info@analyticsindiamag.com +91-95389 89028, +91-96325 33477

Schedules

Event Schedule

We are in the process of finalizing the sessions. Expect more than 30 talks at the summit. Please check back this page again.

19-20th Jan 2022. Virtual

Expand All +
  • Day 1

    Jan 19, 2022

  • Graph neural networks (GNNs) are a class of neural networks that work on graphs. By representing individual entities as nodes and their relationship as edges - GNNs capture patterns within the graph. These techniques have been useful in a variety of industrial applications such as chemistry, recommender systems, fraud detection among others. In this talk, we will do a deep dive on what GNNs are and how they are applied in various parts of the industry.
    Tech Talks

  • Poaching is one activity which has posed a great threat to Wild Animals. According to World Wild Life organization there has been a dramatic increase in the killing of Rhinos for their horn, Tigers for their skin and Elephants for their ivory tusks. Due to extreme poaching activity all these above-mentioned animals have been categorized as endangered species as there are very few left in the jungle. Interventions to poaching have been successful but need accurate intelligence and are human-intensive and dangerous activities. Recent advances made in Computer Vision, Unmanned Ariel vehicles, and Artificial intelligence allows for assisting the said conservation efforts. With that goal in mind, we have come up with an AI solution that would solve or at least help in solving this problem and aid in wildlife Conservation. In this paper we explore a solution based on Thermal camera attached to a drone to detect humans and alert authorities’ personnel in the restricted zone in the forest and national parks. Drones can be put in randomized flight patterns making it hard to circumvent, difficult to detect the presence of or destroy it.
    Paper Presentations

  • In Retail World, promotional offers are important as they increase consumer demand and sales. As promotional strategies directly engage with the customers, it is important to pick the right product and the discount that resonates with the customer needs. One way to do so is by evaluating past promotions and understand their impact on sales and baskets. In this paper, we discussed an approach that evaluates sales uplift due to promotion and helps in understanding its wider impact on the business. Sales uplift is a key metric that calculates the difference between promotional sales (i.e., sales during the promotion period) and baseline sales (sales if there were no promotion in that period). The challenge is to calculate the baseline sales value, as it is not possible to know the actual sales value if the promotion had not taken place. Hence, we proposed a Linear-Regression (LR) based approach to forecast the baseline sales value of a product that is on promotion by considering various factors like seasonality, trend, etc. Our LR based baseline approach has an edge over other Time-series approaches (e.g., Holt- Winters Exponential Smoothening) especially for seasonal products, as LR predicted sales values are not only driven by the recent past sales value. The baseline is used to calculate uplift to evaluate promotion performance and allows retailers to align their promotional strategies. This methodology was tested on transaction data sets for two different retailers. Based on the results the retailers were able to identify and avoid sales dilutive promotions which were neither benefitting the retailer nor the customers.
    Paper Presentations

  • Abstract— In this paper, we present a passive method to detect face presentation attack a.k.a face liveness detection using an ensemble deep learning technique. Face liveness detection is one of the key steps involved in user identity verification of customers during the online onboarding/transaction processes. During identity verification, an unauthenticated user tries to bypass the verification system by several means, for example, they can capture a user photo from social media and do an imposter attack using printouts of users face or using a digital photo from a mobile device and even create a more sophisticated attack like video replay attack. We have tried to understand the different methods of attack and created an in-house large-scale dataset covering all the kinds of attacks to train a robust deep learning model. We propose an ensemble method where multiple features of the face and background regions are learned to predict whether the user is bonafide or an attacker.
    Paper Presentations

