Women in Data Science Worldwide

After she earned her BA, MA, and PhD in economics from the University of Maryland, Menglin Cao spent six years at Fannie Mae before joining Wells Fargo. Over the past 15 years, she has seen a major shift in how financial institutions use data to drive business decisions. In the past, many business decisions were based upon the experience and judgement of senior executives, but today every decision must be backed up by data and analytics. 

Many aspects of the financial services that leverage data are great candidates for AI and ML models, such as customer experience, revenue generation, and risk management. She wants to ensure they build models that are fair, explainable, interpretable, and able to generate value.

She offers several examples about best practices for success.

To be useful, the data must first be focused on a foundational business question. “Data can tell many stories, but until you bring the story to the businesses and they can relate the story to a question, experience or concern that they have, it’s not going to lead to an actionable insight.” The journey to find that answer requires data science, data algorithms, operational procedures, implementations, and integrations of systems, but without the right question to begin with, it doesn’t lead anywhere. 

It’s also important to build a solution in a matrix organization rather than top-down approach, working with the technology team, model developers, subject matter experts, the model governance and validation teams. And data management has become centralized, shifting from silos to a central enterprise data lake where every line of business and all models depend on one source of truth. This removes the redundancies and duplications that silo-based data warehouse may create.

To ensure that models are fair, they continually test them for bias. One way to make sure that these data are representative of the entire population is to conduct a sensitivity analysis. They can see if the model can stand on its own by “shocking it” in different ways to find any vulnerabilities that need to be addressed.

She acknowledges that there are not yet many women leaders in fintech and the financial sector but she is hopeful that can change if women are given the opportunity. “I am in the position that I am in because my leaders trusted me and gave me a chance.” She hopes that leaders will continue to give diverse people a chance to grow and demonstrate what they can do. And then once the opportunity's given to you, then the rest depends on what you make out of it. 

RELATED LINKS
Connect with Menglin Cao on LinkedIN
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

Karen Hao trained as a mechanical engineer and then joined a Silicon Valley startup, thinking that technology was the best means to create social change. While surrounded by smart people who were also passionate about using technology for social change, she soon discovered there were no incentives or pathways to accomplish this. “When you're inside a technology company and you're thinking this is going to help change the world, you're often blind to unattended consequences of your work,” she says.

She decided to transition to a career in journalism where she could create social change by raising awareness about social impacts of technologies like AI, and how big tech companies engage in “ethics washing” to protect their profits. She is intrigued by the way that incentives shape the work that is done at a systemic level. She says every tech giant suffers from issues at the systemic level where there are people who deeply care about ethics within the organization, but it doesn't mean the company is willing to change the way their profitable technologies work. And employees are disincentivized to do this work because they could be fired.

A high-profile example was the ethical AI team at Google was doing great work critiquing some of Google’s practices and tried to get the paper published. Google refused to let them publish it, censored their research and then fired both of the team leads. It later came out that this was just one instance of academic censorship, but Google had told many other researchers to strike a positive tone when talking about technology being developed by Google.

For her article, How Facebook Got Addicted to Spreading Misinformation, she did a nine-month investigation into Facebook's responsible AI team that was supposed to be understanding and mitigating the unintended consequences of Facebook's algorithms. She found the team was focused on specific unintended consequences that are good for Facebook's growth like AI bias. It completely ignored the most important harms of Facebook's algorithms—misinformation, amplification, polarization, exacerbation, especially in the wake of the January 6 capital riots—because addressing them would undermine Facebook's growth. There have been times when Facebook was not only ignoring or negligent of the issues that its algorithms might be causing, but also purposely undermining some of the efforts to try and fix it because of this tension with the company's growth.

She was glad to see policymakers cite her article at a recent congressional hearing, and hopes Congress has the political will to regulate companies like Facebook. She says it’s also important for every new generation of Facebook employees to become educated about these issues so they will hold the company accountable. She thinks AI research has shifted a bit over the last five years to be more focused on taking responsibility for societal impacts and part of that evolution is being driven by people on the inside who raised awareness and advocated for change. 

One of Karen’s inspirations to go into journalism was Rachel Carson's book Silent Spring that sparked a widespread environmental movement. She strives to write stories that activate that same level of change, transforming both the cultural discussions and policy around important issues.

RELATED LINKS
Connect with Karen Hao on LinkedIN and Twitter
Find out more about Karen on her website
Find out more about MIT Technology Review
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

The multi-talented Cecilia Aragon is a data scientist, professor, author and champion aerobatic pilot. In this podcast, she explains how learning to fly gave her the confidence to pursue her career in human-centered data science and as an author.

