Schedule

Automating decisions require a system to consider more than just a data-driven prediction. Real-world decisions require additional constraints and fuzzy objectives to ensure they're robust and consistent with business goals. Brian Keng takes a deep dive into how to leverage modern machine learning methods and traditional mathematical optimization techniques for decision automation.

Recent advances in machine learning frameworks for the browser such as TensorFlow provides the opportunity to craft truly novel experiences within frontend applications. Victor Dibia explores the state of the art for machine learning in the browser using TensorFlow and outlines its use in the design of Handtrack.js—a library for prototyping real-time hand detection in the browser.

Heitor Murilo Gomes and Albert Bifet introduce you to a machine learning pipeline for streaming data using the streamDM framework. You'll also learn how to use streamDM for supervised and unsupervised learning tasks, see examples of online preprocessing methods, and discover how to expand the framework by adding new learning algorithms or preprocessing methods.

Taming massive deep learning models, data, and training times requires new way of thinking. Shital Shah explores new tools and methods to better understand AI. Explaining the decisions made by AI not only helps us accelerate its development but also make it safe and more trustworthy.

Apache Spark is the largest compute engine at Facebook by CPU. Sameer Agarwal dives into the story of how Facebook optimized, tuned, and scaled Apache Spark to run on clusters of tens of thousands of machines, processing hundreds of petabytes of data, and being used by thousands of data scientists, engineers, and product analysts every day.

Unmanaged and foreign devices in the corporate networks pose a security risk, and the first step toward reducing this risk is the ability to identify them. Ryan Foltz walks you through a comprehensive device management machine learning model based on deep learning that performs anomaly detection based on only device names to flag devices that do not follow naming structures.

Sajan Govindan outlines CERN’s research on deep learning in high energy physics experiments as an alternative to customized rule-based methods with an example of topology classification to improve real-time event selection at the Large Hadron Collider. CERN uses deep learning pipelines on Apache Spark using BigDL and Analytics Zoo open source software on Intel Xeon-based clusters.

Giant hogweed is a highly toxic plant. Naoto Umemori and Masaru Dobashi aim to automate the process of detecting the plant with technologies like drones and image recognition and detection using machine learning. You'll see how they designed the architecture, took advantage of big data and machine and deep learning technologies (e.g., Hadoop, Spark, and TensorFlow), and the lessons they learned.

The application of smoothing and imputation strategies is common practice in predictive modeling and time series analysis. With a technique-agnostic approach, Anjali Samani provides qualitative and quantitative frameworks that address questions related to smoothing and imputation of missing values to improve data density.

Time series forecasts depend on sensors or measurements made in the real, messy world. The sensors flake out, get turned off, disconnect, and otherwise conspire to cause missing signals. Signals that may tell you what tomorrow's temperature will be or what your blood glucose levels are before bed. Alfred Whitehead and Clare Jeon explore methods for handling data gaps and when to consider which.

Machine learning (ML) gets a lot of hype, but its classical predecessors are still immensely powerful, especially in the time series space, and classical algorithms outperform machine learning methods in time series forecasting. Anais Dotis dives into how she used the Holt-Winters forecasting algorithm to predict water levels in a creek.

Chad Scherrer explores the basic ideas in Soss, a new probabilistic programming library for Julia. Soss allows a high-level representation of the kinds of models often written in PyMC3 or Stan, and offers a way to programmatically specify and apply model transformations like approximations or reparameterizations.

Jeroen Janssens dives into stochastic outlier section (SOS), an unsupervised algorithm for detecting anomalies in large, high-dimensional data. SOS has been implemented in Python, R, and, most recently, Spark. He illustrates the idea and intuition behind SOS, demonstrates the implementation of SOS on top of Spark, and applies SOS to a real-world use case.

Alejandro Saucedo demystifies AI explainability through a hands-on case study, where the objective is to automate a loan-approval process by building and evaluating a deep learning model. He introduces motivations through the practical risks that arise with undesired bias and black box models and shows you how to tackle these challenges using tools from the latest research and domain knowledge.