  • Driver Monitoring Systems (DMS) track movement of people using a camera inside the vehicle using AI to predict driver alertness to decide the safety of the driver and people on the road. Cameras collect huge amounts of data in all light conditions and activities of people inside the car. This data carries a wealth of insights about driver movement. Hence we propose a new label centric-approach by labeling the camera data with inclusive AI constructs for a more expansive annotation of the same dataset instead of the typical model-centric approach or data-centric approach to improve the performance of this AI. We used the DMD multi-model dataset for driver monitoring scenarios which comes with labeling movement to track 8 actions of the human texting with left or right hand, talking, drinking, phone call with left or right hand, reaching the side of the car or combing hair. We developed a binary classification CNN model for movement in the car. We tested the model trained against an inclusive AI constructed labeling option on the same dataset where we expanded the labeling to track movement of hair flying, scarf fluttering, hand waving or rubbing eyes. The results showed that the inclusive AI construct improved model performance without any change to the model algorithm tuning. Hence we recommend using a label-centric approach to improve labeling of data from camera streams such as the autonomous vehicle to be inclusive on an expanded construct for labels covering all people of all cultures in all lights, all hair, dress, and actions so AI model performance can be improved by capturing more knowledge by being more inclusive of all humans and their actions inside the vehicle.
    Paper Presentations

  • While we all are well versed with projects and POCs, delivering Analytics at scale for large industrial clients is a different ball game all together. It needs a bilingual talent who can combine the domain knowledge along with the analytics/data science competency. This talk is to do a sneak peek into a large analytics delivery organization impacting the client in their day to day life through a combination of analytics services like Data Preparation, Business Intelligence, Forecasting, Optimization and Advisory services by applying conventional analytics tools/techniques as well as the new age AI/ML technologies.
    Tech Talks

  • Heterogenous treatment and its effects have a verity of applications in diverse fields such as medicine, psychology, and marketing. In marketing it is generally used to identify the target customers who will respond to a campaign if they get the exposure. In this paper, we are solving a different problem which is to predict the incremental sales generated by the customers who get the treatment. This problem is tougher to solve since conversion value won’t be available before the intervention so usual supervised machine learning algorithms where we attempt to predict an outcome directly from given observations cannot be applied here. In order to solve this problem, we are using R-meta-learner. This technique estimates marginal effects and treatment propensities to form an objective function which isolates the causal component of signal. We further develop a framework which generalizes this meta learner for non-linear functions. We have implemented this approach using gradient boosting and neural networks and the results show significant improvement from baseline.
    Paper Presentations

  • Quantum Machine Learning (QML) is a multidisciplinary field that attempts to leverage the fundamental properties of the quantum world to empower conventional machine learning algorithms to handle scalability challenges. Strange behavior of quantum particles such as superposition, entanglement and teleportation enable QML to find needle in the haystack in far more efficient way than a classical machine learning on conventional computers can. This was amply demonstrated in 1994 by Peter Shor and in 1995 by Lov Grover through their respective quantum algorithms. This field, even with all the technical challenges of quantum computing, has potential to make non-tractable problems into tractable ones.
    Tech Talks

  • The past few years were focused on the discovery and application of Al to improve productivity. The next few years will drive transformation - where focus will shift towards making AI accessible across organizations. However, organizations are at various spectrums on their journey to accessible AI. Some are experiencing the benefits of accessible AI, while others lack the platforms, skills, or processes. Join Navin Dhananjaya, Chief Solutions Officer, at Ugam, a Merkle company, as he shares practical learnings on enabling transformation by making AI accessible. He will deep-dive into the 3 building blocks that have helped Ugam successfully make AI accessible for leading organizations.
    Tech Talks

  • Engaging with multiple organizations trying to recognize and use predictive power of Analytics, compel us to bring all those journeys in a crisp way here. While we understand many organizations getting their hands dirty in analytics, but around 20% of them only succeed to relish real power of it. Some organizations start analytics with Intellectual curiosity, while some do it with few pilot projects, and few by targeting huge challenges in mind at start of journey. Our observations looking at diversified routes taken by organizations say, all they lack in common is structured approach and clear strategy. Organizations usually see their effort falling short in leveraging data available to them and becomes frustrated. With this study, we propose generic analytics framework that can assist Leaders trying to establish new analytics team or reorganizing existing team. The first half of the paper revolves around framework that is result of extensive work experiences of the authors in starting a new analytics practice and observing counterparts in doing so. Typically, we understand C-suite recognizes great vision and key dimension, but nonetheless sometimes tend to ignore the structured framework of these dimensions and their execution. Our framework is here to help you ease out in identifying those dimensions and giving them a structure that could be apt for a user to apply in current situation. The later half of paper covers framework that is apt for an existing analytics team facing various execution challenges or struggling to grow up. Our work will navigate and assess the present scenario of team in terms of data size and Insights, and binding it with overall vision, it will recommend the route. The model and outcome from this model might look simplistic, but the impact it will have of the strategic vision and structured approach will be monumental. The frameworks proposed can help company become Artificial Intelligence driven and hence improving performance
    Paper Presentations