Her book, Flying Free: My Victory Over Fear to Become the First Latina Pilot on the US Aerobatic Team, is the story of how a timid daughter of immigrants who had terrible phobias overcame her fears to become a champion pilot. Learning to fly and excelling at it helped her overcome emotional barriers from childhood when she was fearful and doubted her abilities.

In her mid-20s, she was pursuing her PhD and felt a lack of confidence, so dropped out of the program. “It wasn't that I had failed life, but I was living a very narrow life. I was just saying no to everything that might be exciting or interesting. And I saw my life stretching in front of me as incredibly narrow. A colleague offered me a ride in a small airplane…It suddenly occurred to me that living life too safely was dangerous for my spirit.” So, she took her first ride in the small plane and knew that she was going to learn how to fly.

She was very fearful as she was learning to fly and realized it was the same feeling she had in grad school when she didn't think she knew enough to be there. She practiced a discipline of constant learning, trying, making mistakes, relearning, and trying again until you get it right. She builds in safety protocols to anticipate potential problems, and most of all, never gives up.

She applies these same techniques in data science. As director of the Human-Centered Data Science Lab, Cecilia explains that every algorithm you write has potential human impact. A small error can be magnified and can have dramatic effects for thousands or millions of people.

She has co-authored a book called Human Centered Data Science: An Introduction to help experienced and new data scientists learn how to plan for and manage the unintended consequences from the automated collection, analysis, and distribution of very large data sets. There are human decisions at every stage of the work of data science, and we discuss how bias and inequality may result from these choices and what to do to help prevent this. She says we need to put human needs and ethics at the center of data science and place data in its social context.

RELATED LINKS
Connect with Cecilia Aragon on LinkedIN and Twitter
Find out more about Cecilia on her University of Washington profile page
Find out more about Cecilia's book
Learn more about the Human-Centered Data Science Lab at the University of Washington
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

Kristian’s interest in statistics and algorithmic fairness has taken her on a winding career path from academia to business, to public service, and back to academia. As she has made different career changes, she didn’t decide between academia vs. industry vs. non-profit, it was more about the problem she was interested in working on at the moment, and what else is happening in her life. 

After she earned her PhD in Statistical Science from Duke University, she worked as a research professor at Virginia Tech where she did microsimulation and agent-based modelingin a simulation lab. After that, she tried a data visualization and analytics startup called DataPad that was quickly acquired. When she was thinking about her next step in her career, she wanted to do something with social impact.

She was fascinated by the work of the Human Rights Data Analysis Group (HRDAG) that was applying statistical models to casualty data to estimate the number of undocumented conflict casualties. She spent a summer working for HRDAG in Colombia and then decided to join the organization full time. She spent five years as HRDAG’s lead statistician leading the group’s project on criminal justice in the United States focused on algorithmic fairness and predictive policing. Predictive policing uses algorithms to help the police decide where to deploy their resources based on crime statistics, so if you look at where crimes are most likely to occur, this is where you police more often. Kristian’s work showed that these algorithms could actually perpetuate historical over-policing and racial bias in minority communities. 

Early this year, she moved from HRDAG back to academia. She started her new position at the University of Pennsylvania in the Computer and Information Science Department on March 2 and a week later Penn closed down for COVID. Over this year, she has learned that she needs to adjust her expectations for herself, and not be so frustrated when she can't get things done that maybe under normal circumstances she could. It's not just working from home with her daughter nearby, it's the stress of everything that's going on, the additional mental fatigue of having to do all these risks calculations. This year has also made her appreciate the increasingly critical role of data science in driving data-driven decision making.

RELATED LINKS
Connect with Kristian Lum on LinkedIN and Twitter
Learn more about Penn Engineering
Learn more about HRDAG
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

A data scientist by training, Lillian brings a passion for human-centric machine learning and algorithmic effects to her role leading Spotify’s Personalized Home team. She also talks about the experience of leading her team from home during the pandemic.

Lillian’s team of data scientists and user researchers collaborate with a mixed-methods research approach to try to discern user needs and create the best matches between listeners and creators. They observe and analyze user behaviors to develop product insights. Her team’s user researchers develop qualitative insights about user needs and then work with the data scientists to figure out the scale of this need or behavior. “What are users telling us they actually need in that moment by doing certain actions, by looking and exploring at certain recommended items?” 