Sandra Carrico explores mixed formal learning, explains it, and outlines one machine learning example that previously used large numbers of examples and now learns with either zero or a handful of training examples. It maps apparently idiosyncratic techniques to mixed formal learning, a general AI architecture that you can use in your projects.

Mumin Ransom gives an overview of the data management and privacy challenges around automating ML model (re)deployments and stream-based inferencing at scale.

Graphs are a powerful way to represent knowledge. Organizations, in fields such as biosciences and finance, are starting to amass large knowledge graphs, but they lack the machine learning tools to extract insights from them. David Mack offers an overview of what insights are possible and surveys the most popular approaches.

You're a SaaS company operating on a cloud infrastructure prior to the machine learning (ML) era and you need to successfully extend your existing infrastructure to leverage the power of ML. Jing Huang and Jessica Mong detail a case study with critical lessons from SurveyMonkey’s journey of expanding its ML capabilities with its rich data repo and hybrid cloud infrastructure.

Frequently, Kafka is just a piece of the stack that lives in production that often times no one wants to touch—because it just works. Alon Gavra outlines how Kafka sits at the core of AppsFlyer's infrastructure that processes billions of events daily.

Narvar provides next-generation posttransaction experience for over 500 retailers. Karthik Ramasamy and Anand Madhavan take you on the journey of how Narvar moved away from using a slew of technologies for their platform and consolidated its use cases using Apache Pulsar.

Jonghyok Lee Chon Yong Lee discuss T-CORE, SK Telecom’s monitoring and service analytics platform, which collects system and application data from several thousand servers and applications and provides a 3D visualization of the real-time status of the whole network. Join in to hear lessons learned during development.

Stavros Kontopoulos and Debasish Ghosh explore online machine learning algorithm choices for streaming applications, especially those with resource-constrained use cases like IoT and personalization. They dive into Hoeffding Adaptive Trees, classic sketch data structures, and drift detection algorithms from implementation to production deployment, describing the pros and cons of each of them.

Data scientists create and test hundreds or thousands more models than in the past. Models require support from both real-time and static data sources. As data becomes enriched, and parameters tuned and explored, there's a need for versioning everything, including the data. Jim Scott examines the very specific problems and approaches to fix them.

Machine learning (ML) will fundamentally change the way we build and maintain applications. Diego Oppenheimer dives into how you can adapt your infrastructure, operations, staffing, and training to meet the challenges of the new software development life cycle (SDLC) without throwing away everything that already works.

As an increasing level of automation becomes available to data science, the balance between automation and quality needs to be maintained. Applying DevOps practices to machine learning workloads brings models to the market faster and maintains the quality and integrity of those models. Sireesha Muppala, Shelbee Eigenbrode, and Randall DeFauw explore applying DevOps practices to ML workloads.

Leveraging polyglot solutions for your time series data can lead to issues including engineering complexity, operational challenges, and even referential integrity concerns. Michael Freedman explains why, by re-engineering PostgreSQL to serve as a general data platform, your high-volume time series workloads will be better streamlined, resulting in more actionable data and greater ease of use.

Omkar Joshi and Bo Yang offer an overview of how Uber’s ingestion (Marmary) and observability team improved performance of Apache Spark applications running on thousands of cluster machines and across hundreds of thousands+ of applications and how the team methodically tackled these issues. They also cover how they used Uber’s open-sourced jvm-profiler for debugging issues at scale.

Building a reliable big data platform is extremely challenging when it has to store and serve hundreds of petabytes of data in real time. Reza Shiftehfar reflects on the challenges faced and proposes architectural solutions to scale a big data platform to ingest, store, and serve 100+ PB of data with minute-level latency while efficiently utilizing the hardware and meeting security needs.

Vitaliy Baklikov and Dipti Borkar explore how DBS Bank built a modern big data analytics stack leveraging an object store even for data-intensive workloads like ATM forecasting and how it uses Alluxio to orchestrate data locality and data access for Spark workloads.

Big data computing and high-performance computing (HPC) evolved over the years as separate paradigms. With the explosion of the data and the demand for machine learning algorithms, these two paradigms increasingly embrace each other for data management and algorithms. Supun Kamburugamuve explores the possibilities and tools available for getting the best of HPC and big data.