  • The Tredence team would like to introduce Tredence Studio, a co-innovation platform that provides ready-to-deploy AI solutions and accelerators that enable businesses to accelerate their digital transformation. With Tredence Studio, enterprises can re-imagine analytics interventions across customer touchpoints and business processes to drive sustainable business outcomes.
    Tech Talks

  • The usage of drones or Unmanned Aerial Vehicles (UAVs), both for military and civilian purposes, has increased in India in the past decade. They are being used for reconnaissance, imaging, damage assessment, payload delivery (lethal as well as utilitarian), and recently among the COVID-19 pandemic, for contact-less delivery of medicines. As UAVs are getting more affordable and easier to fly, and more adaptable for crime, terrorism, or military purposes, defence forces are getting increasingly challenged by the need to quickly detect and identify such aircraft. A number of counter-drone solutions are being developed, but the cost of drone detection ground systems can also be very high, depending on the number of sensors deployed and powerful fusion algorithms. With recent research showing attention-based neural networks possessing better object detection and tracking capabilities than traditional convolutional neural networks, a novel attention-based encoder- decoder transformer model is developed to detect drones in a given field of vision and track their future trajectory in real- time. This allows surveillance systems in red zones (such as military bases, airports, etc.) to detect and neutralize possible dangers associated with drones.
    Paper Presentations

  • The purpose of this paper is to demonstrate the methodology used to identify accounts that have a propensity for cross-selling in a B2B SAAS industry. Data spanning across various categories - Demographics, Client Relationship, Product usage & Transaction pertaining to customer accounts was considered, to ensure 360-degree visibility of each account. A probabilistic model was built to predict the account propensity (trained on accounts that had shown historical transaction of cross-sells for a given product). Performance of the model was gauged on parameters - F1 score, precision & recall. Ultimately, the model with higher recall with reasonable precision was shortlisted (with the sole intention to increase the coverage on potential customers that are highly likely to buy other products). This exercise was done with 5 products and separate models for each product were built, to create an ensemble of cross-selling models. The accounts were further bucketed in high/medium/low based on their probability score. These results were presented in a dashboard along with additional key information around the accounts (such as industry operating in, region, tenure, etc.) for the stakeholders to make better sense of the cross-sell accounts. Several marketing and sales campaigns have been initiated for the high propensity accounts.
    Paper Presentations

  • This Modern day digital marketing is an enormous system of channels comprising digital channels such as search engines, websites, social media, email, and mobile apps. Marketers can simply onboard their brands, advertising online is much more complex than the channels alone. Consumers heavily rely on digital means to research products. For example, Think with Google marketing insights found that 48% of consumers start their inquiries on search engines, while 33% look to brand websites and 26% search within mobile applications. With the burgeoning options in digital marketing, and relatively lower cost compared to the traditional media, every advertiser has the opportunity to reach their target audience. Every advertiser is competing for the attention of their target audience. It is estimated that an average user is expected to encounter 6000 to 7000 advertisements every day . Since it is not humanly possible to hold the user’s attention to all the ads, it thus becomes imperative to reach the user when they are most likely to engage with the ad. The attention window of every user is different and it varies with the channel. The behaviour of individual users is also dynamic. Thus the solution that suggests the best time to send has to be dynamic and adaptive. Optimizing the time when the ad is delivered to every user on a specific channel thus helps marketers in achieving the best possible engagement and thus aids in attaining optimum conversion goals. This paper presents a method to estimate the best time to send a campaign in order to maximize user engagement with the ads.
    Paper Presentations

