems which enhances the usability and reliability of Conversational AI powered applications, especially in Customer Support systems like Smart IVRs and AI Chatbots. Also, how to go about adding support for Vernacular languages to increase customer engagement.

Speaker Profile

Nishant is an experienced Big data stack developer working at Microsoft Azure Storage Platform team. He has been instrumental in research for developing Vernacular Language Understanding Infrastructure for Indian languages using latest advancements in NLU domain.

X Topic Abstract

There has always been a trade-off between “Accuracy” and “Interpretability” in the field of applied ML. The dichotomy presents significant challenges, when it comes to application of ML for creating sustainable and scalable impact for businesses. While the linear models are fairly explainable as those often depict the average behaviour based on data, the accuracies of those models are certainly sub-optimal. Non-linear models are significantly superior on accuracy, however are extremely difficult to explain. The lack of transparencies (read black box!) in complex ML models hit the trust factor amongst stake-holders hard.

In this talk Satyamoy will simplify the physical significance of the topic. He will bring in clarity around what it means for businesses and how one can leverage the progressive research in this field to drive explainability without sacrificing on accuracy. He will also walk the participants through Analyttica’s innovative IP solution addressing this field through one of the real business cases.

X Topic Abstract

Insurance has been a traditional, century old business affair and has largely been unperturbed through the immense technological disruptions & advancements that have been made in the last decade or so. It has to take a leap of faith to evolve out of the traditional models of insuring property, casualty and commercial line items through leveraging AI & emerging technologies such as distributed ledger blockchain in order to revamp its presence to customers My talk will focus on my work in niche microinsurance based product offerings which are tailor made for a focussed audience group, such as: - Easy-to-purchase flight delay insurance using blockchain architecture (ref. fizzy.axa) focussing on millenials and frequent flyer groups - Usage based motor insurance as per the driving behaviour & risk score of the user - Parametric event interruption insurance (such as rains during cricket match etc.) I shall specifically stress on how are we building active machine learning workflows in designing the experience these products cater to their users while also touching upon the architecture which makes it feasible for implementation and monitoring production in the long run.

Speaker Profile

Amitanshu heads the New Products and Business Initiatives team at Bharti Axa General Insurance and has a diverse experience in speaking at various data science keynotes for corporates and student groups alike.

After completing his bachelors in Mathematics and Scientific Computing from Indian Institute of Technology, Kanpur (IITK), Amitanshu has extensive experience in both academia and corporates – where he worked across verticals such as Telecommunications, Retail, E-Commerce and Insurance.

In his current role, Amitanshu leads the development of innovative micro-insurance product portfolios encompassing delays (flight delay insurance, cab trip delay insurance etc.), consumer durables & electronics such as gadget insurance, bicycle insurance etc. offering seamless customer claims experience with minimal customer involvement or paperwork.

Prior to this, Amitanshu has lead the end-to-end development & deployment of various Data Products leveraging sophisticated Machine Learning Algorithms on various fronts viz. Chatbots, Voicebots, Recommender Systems etc.

On a typical work day, you can find Amitanshu busy in strategizing optimal insurance cross-sell/up-sell strategies along with hands-on problem solving with Data Science & Product teams while implementing robust data architectures for optimum digital-first insurance product offerings. Apart from work, Amitanshu is a professional guitarist, with a passion for classic rock bands such as Pink Floyd, Led Zeppelin, Radiohead, etc. and an avid adventure sports enthusiast.

X Topic Abstract

What makes explainability difficult in AI/ML

How to go about adding explainability and interpreting it

Adding MLI (Machine learning interpretability) to Deep Learning models across text and vision projects

Pros and Cons of adding MLI in AI/ML models

X Topic Abstract

NLP based virtual agents are servicing robo advisory, be it explanation of financial products or of health plan benefits. The use cases will discuss on importance of interplay of ML and conventional engineering to solve end user problems, and explore an NLG approach to tackle the data non-availability issue for training purposes.