The Large Scale Survey Telescope (LSST) is one of the most important future surveys. Its unique design allows it to cover large regions of the sky and obtain images of the faintest objects. After 10 years of operation, it will produce about 80 PB of data in images and catalog data. Petar Zecevic explains AXS, a system built for fast processing and cross-matching of survey catalog data.

Jason Wang and Sushant Rao offer an overview of cloud architecture, then go into detail on core cloud paradigms like compute (virtual machines), cloud storage, authentication and authorization, and encryption and security. They conclude by bringing these concepts together through customer stories to demonstrate how real-world companies have leveraged the cloud for their big data platforms.

Nikki Rouda and Janisha Anand demonstrate how to deduplicate or link records in a dataset, even when the records don’t have a common unique identifier and no fields match exactly. You'll also learn how to link customer records across different databases, match external product lists against your own catalog, and solve tough challenges to prepare and cleanse data for analysis.

In this session, Scott Castle, General Manager at Sisense and former VP of Product at Periscope Data, will discuss lessons learned from scaling up Periscope Data to support incredibly large volumes of data and queries from its data teams.

Spark on Kubernetes is a winning combination for data science that stitches together a flexible platform harnessing the best of both worlds. Jordan Volz gives a brief overview of Spark and Kubernetes, the Spark on Kubernetes project, why it’s an ideal fit for data scientists who may have been dissatisfied with other iterations of Spark in the past, and some applications.

Data has always been and will always be relational. NoSQL databases are gaining in popularity, but that doesn't change the fact that the data is still relational, it just changes how we have to model the data. Rick Houlihan dives deep into how real entity relationship models can be efficiently modeled in a denormalized manner using schema examples from real application services.

With cloud object storage (e.g., S3, ADLS) one expects business intelligence (BI) applications to benefit from the scale of data and real-time analytics. However, traditional BI in the cloud surfaces nonobvious challenges. Shant Hovsepian examines service-oriented cloud design (storage, compute, catalog, security, SQL) and how native cloud BI provides analytic depth, low cost, and performance.

With cheap and scalable storage services such as S3 and ADLS, it's never been easier to dump data into a cloud data lake. But you still need to secure that data and be sure it doesn't leak. Tomer Shiran and Jacques Nadeau explore capabilities for securing a cloud data lake, including authentication, access control, encryption (in motion and at rest), and auditing, as well as network protections.

Fine-grained data protection at a column level in data lake environments has become a mandatory requirement to demonstrate compliance with multiple local and international regulations across many industries today. Owen O'Malley dives into how column encryption in ORC files enables both fine-grain protection and audits of who accessed the private data.

Ruixin Xu, Long Tian, and Yu Zhou explore an experiment run using Spark and Jupyter notebooks as a replacement for existing IDE-based tools for internal DevOps. The Spark-based solution improved the diagnosis performance significantly, especially for a complex job with a large profile, and leveraging the Jupyter notebooks brings the benefit of fast iteration and easy knowledge share.

Gayle Bieler explains how she built a thriving center for data science within a large, well-respected nonprofit research institute and shares some of its most impactful projects and best adventures to date, that have solved important national problems, improved local communities, and transformed research.

This talk will explore some of the philosophy around the concept of explaining a model given the colloquial definition is partially recursive. It will cover the lens banking regulation places on this philosophical basis and expand into techniques used for these well governed aspects.

Paco Nathan outlines the history and landscape for vendors, open source projects, and research efforts related to AutoML. Starting from the perspective of an AI expert practitioner who speaks business fluently, Paco unpacks the ground truth of AutoML—translating from the hype into business concerns and practices in a vendor-neutral way.

Jonathan Tudor and Ross Schalmo explore how GE Aviation made it a mission to implement self-service data. To ensure success beyond initial implementation of tools, the data engineering and analytics teams created initiatives to foster engagement from an ongoing partnership with each part of the business to the gamification of tagging data in a data catalog to forming a published dataset council.

Dean Wampler dives into how (and why) to integrate ML into production streaming data pipelines and to serve results quickly; how to bridge data science and production environments with different tools, techniques, and requirements; how to build reliable and scalable long-running services; and how to update ML models without downtime.

