Welcome by Jenny Lundberg, Senior Director, Global Platform Technology Enablement, Global Product Strategy at SAP Concur

• History & Motivation
• Plan for the day & Logistics

Welcome to WiDS: Inspiring Diversity in Data Science.

• Judy Logan, Co-Director, Women in Data Science (WiDS) Conference
  Stanford ICME

Panel Abstract: The introduction of AI technology in products affects the way companies operate and its culture. Often, Data Scientists develop models for PoCs. However, a very small percentage of these models reach the production phase. The gap between a good Machine Learning (ML) model and a production-ready ML application is still widely misunderstood. Ideally, deploying ML models should be a collaboration between Data Science, Engineering, Product Management, Design and the Legal team. Yet, often the engineering organization is tasked with turning ML models to a software product.  As part of a data-driven organization how do Data Scientist deal with topics such as sample bias, user privacy and the legal implications associated with their models? This panel will highlight the risks that come along with putting Data Science in production.

• Michelle Grover, SVP of Development, Concur, and board member for Techtonica
• Srujana Kaddevarmuth, Director of Data Science & Value Realization, Walmart Labs

Moderator: Sally Lawler Kennedy, Director, SAP Innovation and Design Research, SAP Design and Co-Innovation Center, Americas

Panel Abstract: There are 9 key issues in AI ethics according to the World Economic Forum that span the entire lifecycle AI: how do we build AI, for what purpose do we use AI, what impacts does AI have on society? We’ll dig into these questions, as well as how to incorporate these questions in your own work, with a diverse group of women.

• Lisa Nash, Senior Designer & Data Scientist, IDEO
• Eva Sasson, Product Growth & Marketing, Persona
• Natalia Bilenko, Freelance Machine Learning Engineer

Moderator: Nicole Helmer, Head of Strategy, Innovation Center Network, SAP

Title: What’s the Cost of Being Wrong? How Do We Get it Right?

Abstract: The Human Rights Data Analysis Group (HRDAG) uses methods from statistics and computer science to help answer questions about mass violence. Examples from HRDAG’s work will be presented to illustrate the many steps involved, including evaluating data quality, determining what’s missing from observed data, conducting analyses that withstand adversarial political and/or legal climates, and explaining analyses to non-technical audiences, including judges and prosecutors. The potential harm that can be done when inappropriately analyzing and interpreting incomplete and imperfect data will be especially highlighted, including questions such as: How can we develop approaches to help us identify the cases where analytical tools can do the most good, and avoid or mitigate the most harm? We propose starting with two simple questions: What is the cost of being wrong? And who bears that cost?

• Megan Price, Executive Director of the Human Rights Data Analysis Group

1. Why your project needs to be an app

Before you get to production, you must explain and sell your work to teammates and customers. Ensure your success by creating apps that demo your work and allow others to explore your project. Streamlit is an open-source tool that allows you to create beautiful data apps in just a few lines of code. (30 minutes)

Speaker: Amanda Kelly, Streamlit

2. SAP Ariba Reinforcement Learning Approach Towards Recommendations in a Procurement Marketplace

Recommendation systems are a standard element of marketplaces. At SAP Ariba we use a Reinforcement Learning to provide recommendations to new users with minimum data while staying compliant to GDPR. Join us to learn the techniques used to capture user behavior and provide relevant recommendations. (30 minutes)

Speaker: Cynthia Wood, SAP

3. NLP as a powerful tool in Health Data Science: Unstructured Text Analysis in Electronic Health Records to Characterize Sepsis Presentation 

With the realization of the value of unstructured data in a wide range of sectors, NLP has become essential to the data science toolkit. We will explore NLP as a tool in health DS, and how Kaiser Permanente research is using NLP to characterize diseases that are tough to identify, such as sepsis. (30 minutes)

Speaker: Meghana Bhimarao, Kaiser Permanente’s Division of Research

4. What Makes Data Science "Fast" ?

ML algorithms often need to execute several billion operations per second. For them to be viable solutions for real time applications, one usually needs to exploit: efficient architecture search, full utilization of GPU cores and maximum CPU parallelism. Let’s explore what makes data science “fast”. (30 minutes)