  • Join us to learn how Deep learning is making its way into solving real-world problems in industrial settings and how companies like Drishti are using AI and computer vision to solve problems in manufacturing, one of the world's most complicated industries. We will be covering the following topics: · How Drishti uses computer vision to perform automated video analysis of factory floor videos. · Nuances of action recognition and how challenging the problem is given the accuracy requirements for a production-ready solution. · Drishti's cutting-edge models which are trained using minimal labeled data (semi-supervised learning). · Drishti's homegrown ML Ops infrastructure for continuous training and evaluation of models. · How manufacturing industries can improve their approach to efficiency using the data generated by these models.
    Tech Talks

  • A customer lifecycle journey consists of many events arising out of customer activity and behaviour. Such events (like upgrading subscription, cancelling the product or customer needing support) helps an organisation in providing the best possible experience to customers. However, predicting such events ahead-of-time using AI requires building multiple bespoke propensity models for each such event. In this talk we will show how we built a service for quickly building propensity models in a self-serve way for any desired event.
    Tech Talks

  • According to Gartner, all personnel hired for AI development and training workshops will be required to demonstrate expertise in responsible AI by 2023. As AI adoption grows in the industry, the emphasis has shifted heavily to developing and deploying responsible AI applications. In the MLDS workshop that Tredence will conduct, we will introduce the different elements of Responsible AI: A comprehensive introduction to the theory of Explainable AI and Monitoring, specifically Drift's concept Demonstrate how Explainable AI and Monitoring work with examples Introduction to the other elements of Responsible AI Who should attend: ML Practitioners who want to learn about Responsible AI and the theory behind Explainability and Model Monitoring What will the attendees get at the end of the workshop: The ability to explain different elements of Responsible AI Be comfortable in the theory of how Explainability and Monitoring works Implement Explainability around models built on Structured, Text, and Image data
    Workshops

  • Advanced analytics at the edge has become a reality due to a Cambrian explosion in the area of hardware-based AI accelerators. Depending on the type of workload, cost and power considerations, there are a variety of hardware options that can be considered, and the race for the highest TOPS/Watt seems to have just begun. In this talk, we will look at Perception AI as the compute workload, characterized by data from sensors like microphones, 2D camera, 3D lidar and more, in relation to the trends of hardware accelerators, and finally explore the ends of the spectrum, i.e., low power and high performance, using two relevant real world use cases. Who should attend: Deep Learning and Embedded AI practitioners, who want to a peek into the world of hardware accelerators and how Perception AI applications are leveraging them.
    Tech Talks

  • Why data products are increasingly becoming important Types of data products Framework for building data products Unique challenges in building data products Importance of product mindset for data scientists
    Tech Talks

  • Pre-trained models and transfer learning have been all the rage in recent years with transformers breaking all performance benchmarks for state-of-the-art (SOTA) performance across almost every possible NLP task. In this session we will journey through the motivation behind transformers, unbox the architecture of how it works behind the scenes, discuss latest and best models like BERT, T5 and GPT3 and look at ways in which it can be leveraged to tackle a variety of diverse NLP tasks using the same model (and also a brief about how it has become popular in computer vision!).
    Tech Talks

  • Day 2

    Jan 20, 2022

  • In this 2 hour workshop, you will have the opportunity to experiment with data using a combination of techniques to look for the answer. You will extract key pieces of information that define the essence of each race and visualize races and seasons using advanced techniques to determine how interesting each race is. You will learn the basics of ANALYTICS and MACHINE LEARNING and become familiar with the entire process, from data preparation and visualization, to model building, prediction and evaluation. You'll experience first-hand what challenges data scientists! So come join us in this fun workshop and solve the challenge! No previous experience is required. We will provide instructions to get your Oracle Cloud free tier account.
    Workshops

  • AI adoption is happening at a rapid pace across industries. While there has been immense benefits from AI technologies for people, businesses and governments, there has been some concerns, fears and risks flagged. Governments are mulling regulating AI systems so that there are adequate safeguards and controls to ensure AI is used in an ethical and trustworthy way, does not endanger their safety and makes a difference in their lifestyle. AI regulations are going to be a reality soon. To be future ready, it is imperative on AI practitioners to keep a watch on regulatory trends, define processes to adhere, create systems to track and continue to innovate without any hindrance.
    Tech Talks