Language shapes our thinking, our relationships, our sense of self. Conversation connects us in powerful, intimate, and often unconscious ways. Jonathan Foster explains why, as we design for natural language interactions and more humanlike digital experiences, language—as design material, conversation, and design canvas—reveals ethical challenges we couldn't encounter with GUI-powered experiences.

If you've ever been confused about what it takes to be a data scientist or curious about how companies recruit, train, and manage analytics resources, Usama Fayyad and Hamit Hamutcu are here to explore insights from the most comprehensive research effort to date on the data analytics profession and propose a framework for the standardization of roles and methods for assessing skills.

T-Mobile successfully improved the quality of voice calling by analyzing crowdsourced big data from mobile devices. Alex Yoon walks you through how engineers from multiple backgrounds collaborated to achieve 10% improvement in voice quality and why the analysis of big data was the key to the success in bringing a better voice call service quality to millions of end users.

At MakeMyTrip customers were using voice or email to contact agents for postsale support. In order to improve the efficiency of agents and improve customer experience, MakeMyTrip developed a chatbot, Myra, using some of the latest advances in deep learning. Madhu Gopinathan and Sanjay Mohan explain the high-level architecture and the business impact Myra created.

Companies that harness creativity and data in tandem have growth rates twice as high as companies that don’t. David Boyle shares lessons from his successes and failures in trying to do just that across presidential politics, with pop stars, and with power brands in the world of luxury goods. Join in to find out how analysts can work differently to build these partnerships and unlock this growth.

As an early adopter of data science, machine learning, and AI, Deutsche Bank's analytics function is trailblazing new ways to drive revenues, lower costs, and reduce risk across all areas of the group. John Allen shares how his team combines commercial offerings with open source technologies to revolutionize legacy processes and transform the way the bank uses technology to drive innovation.

Miscommunication between business leaders and technical experts can doom even the best data science project. Don’t let it drive you insane! In this session, we’ll dissect many flavors of communication failure, from goal misalignment to technical misunderstanding. Then, we’ll explore practical ways to bridge these gaps.

Failures or issues in a product or service can negatively affect the business. Detecting issues in advance and recovering from them is crucial to keeping the business alive. Join Akshay Rai to learn more about LinkedIn's next-generation open source monitoring platform, an integrated solution for real-time alerting and collaborative analysis.

As new digital platforms emerge and governments look at new ways to engage with citizens, there's an increasing awareness of the role these platforms play in shaping public participation and democracy. Audrey Lobo-Pulo, Annette Hester, and Ryan Hum examine the design attributes of civic engagement technologies and their ensuing impacts and an NEB Canada case study.

California is following the EU's GDPR with the California Consumer Protection Act (CCPA) in 2020. Penalties for non-compliance, but many companies aren't prepared for this strict regulation. This session will explore the capabilities your data environment needs in order to simplify CCPA and GDPR compliance, as well as other regulations.

AI isn't magic. It’s still hard work. Daniel Hernandez explains why having the technology alone isn't enough; it requires a thoughtful and well-architected approach.

Jed Dougherty takes a deep dive into an often overlooked aspect of the data science lifecycle: model deployment. Once they’ve constructed a data science model that does a good job accurately predicting their test set, many data scientists think the job is over. But really, it’s just begun.

Transforming data into a trusted business asset that informs decision making requires giving teams access to a powerful platform that makes it easy to harness data across the enterprise. Jim Cushman and Piyush Jain detail how Progressive uses Collibra to transform the way data is managed and used across the organization, driving real business value.

Edward Jezierski and Jackie Nichols demonstrate how Cognitive Services Personalizer works with your content and data, how it autonomously learns to make optimal decisions, how you can add it to your app with two lines of code, and what’s under the hood. Then they share the results Personalizer achieved on the Xbox One home page as well as best practices for applying it in your applications today.

Paul Scott-Murphy dives into the options that exist for cloud migration and their advantages and disadvantages, what cloud vendors do and don't offer to support large-scale migration, the business risks associated with large-scale cloud migration, and how to migrate analytics data at scale for immediate use in Spark without disrupting on-premises operations.