She sees that Spotify can have a big impact on inspiring new experiences between listeners and the creators of music and podcasts. By sharing music and podcast stories with the world, Spotify often making a match between a listener and a creator for the first time. It could be the first time someone tried a podcast. “These aren't just pixels and files that you're moving around, you're really creating creative experiences for people, and you're sharing somebody's creative output with people. There's a cultural consequence to the things that we do.” 

Spotify’s recommendation systems work partly by design and part of it is random. She says it’s the most exciting area of research for her right now because it's stretching the limits of what a recommendation system can do. She explains that the kinds of recommendations you get from friends are contextualized and teaching a computer to do that is very difficult. It's even harder to measure if you're doing a good job, because you have to understand how to interpret the reactions that people are giving. 

Lillian also cares deeply about being inclusive and offering opportunities for all types of people to be included in the tech conversation. As a female mathematician and proud Puerto Rican, she has seen how the current tech culture can be difficult for some to adapt to. She believes that instead of trying to adapt, it’s more important for each person to be authentic and that the culture itself needs to adapt to welcome all voices and their technical talents. 

While it’s been challenging leading her new team completely over Zoom, she says she is being very intentional about what kind of leader she is and what kind of visible working mom she is. She thinks it’s harder for more junior colleagues because they don't have the opportunities for workplace mentorship. Overall, she thinks this shift to remote work will be empowering and an opportunity to give people more flexibility about where they can live and work.

RELATED LINKS
Connect with Lillian Carrasquillo on LinkedIn
Connect with Lillian Carrasquillo on Twitter (@lillian)
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

After earning her PhD in Aerospace Engineering at Delft, Femke spent several years in oceanography, climate research, and subsurface modeling. She developed an expertise in data assimilation that she's now applying to improve COVID-19 pandemic forecasting models. 

Femke explains that data assimilation originated in weather forecasting, where a model is updated with the current day’s weather observations to provide a more accurate forecast for the next day. Data assimilation tools tune the model to provide a more accurate forecast. This concept can be applied in many areas including financial markets, the oil industry, and for COVID-19 research.

To help improve COVID-19 forecasting, she is using a compartmental model where there are compartments for different groups: those susceptible to COVID-19, those exposed to it, those infected, those who recovered, those in quarantine, and those who are deceased. The model is like a set of boxes, and the transition from one box to the other is governed by an ordinary differential equation. Then in those equations, you have parameters, which are typically not well-known. 

The data assimilation approach is to work more from the “outside in” instead of from the “inside out”. So, if you know the number of people that have died since the start of COVID, then according to this data, you can determine what the parameters would have looked like three weeks ago. With this type of inverse modeling, you can actually tune the parameters in that compartment model, and find the most likely reproduction number or the likely number of infected in the first place. The approach of having these simple relationships between the different compartments is a good framework for a very complex process. However, you cannot expect the data to tell you the story if you don’t have any prior domain knowledge. In order to take their research to the next level, it will be critical for Femke and her colleagues to collaborate with the medical experts that built the models who know how to express certain relationships.

As she has transitioned from one field to another in her career, Femke has needed to learn how to apply her expertise to entirely different research areas. She says one of the most important skills she has developed is to ask a lot of questions and not worry about being wrong and she advises young researchers to do the same. Sometimes those questions can help people already in the field think differently, and lead to new insights. 

Femke’s experience as an endurance athlete has also taught her valuable lessons for her work as a scientist. “People who excel in sports lose more races than they win. You have to make mistakes and fail, that’s the way you actually grow.” It also teaches you perseverance, to hang in there when it gets tough, and be happy with small increments of your own progress rather than always comparing yourself to your competitors.

RELATED LINKS

Connect with Femke Vossepoel on LinkedIn
Find out more about Femke on her TU Delft profile
Read more about TU Delft Civil Engineering and Geosciences
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

With the onset of the COVID-19 pandemic, Francesca Dominici and Rachel Nethery saw a way to connect the research they were doing on air pollution and health with the pandemic. They are studying the effects of air pollution exposure on different causes of hospitalization to see if pollution could increase a person’s vulnerability to COVID-19. While the research is at a preliminary stage, there is a lot of information that points towards the possibility that long-term exposure to air pollution could increase the mortality risk for COVID-19. 

Since this research is so important and timely, they have sometimes felt under pressure from the media and from high level government officials (from both parties) to answer questions with certainty about how many lives would be saved by reducing air pollution. They say it can be stressful to talk with certainty about their preliminary findings because the research is ongoing. But as a scientist, you should do your work with the utmost rigor and be able to communicate the uncertainties and the possibility that there's bias in the results, knowing this is still valuable scientific information and should be considered within a larger body of evidence to make policy.