  • Personal finance is an area that is data-rich and with significant untapped potential from an AI lens. In this talk, I will detail several avenues where I have contributed toward the AI augmentation of Intuit's personal finance ecosystem.
    Tech Talks

  • Artificial intelligence (AI) has seen rapid adoption in recent years and the conversations about what is and isn’t part of AI are a constant point of discussion in the media. One such perception is around Operations Research (OR) is that, it is not very significantly useful for a Data Science (DS) tasks. Additionally, the overlap between DS and OR is often misunderstood. One reason for this misunderstanding lies in the marketing of OR products and services when applied to real world problems. Since end users and customers often do not have a significant understanding of what encompasses the terms OR and DS, many products are commonly marketed as DS. Another reason is the fact that all the readily built ML models are available as packages of various platforms like Python/R. Such packages don’t contain any specific OR models and hence it seems as though DS and OR are not that interlinked. Operation Research (OR) problems are applicable to AI/DS and a lot of ideas in solving problems in AI/DS have cross pollinated from OR due to the large overlap in the techniques, approaches and methods used.
    Tech Talks

  • Forecasting demand for trends in their introduction phase plays an important role for product manufacturers to identify future business opportunities and plan their strategies. In this talk by Karthik, a novel approach to forecast the demand of emerging trends and how the performance compares against conventional forecasting techniques will be presented.
    Tech Talks

  • Language models based on deep learning are widely used in a number of applications, including text summarization, conversational bots, Q&A, text generation, translation, semantic search, information retrieval etc. Many research and industry participants use pre-trained language models (PLMs) to build and architect these use cases. With the widespread use of pre-trained language models (PLM), there has been increased research on how to make them applicable, especially in limited resource or low latency high throughput scenarios. One of the dominant approaches is knowledge distillation (KD), where a smaller model is trained by receiving guidance from a large PLM. While there are many successful designs for learning knowledge from teachers, it remains unclear how students can learn better. Inspired by real university teaching processes, in this work we further explore knowledge distillation and propose a very simple yet effective framework, SEAD, to further improve task-specific generalization by utilizing multiple teachers. Our experiments show that SEAD leads to better performance compared to other popular KD methods and achieves comparable or superior performance to its teacher model such as BERT on total 13 tasks for the GLUE and SuperGLUE benchmarks.
    Paper Presentations

  • As we embark on an era of data-driven decisions, preserving the privacy of the underlying data is of utmost importance. Differential privacy has emerged as the go-to solution for tech giants and data vendors as it not only provides mathematical definition to privacy but also grants the ability to control the privacy parameter (noise) in the underlying data. The differentially private data aggregates sensitive individual data over multiple features and masks them with statistical noise. The conventional ML algorithms need to be tweaked to handle such aggregated data with noise. In this paper, probabilistic and Isolated learning techniques are leveraged to model differentially private data to improve the click-through rate of Ad-tech campaigns. The ability to predict the click event for different feature combinations such as URL, Ad type, Device type, etc. beforehand from the aggregated noisy data will help in developing a custom bidding strategy in a highly competitive programmatic setting (Real-time bidding)
    Paper Presentations

  • We would like to present techniques and tools to use data from large-scale, accurate simulations of models and synthetically generated, physics-based data to train and evaluate machine learning models for industrial and automotive applications. Since generating and labelling real-world data is both costly and difficult in many scenarios, the use of a digital twin and simulation data can enable the use of ML and AI models in areas which have not seen their wide-spread use so far.
    Tech Talks

  • Forecasting plays a significant role in making effective business decisions. Many products are launched in the market every year and the initial sales of the products are very much essential for gaining insights on the sustainability of the product in the market. But because of the data limitation, forecasting with traditional models becomes a huge problem and it is fraught with risks and hence, estimates can often be off the mark. So, generating accurate forecasts becomes crucial as they can help the companies in assessing overall performance, budgeting, risk management, and cost reduction. In this paper, Forecasting of Recently Launched Products with limited data using different techniques is presented. Multiple approaches have been tested and compared to a baseline strategy to solve the above problem. For testing, the results from all the models implemented have been compared, using various statistical error measures such as the Mean Absolute Percentage Error (MAPE), the Weighted Mean Absolute Percentage Error (WAPE), and Mean Absolute error (MAE).
    Paper Presentations

