This talk will explore the competitive landscape among graph database solutions providers, discuss key trends in the graph database industry, and criteria for selecting graph database solutions. There are over thirty graph databases available currently with a smaller subset focused on RDF graphs. An optimal solution often depends on a specific use case, but some common characteristics can help identify it, such as openness, horizontal scalability, query language support, deployment options, and community support.
In healthcare, a machine learning model can mean the difference between prescribing a lifesaving medical treatment or death. This session will cover high and low stakes models in production, how they change over time, how to monitor their performance, and what to do when they underperform.
In this talk, we will trace the development of Wayfair’s deep learning capability and how deep learning is driving strong results in projects across merchandising, personalization and marketing for the company. Wayfair has invested in building deep learning experience over multiple years and has a large team of data scientists, machine learning engineers, product managers and more dedicated to deploying these solutions at scale. We will discuss learnings from this growth process, highlight successes (and failures) and how deep learning has evolved to be a core aspect of the work the Wayfair team is doing.
Dan oversees Wayfair’s Data Science team, which works on projects stretching across the customer experience in Marketing, Merchandising & Storefront. Prior to Wayfair, Dan was a strategy consultant at the Boston Consulting Group in Chicago. Dan holds a B.A. from Columbia University... Read More →
Thursday August 23, 2018 10:30am - 11:00am EDT
Room 406Boston University Questrom School of Business, 595 Commonwealth Avenue Boston
Organizations of various sizes and across verticals face challenges with how their data evolves as they scale. One of the most significant challenges as they look to integrate data from additional systems is duplication of entities across systems. These duplicated entities may not have clear criteria to help engineers identify and combine them. However, as organizations are increasingly looking to aggregate data sources—both internal and external—addressing this challenge is critical to realizing value from the data. At medium- to large-scale data, combining the data sets can't be done without the aid of software. Traditional techniques for tackling this challenge involved rule-based systems to look at every pairwise combination and determine whether there was a match or not. More recently, machine learning approaches that block and score probable matches allow engineers to set an acceptable probability threshold, as well as validate and tune the algorithm to help it continuously improve.
In this presentation, we will present entity resolution in the context of a detailed example scenario using physician data matching. We will walk through a machine learning model we created by forking and building on an open source entity resolution library, dedupe.io. We will describe how the model initially performed, how we evaluated it, and how we improved it. Finally, we will walk through high-level workflows for performing entity resolution on your own data.
DevOps is a fairly well understood domain. Analytic Ops on the other hand is an emerging area. Analytic Ops attempts to address the challenges of bringing process, people and technology together to create the foundation of an organization’s analytic framework. This session will discuss how crucial it is for an organization to have a well defined analytics process to retain its competitive edge and how to go about setting it. It will also include a short demo of Think Big’s Analytic Ops Accelerator that provides an end-to-end, flexible framework for the orchestration, deployment and management of analytic models at scale.
What if I told you I had evidence of a serious threat to American national security – a terrorist attack in which a jumbo jet will be hijacked and crashed every 12 days. Thousands will continue to die unless we act now. This is the question before us today – but the threat doesn’t come from terrorists. The threat comes from climate change and air pollution.
We have developed an artificial neural network model that uses on-the-ground air-monitoring data and satellite-based measurements to estimate daily pollution levels across the continental U.S., breaking the country up into 1-square-kilometer zones. We have paired that information with health data contained in Medicare claims records from the last 12 years, and for 97% of the population ages 65 or older. We have developed statistical methods and computational efficient algorithms for the analysis over 460 million health records.
Our research shows that short and long term exposure to air pollution is killing thousands of senior citizens each year. This data science platform is telling us that federal limits on the nation’s most widespread air pollutants are not stringent enough.
This type of data is the sign of a new era for the role of data science in public health, and also for the associated methodological challenges. For example, with enormous amounts of data, the threat of unmeasured confounding bias is amplified, and causality is even harder to assess with observational studies. These and other challenges will be discussed.
Unstructured free text is plentiful and valuable, for example: Doctor's notes, news stories, call-center notes. We would like to take advantage of such text for warnings of serious medical conditions, classifying reports, and identifying customers likely to leave for a competitor.
Text can be hard to use for predictive modeling, but in some respects it is also easier to use than structured numeric data. This talk will give a high-level overview of opportunities and challenges in using text in predictive models, and survey technical approaches for representing and modeling text.
Many important care quality measures, especially those that relates to communication, are not properly operationalized in the clinical context because they exist as unstructured free text in notes and other places. Our work aims to develop techniques that we can extract documentation of some of these important conversations for seriously ill patients. Our deep neural network attained 90%+ precision and recall for the documentation of care preferences.
The increasing number of Electronic Health Record (EHR) clinical free text documents has urged the need to build novel clinical Natural Language Processing (NLP) solutions towards optimizing patient outcomes. Deep Learning (DL) techniques have demonstrated superior performance over other Machine Learning (ML) approaches for various NLP tasks in recent years. This talk will present a brief overview of various DL-driven clinical NLP algorithms developed in the Artificial Intelligence lab at Philips Research - such as diagnostic inferencing from unstructured clinical narratives, clinical paraphrase generation, and medical image caption generation.
Senior Scientist, Artificial Intelligence Lab, Philips Research North America
Sadid Hasan is a Senior Scientist at the Artificial Intelligence Lab in Philips Research, Cambridge, MA. His recent work involves solving problems related to clinical question answering, paraphrase generation, and medical image caption generation using Deep Learning. Before joining... Read More →
Thursday August 23, 2018 2:00pm - 2:45pm EDT
AuditoriumBoston University Questrom School of Business, 595 Commonwealth Avenue Boston
Ride-sharing platforms employ surge pricing to match anticipated capacity spillover with demand. We develop an optimization model to characterize the relationship between surge price and spillover. We test predicted relationships using a spatial panel model on a dataset from Uber's operation. Results reveal that Uber's pricing accounts for both capacity and price spillover. There is a debate in the management community on the efficacy of labor welfare mechanisms associated with shared capacity. We conduct counterfactual analysis to provide guidance in regards to the debate, for managing congestion, while accounting for consumer and labor welfare through this online platform.
Essential Administrative functions (aka back office functions) for health care we very well suited for assistance and automation with Deep Learning. These important functions are often too complex for rules but with lots of well-labeled data, they are a good fit for Deep Learning. OptumLabs Center for Applied Data Science will share how they design and train Deep Learning Neural Networks for these use cases.
The recent hype about automated machine learning has focused on the modeling part of the pipeline, while the critical step of feature engineering is left out of the discussion. Feature engineering, the process of extracting predictor variables from a dataset, largely determines the success or failure of a machine learning project. That is why automating feature engineering has long been a dream for machine learning practitioners. In this talk, we will walk through several real-world use cases using Featuretools, an open source library for automated feature engineering developed and maintained by Boston-based startup Feature Labs. We will show how automated feature engineering can reduce development time by 10x, deliver better predictive performance, create meaningful features with real-world insights, and prevent data leakage.
This presentation will focus on showing both supervised and unsupervised learning methods to work with claims data and how they can complement each other. A supervised method will look at CKD patients at-risk to develop ESRD, and unsupervised approach will look at classification of patients that tend to develop this disease faster than others.