Francesca feels optimistic that this research is going to make a tremendous positive contribution and the most important thing is to do work that matters and has an impact. She thinks that the field of public health is getting much more attention and respect. “This is the first time we’ve had the opportunity to speak with high-ranking senators about the importance of issues regarding climate change, health disparity, data science and air pollution. There is a silver lining in recognizing the value of data and the value of science in a way that has never been done before,” she says.

Rachel agrees. “I feel privileged to be starting my career at a time when suddenly everybody in the US knows what an epidemiologist is and appreciates the value of health data science. Policy makers and the media are very interested at this moment in what data can tell us about our future with COVID. I hope that moving forward this might ignite curiosity about other ways that we could integrate health data science into our policymaking and our lives,” says Rachel. “One of the most important things that government can be doing right now is to collect data in a systematic and reliable way about health and public safety.”

When asked what she would put in an executive order in the area of health, Francesca says she would first lower the level of fine particulate matter allowed in the National Ambient Air Quality Standard, especially during the pandemic. Second, she would ensure we are not just cleaning the air for the wealthy as poorer minorities are most affected by air pollution. And finally, she would put accomplished women scientists at the head of all our health agencies: NIH, CDC, FDA, and EPA, because they understand the issues and would get things done.

RELATED LINKS

Connect with Francesca Dominici on Twitter (@francescadomin8) and LinkedIn
Find out more about Francesca on her Harvard Profile
Connect with Rachel Nethery at Harvard
Read more about Harvard T.H. Chan School of Public Health
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile

Manisha Desai is a professor of medicine (research) and of biomedical data science, and director of the Quantitative Sciences Unit at Stanford University. She is an expert in the design and analysis of clinical trials and epidemiologic studies across multiple diseases, including COVID-19. In this podcast, she provides some insights into the challenges and progress of COVID-19 clinical trials.

When COVID happened, Manisha knew her team’s expertise in clinical trials would allow them to get up to speed quickly to study this new disease. So, she augmented her team with additional data science experts to focus on trials to identify safe and effective drugs to help prevent hospitalization and disease progression.“This whole period has been so anxiety-inducing for so many of us, it's actually been nice to focus that energy and anxiety towards work that could be part of the solution to some of the crisis,” she says. In studies so far, they have learned that there's a real disparity in outcomes by ethnicity and race and they are trying to determine if that’s due to genetic, cultural, or environmental factors. They are also launching trials to figure out questions about immunity, seroprevalence and incidents. 

Clinical trials are the gold standard for identifying safe and effective therapies. Since these trials require a lot of time and rigor, the findings have high internal validity. In a typical clinical trial, there is a sequential plan which starts by designing the study, setting up a data and safety monitoring board, database capture, and analysis plan. It's a totally different in a pandemic when you want to launch all of these events quickly, so developing the analysis plan and designing the study don't happen sequentially. “You're talking to the FDA and the IRB at the same time that you're developing the protocol, just as the ideas are sort of coming together. Everyone needs the study launched, we're all sheltering in place and solutions need to be found, so everyone is rushing to do this,” she says.

She cautions that we need to be careful. “These protocols are in place for the safety of the patients, and the integrity of the data and the study. The worst thing would be spending all these resources, having patients contribute to the trial, and then at the end of the day having inconclusive evidence,” she says. People are anxious about finding answers, they want to publish and disseminate their findings and be the first to solve the problem. And as data scientists, it’s critical to ensure the rigor, that the study is designed well, that the limitations are understood and that there's an appropriate interpretation of the findings. All of the misinformation that has been going on during this pandemic can contribute to further loss of lives.

Collaboration on COVID-19 research across groups is essential because we need to pool data. When you combine data across institutions and countries, we need a level of certainty about who is collecting what, how often, on whom, under which policy. This will really impact how and if you even can pool the data, and how you can interpret it. She says one of the most important foundations to the success of the research is ensuring that the data has integrity. People spend so much time spent thinking about the models but we need to get the data management process right before applying the models. 

There is not currently one centralized, robust data repository in the US that makes it easy for researchers to study questions when they arise. Manisha says people are trying to get these registries up and going quickly, and it's not easy because we ascertain data differently across different sites and countries. This is a very worthwhile investment to put time into thinking about how to collect data in standardized forms. 

Manisha has seen an enormous response to solving the COVID crisis. She is optimistic that we will identify safe and effective treatments and a vaccine on the horizon. “We’ve learned that the FDA and regulatory agencies can be quite responsive. And maybe even in a non-COVID world, we can learn about how to gain efficiency and have some work arounds when necessary, and get rid of things that might be unnecessary. There will be many lessons learned, not just for COVID, but there will be other pandemics, and we will be able to learn from this experience.”