  • In this paper, we present our approach on using time series forecasting for managing resource allocation for one of the Social Media UPI. With the rapid increase in online payments, the business is facing challenges in resource allocation across three vendors to monitor its payment requests categorized under seven hierarchical processes labelled as Towers. Each of these Towers is further divided into various sub-processes called Products and then Alert Queues. Since the workforce allocation depends upon the incoming volume requests for these processes, it is vital to have a proper volume forecasting system to plan for the optimal resource allocation in advance. We forecasted volumes for all Products and Alert Queues across all the Towers and categorized them under three different risk buckets depending upon the Mean Absolute Percentage Error (MAPE) observed on test data. Using the forecasted volumes and the risk category the Product falls under, capacity planning for the future dates is made easier and the client was able to reduce its resources by 26%.
    Paper Presentations

  • Deep Learning has revolutionized the fields of computer vision, natural language understanding, speech recognition, information retrieval and more. Many techniques have evolved over the past decade that made models lighter, faster, and robust with better generalization. However, many deep learning practitioners persist with pre-trained models and architectures trained mostly on standard datasets such as Imagenet, MS-COCO, IMDB-Wiki Dataset, and Kinetics-700 and are either hesitant or unaware of redesigning the architecture from scratch that will lead to better performance. This scenario leads to inefficient models that are not suitable on various devices such as mobile, edge, and fog. In addition, these conventional training methods are of concern as they consume a lot of computing power. In this paper, we revisit various SOTA techniques that deal with architecture efficiency (Global Average Pooling, depth-wise convolutions & squeeze and excitation, Blurpool), learning rate (Cyclical Learning Rate), data augmentation (Mixup, Cutout), label manipulation (label smoothing), weight space manipulation (stochastic weight averaging), and optimizer (sharpness aware minimization). We demonstrate how an efficient deep convolution network can be built in a phased manner by sequentially reducing the number of training parameters and using the techniques mentioned above. We achieved a SOTA accuracy of 99.2% on MNIST data with just 1500 parameters and an accuracy of 86.01% with just over 140K parameters on the CIFAR-10 dataset.
    Paper Presentations

  • How machine learning is making customer experience more emerged and engaged.
    Tech Talks

  • Document Similarity could be a building block for many useful applications, Information Retrieval, Document Clustering, Question-Answering Systems to name a few. In modern digital world, Informative Documents are composed of Text, Images and Videos. In such a scenario, similarity based on purely Text, Image or Video may not be adequate. Hence a metrics blending similarity on all these aspects should be used. In this paper, a weighted similarity measure based on Texts and Images has been developed, using some popular opensource Machine Learning (ML) libraries. This provides a flexible and easy method, without using large training data, which often is the case with ML tasks.
    Paper Presentations

  • With increasing efforts in customer service automation, the task of customer feedback classification has assumed increasing importance. Superior text classification helps companies deliver better chat experience, improve product by correctly labelling user queries and feedback/complaints, etc. However, developing such customer feedback classification models is very difficult as often many of long tail classes do not have sufficient labelled data. We have developed an NLP approach based on the concept of contrastive learning to tackle long tail (with low volume labelled data) classification problems. We found that this approach gives good results and is computationally very efficient. This approach can be leveraged across many NLP use cases with limited data and high class skewness.
    Tech Talks