Speaker: Tanya Piplani, SAP

1. Productize Machine Learning and Drive Customer Success in CRM

One way to leverage ML is to productize it. However building a successful model offline is not enough to be successful in production. What are the gaps between offline and production and what are the steps to take? Let’s share our ML stories from CX and make your model valuable to customers too! (30 minutes)

Speaker: Jing Ye, SAP

2. Like a Pro: Making Your Data Science Code Production-Ready 

PoCs are a great first step in showcasing your latest machine learning model. But what good is that model if it’s never used? When making your model generally available, your code must be robust and easily delivered. This talk contains tips for making your data science code “production-ready.” (30 minutes)

Speakers: Richa Namballa & Shruti Bhargava, SAP

3. Research to Production with Open Source Software: High performance model inferencing using ONNX

As ML developments shift from research to real world applications, we encounter a host of deployment challenges. Using the interoperable ONNX model format and the cross platform ONNX Runtime inference engine, we can streamline the process to deploy models performantly across a variety of hardware. (30 minutes)

Speakers: Faith Xu & Emma Ning, Microsoft

1. How do algorithms become biased?

There’s bias in algorithms - how does this happen? In this talk, we walk through a step by step example of building a prediction algorithm, focused on areas where bias could be inadvertently introduced. We then look at when algorithms were used to make biased decisions, and what we can do about it. (30 minutes)

Speaker: Eva Sasson, Persona

2. Fueling Machine Learning with Feature Engineering

Modern times are buzzing with Machine Learning & for these state-of-the-art algorithms to work their magic it’s important to lay a strong foundation via Feature Engineering. This talk reviews the what is Feature Engineering & how to actually harmonize valuable information from volumes of crude data. (30 minutes)

Speaker: Jigyasa Grover, Twitter

1. Understanding and Addressing Bias in AI Diagnostic Systems

Experts predict that artificial intelligence (AI) will completely revolutionize the field of medical diagnostics. However, we must consider the impact of bias. How do we recognize and address different kinds of bias, and avoid models that perpetuate bias in healthcare and other fields? (30 minutes)

Speaker: Julie Vaughn, MIT

2. Data Science in Cloud Data Protection

In this talk, it will be talking about how to implement data science techniques to protect data on public cloud. In the meanwhile, it will also discuss how to deal with risks come along with that, given the needs to comply with various data-protection regulations like GDPR. (30 minutes)

Speaker: Shmily Wang, SAP

3. Adverse Impact - a data science perspective on a legal data-evaluation process

Legal requirements are becoming a critical part of data science productization. We will examine current legal requirements for evaluating unintentional discriminatory effects via adverse impact, to build a better understanding of what the future may hold for algorithm evaluation. (30 minutes)

Speaker: Elinor Brondwine, Workday

1. Open Neural Network eXchange (ONNX): Accelerate and operationalize models for deployment

The modern data scientist has a large toolbox, from frameworks like PyTorch and TF to Scikit-Learn and Keras. These are useful for model training but not optimized for latency-sensitive deployment with real customers. The solution? ONNX. Come to our hands-on workshop to learn more. (1 hour)

Speakers: Vinitra Swamy, Ashwini Khade & Cecilia Liu, Microsoft

2. Turn Python Scripts into Beautiful Data Apps

Quickly create your own custom data tools for visualization, debugging, comparing models and presenting data - all in Python. This tutorial covers how to use Streamlit, a new open-source app framework built specifically for data scientists to create beautiful, performant tools. (1 hour)

Speaker: Amanda Kelly, Streamlit

Panel Abstract: There’s no doubt that COVID-19 will change the Future of Work. According to a recent McKinsey Global Survey, 87% of executives said they were experiencing skill gaps in the workforce or expected them within a few years.  Join us to learn how priorities for CHROs have changed, how companies can proactively manage the transition to a new future, and the latest trends in HR Tech.

• Joanna Riley, CEO and Founder of Censia
• Margaret Tomaszczuk, Head of Product and Customer Experience at Cultivate
• Sirisha Ayyagari, VP of Engineering, Cross Suite Development at SAP SuccessFactors

Moderator: Domitilla Caputo, Head of SAP.iO Foundry San Francisco, SAP

Panel abstract: Showcase of SAP’s culture, the work of a data scientist, and ways to enter or grow a DS career in SAP.

Moderator: Caitie Sullivan, SAP