RELATED LINKS

Connect with Manisha Desai on LinkedIn
Find out more about Manisha on her Stanford Profile
Read more about Stanford School of Medicine
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile
Find out more about Margot on her personal website

Newsha Ajami is a hydrologist specializing in sustainable water resource management, water policy, the water-energy-food nexus, and urban water strategy.

When she was studying hydrology in grad school, she took a water policy class that changed the trajectory of her career. “I would say that was one of the most important events in my professional career. I realized that laws and policies are what change the way we manage resources,” she says. All the data optimization and modeling means nothing unless you can understand the policy layer imposed on how our natural systems operate. 

This interdisciplinary approach guides her research at Stanford’s Urban Water Policy group where she brings together expertise in hydrology, data science, engineering, public policy, human behavior and economics to improve urban water systems. 

Newsha explains that we’ve spent a lot of time focusing on building more capacity to meet increasing demand for water because our 20th century approach to water resource management has been very one-dimensional and engineering focused. She has learned how it's better to work with nature to get access to clean water, rather than constantly trying to engineer our way out of our natural limitations.

As we build future cities and communities, she says we need to be mindful to not impose our outdated thinking. She says we don’t need to build infrastructure like dams or centralized systems that disconnect people from their water resources. Instead, build decentralized systems, green infrastructure and capture and recycle water as much as possible.

Newsha’s research helping utilities use data science to improve demand management by increasing the understanding of how and why customers change the way they use water. She recommends that instead of building for more demand that instead we focus on changing mindsets to increase efficiency with the water resources we have.

Her research leverages data science across disciplines to understand how water demand patterns are changing over time and then communicates this effectively to decision makers. She wants them to see how we can build cities or communities that are data centric and connect water systems, energy systems and transportation systems to work together more sustainably in the future.

RELATED LINKS

Connect with Newsha Ajami on  LinkedIn
Read more about Stanford University's Water in the West
Read more about Stanford Woods
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile
Find out more about Margot on her personal website

When Andrea was studying math as an undergrad, she was required to take an arts class in order to graduate, and soon discovered that she loved poetry. She learned that the process of writing a poem was often similar to solving a complex math problem—just starting with one part, and then doing one more, and gradually the rest is revealed. She enjoyed it so much that her first machine learning project in graduate school was on poetry/sonnet generation. 

Andrea wanted to blend technology and art in her career and Getty Images turned out to be the perfect place to combine her two interests. Getty Images curates and manages a huge library of images and videos that are used in editorial news, websites, social media, billboards and more. She started as a data scientist two years ago, and is now leading the AI/machine learning team to develop new tools to help clients more effectively use Getty’s creative assets. She explains that many of their creative clients come to the site and don't have the language to describe the types of images they're looking for. Andrea’s team is building models to help clients find images when they don't have the words to articulate their vision. Another machine learning project they are working on is how to identify awkwardly posed “stocky” photos vs. the more realistic photos that clients are looking for.

She is currently doing more painting than poetry, and also sees parallels between painting and computer vision. She explains that when you start with a blank canvas, you have to think about shapes, lines, negative space and colors. It’s a similar process to how a machine comprehends pixels and the relationships between colors, contrast, shapes and textures.

Andrea created a special photo exhibit for the WiDS Stanford 2020 Conference that illuminates concerns about image manipulation while also posing provocative questions about gender and leadership. In the exhibit, she used machine learning (style GAN) to transform pictures of US presidents, ranging from George Washington to Donald Trump, into female versions of those presidents. The style GAN is a machine-learning model that can manipulate an image in different dimensions, in this case, from masculine to feminine.

The project was born out of conversations around Generative Adversarial Networks (GANs), synthetic image generation and concerns about the implications of deep fakes in politics and our culture. She wanted a way to expose that concern in a humorous way. She also saw this as an opportunity to re-imagine our history. What would the world be like today if our presidents had all been women? 

She says the first response to the exhibit is usually laughter, but then it also sparks questions like: What would it have been like if females had founded the country? What wars would have happened or not happened? What would our constitution be like? How would capitalism have evolved? It catalyzes a conversation about the qualities of great leaders what leadership means through a female lens. 

RELATED LINKS

Connect with Andrea Gagliano on  LinkedIn
Read more about Getty Images
Connect with Margot Gerritsen on Twitter (@margootjeg) and LinkedIn
Find out more about Margot on her Stanford Profile
Find out more about Margot on her personal website