  • Pharmacovigilance has gained much prominence to identify the adverse signal originating from the use of a drug and subsequent reviews. Pharmacovigilance analytics can be applied to gain insights by integrating data related to medical products from multiple sources and applying techniques to search, compare and summarize them. Identifying the drug adversities in prior with the help of pharmacovigilance analytics can help to reduce the risk of unfavorable outcomes in selecting line of treatment. However multiple sources (medical reports, review websites, online platforms, etc.) exist for adverse signal detection which makes it challenging to extract the information of interest manually. The gap is readily identified by the authors and is addressed in this work by proposing a temporal graph neural network based approach. The existing Named Entity Recognition (NER) techniques are not efficient enough to Identify and extract the causal relationship between the drug and the events based on but not limited to onset of the event, giving insights like type of ADR. In contrast, we propose a temporal graph neural network-based approach which includes both entity recognition as well as event extraction followed by finding the causal relationship between the drug and the events. The technique uses temporal topological information, word dependency parsing, edge information, which is representing the relationship among the entities, Node Embeddings which are vectors which reflect properties of nodes in a network. The proposed methodology has exhibited improved result in comparison to the state of the art and has shown statistical significance in performance measure.
    Paper Presentations

  • The credit card fraud detection at transaction level is a fairly complex problem in the consumer lending industry across the globe. This classification problem combines the intricacies of finding a needle in a haystack (highly imbalanced classes) and hitting a moving target (dynamic fraud pattern recognition) at the same time. To add to the issues, any wrong decision (misclassification) in approving or declining a credit card transaction either leads to an immediate and direct monetary loss to the lender or compromises the core of the business – “the genuine customer’s experience”. The credit card industry has come a long way in identifying fraudulent transactions by adopting various advanced statistical techniques over years. The fraudsters, however, are not laggards. They invent channels and target new loopholes. The alerts generated by analysis of location and amount of a transaction do not suffice anymore. This paper aims at a) highlighting the significance of feature engineering in credit card fraud pattern recognition based on our industry experience, corroborated by deeper analysis of the information collected for credit card transactions, and b) comparing performance of different machine learning algorithms and quantifying their value addition to naïve models.
    Paper Presentations

  • A company which has a huge catalog with over 14 million items. The site features a diverse array of products .Some of these categories include hundreds of thousands of products; this broad offering ensures that they have something for every style and customer. However, the large size of the product catalog also makes it hard for customers to find the perfect item among all of the possible options. In this talk I will explain how to use reinforcement learning (exploration, exploration and reward methods) to have better search results.
    Tech Talks

  • Millions of packages are delivered successfully by online and local retail stores across the world every day. The proper delivery of packages is needed to ensure high customer satisfaction and repeat purchases. These deliveries suffer various problems despite the best efforts from the stores. These issues happen not only due to the large volume and high demand for low turnaround time but also due to mechanical operations and natural factors. These issues range from receiving wrong items in the package to delayed shipment to damaged packages because of mishandling during transportation. Finding solutions to various delivery issues faced by both sending and receiving parties plays a vital role in increasing the efficiency of the entire process. This paper shows how to find these issues using customer feedback from the text comments and uploaded images. We used transfer learning for both Text and Image models to minimize the demand for thousands of labeled examples. The results show that the model can find different issues. Furthermore, it can also be used for tasks like bottleneck identification, process improvement, automating refunds, etc. Compared with the existing process, the ensemble of text and image models proposed in this paper ensures the identification of several types of delivery issues, which is more suitable for the real-life scenarios of delivery of items in retail businesses. This method can supply a new idea of issue detection for the delivery of packages in similar industries.
    Paper Presentations

  • We’ve all embraced digital life. Video calling is as common today as audio calls were a decade earlier. And it is only a matter of time before “metaverse” becomes a daily reality. With that will accompany the human need to effectively express ourselves in the online world. This will happen with an amalgamation of content creation, computer vision and virtual reality breakthroughs, leading to a massive shift in the business models and economics of online presence. Most of it will be fueled by ML. We examine top such trends and identify unique opportunities for the MLverse within.
    Tech Talks