Stephan Ewen details how stream processing is becoming a "grand unifying paradigm" for data processing and the newest developments in Apache Flink to support this trend: new cross-batch-streaming machine learning algorithms, state-of-the-art batch performance, and new building blocks for data-driven applications and application consistency.

Healthcare’s reliance on comprehendible data is critical to the mission of providing optimal and affordable care. Charles Boicey takes a deep dive into how the application of technology, such as machine learning, is paramount to the modernization of healthcare that provides its professionals with fully integrated and complete medical records.

Paul Wolmering explores the key characteristics for building an Agile data warehouse and defines a reference architecture for hybrid data.

Clinical trial data analysis can be a complex process. The data is typically hand-coded and formatted differently and is required to be delivered in an FDA-approved format. Matt Derda and Yogesh Prasad explain how IQVIA built its Clean Patient Tracker and how it enabled agility and flexibility for end users of the platform, from data acquisition to reporting and analytics.

Learn how you can overcome the challenges of traditional OLAP solutions and scale BI to deliver quick insights to business users across your enterprise

After 40 years of apps, enterprise companies now realize that building or buying an application for every use case has become a major threat to their ability to leverage and protect their core data assets. Dan DeMers provides a live demo of Cinchy, the world’s first data collaboration platform.

Program chairs Ben Lorica, Doug Cutting, and Alistair Croll welcome you to the second day of keynotes.

Machine learning and artificial intelligence are no longer science fiction, so now you have to address what it takes to harness their potential effectively, responsibly, and reliably. Based on lessons learned at Google, Cassie Kozyrkov offers actionable advice to help you find opportunities to take advantage of machine learning, navigate the AI era, and stay safe as you innovate.

Daniel Hernandez takes a deep dive into how, with a unified, prescriptive information architecture, organizations can successfully unlock the value of their data for an AI and multicloud world.

In this keynote, we’ll introduce you to the new 100% open source Cloudera Data Platform (CDP), the world’s first enterprise data cloud. CDP is hybrid and multi-cloud, delivering the speed, agility, and scale you need to secure and govern your data anywhere from the edge to AI.

Barbara Eckman shares lessons learned from early big data mistakes and the progress her team at Comcast is making toward a postrevolutionary big data vision.

Microsoft has an ecosystem spanning research, gaming, and the cloud that's advancing reinforcement learning (RL) and putting it into everyday use. Join Edward Jezierski to see where RL is used practically across Microsoft and imagine the opportunities that exist for your business today.

The Strata Data Awards recognize the most innovative startups, leaders, and data science projects from Strata sponsors and exhibitors around the world. Join us during keynotes for the announcement of the winners.

Language shapes our thinking, our relationships, our sense of self. Conversation connects us in powerful, intimate, and often unconscious ways. Jonathan Foster explains why, as we design for natural language interactions and more humanlike digital experiences, language—as design material, conversation, and design canvas—reveals ethical challenges we couldn't encounter with GUI-powered experiences.

Jed Dougherty presents the trailer of the upcoming _Data Science Pioneers_ documentary about the passionate data scientists driving us toward technological revolution. Cut through the hype with _Data Science Pioneers_ and see what it really means to be a data scientist.

Based on a critical evaluation of the iconic yield curve chart, Alan Smith argues that combining visualization (data to pixels) with sonification (data to pitch) offers potential to improve not only aesthetic multimedia experiences but also an opportunity to take the presentation of data into the rapidly expanding universe of screenless devices and products.

Program chairs, Ben Lorica, Doug Cutting, and Alistair Croll, offer closing remarks.

Topic Table discussions are a great way to informally network with people in similar industries or interested in the same topics.

AI will be the most disruptive class of technologies over the next decade, fueled by near-endless amounts of data and unprecedented advances in deep learning. Brittany Bogle walks you through how to address some of the major AI challenges, like trust, talent, and data.

Gather before keynotes on Thursday morning to enjoy casual conversation while meeting fellow attendees. If one of your goals at Strata is to meet new people, this session will jumpstart your networking with other attendees.

Topic Table discussions are a great way to informally network with people in similar industries or interested in the same topics.

Join Strata Business Summit speakers and attendees for a networking lunch on Thursday.