  • Incidents leading to a health and safety risk are a common occurrence in most large-scale Construction & Engineering work sites, which can lead to millions of dollars in damages and lawsuits affecting financial performance of the overall project. Early detection of these potential issues and proactive interventions could lead to avoiding of accidents or safety breaches that can occur in worksite, which will not only have huge financial benefits but also will ensure timely delivery of projects. We, part of the Construction Engineering Global Business Unit (CEGBU) in Oracle, have applied NLP based state-of-the-art Machine Learning (ML) models to classify text data from textual construction injury reports as well as correspondence data between construction project participants. The health and safety risk detection sub-system can predict if the text data is associated with (any impending) risks with high accuracy. The sub-system also enables human decision- maker to provide feedback for these correspondences, based on human experience and intuition, in case the prediction(s) made by the subsystem is incorrect. This feedback will later be fed back to the sub–system as (new) labelled training data to improve the prediction accuracy and re-inforce the sub-system over time.
    Paper Presentations

  • As deep and deeper neural networks and transfer learning are taking the world by storm, let's just take a step back and visit some fundamentals. In our hurry to learn, build and deploy, we forget to check what we have built. Is the complexity worth it or are you getting tricked by the Maya of million parameters? Join me on a discussion on the performance measurement of machine learning models.
    Tech Talks

  • Interpreting image data to neural networks is challenging. Deep convolutional neural network methods have shown promising results in interpreting image data to neural networks, with convolution and pooling operations over the traditional fully connected dense layers. The performance of these deep convolutional methods, however, is often compromised by the constraint that the convolution and pooling operations interpret the image data to neural networks by compressing the data into lower dimensions that lead to information bottleneck. To mitigate this, neural networks tend to be larger with more parameters (filters) thus increasing the computational cost (GPU Resources). In this paper, I present an information-preserving way to interpret image data to a convolutional neural net without any information bottleneck and with relatively fewer parameters, that also ensures no loss in information due to convolution or pooling operations. Uniquely my method adds two additional operations, one over the first regular convolution operation and the other operation next to the deeper convolution operation in the neural net. The first operation is to divide the image into individual frames by frame-based crop operation and then apply regular convolution, pooling operations to interpret individual frames of the image into low dimensional tensors that preserve information from being bottlenecked since operations use frames of the image instead of using total image data. The second operation is a convolution operation applied after joining all low dimensional tenors of individual frames to extract information that later passes through the rest of the layers. This idea of using frames can better help model tasks like image generation and completion as well. Using frames of data instead of the total image helps in parallelizing computations thereby drastically decreasing the computational cost and depth of a neural network. Experiments conducted on inception networks worked better with relatively small network architecture on vision tasks.
    Paper Presentations

  • This problem concerns with predicting whether the patient will get stroke in the future with predictors like age, gender, smoking status, body mass index, whether they had heart disease, whether they had hypertension etc. Since the output variable is categorical in nature, it is a classification problem. Many classification techniques are used with the help of three main Business Analytics tools such as Excel, R and Python. The data is understood through Exploratory Data Analysis, then Data Pre-Processing is done to prepare the data, various models are built on the data and finally Error metrics are used to compare the results.
    Paper Presentations

  • AI/ML has not only changed the way business is operating but also the software development lifecycle. In standard software development, it was assumed to build a small prototype for testing and then just extrapolate it if it's working. Such extrapolation doesn't work in AI/ML. You have to start thinking about the scale from day 1, even during MVP (Minimum Viable Product) development. Scaling is a huge beast that can't be tackled in the later stages. Let's discuss various challenges you face if you ignore scale from day 1, what frameworks can you use to scale your thought process, as that of your business, while designing an A/ML solution/product.
    Tech Talks

  • Despite the advances in data acquisition, storage and algorithm capabilities, the impact of Machine Learning (ML) initiatives is sometimes debated across sponsoring units and organizations. While there is a possibility of the data and models not having the desired predictive value, there is a possibility that the desired business value is not realized due to an improper implementation of the interventions from the predictions intended. We discuss the chronological nature of data richness and the increase in predictive capability, and conversely there exists an inverse relationship between most effective interventions and the richness of data available for actionable predictions across process & product lifecycles. It thus becomes imperative to view ML implementations not as technology implementations but as Business Process Re- engineering initiatives to deliver the optimal business / process returns.
    Paper Presentations

Check Schedule from 2019

Schedule 2019

Extraordinary Speakers

Meet the best Machine Learning Practitioners & Researchers from the country.