How to Empower Women in Big Data — 6 VP and Directors in Data Science at Netflix Shared Their Experience with Machine Learning
When it comes to the topic of Women in STEM and specifically in technology, there have been challenges and obstacles. To enhance women’s skills and status in Data Science and Machine Learning, Women in Big Data, a non-profit organization cooperated with Netflix on how to bring more diversity in the data field. Let’s take a look on what is covered in the #SheRules with Machine Learning: From Infrastructure to Insights event on the afternoon of September 18th in Netflix HQ.
Keynote Speaker — Caitlin Smallwood
Do you know that each year, there are 44,000 teams, meaning 44,000 unique machine learning algorithms, trying to solve the same problem with the same data-set for winning the Netflix Prize? Caitlin Smallwood, the VP of Science and Analytics, also a mother of two teenage children said, the competition of Netflix Prize inspired more Machine Learning innovation internally and enabled a maturer recommender system of Netflix. As Netflix expands globally, new challenges and opportunities are created, so the same model for recommendation system that applies to the United States was changed for adapting different regions, such as Canada and Latin America. The use of Machine Learning has also been expanded to many other areas of Netflix’s Business, such as Content Catalog Planning & Optimization — e.g. catalogs in Food & Travel, Title Specific Investment Level — in order to find how sticky the contents are to the users, Studio Production or Streaming Optimizations. As Caitlin said, the current emphasis using Machine Learning is on further expanding and scaling its business.
Julie Pitt — ML from prototype to production with Metaflow
When it comes to Machine Learning and Data Science research, there is usually a gap between the academia research and the real product. However, as Julie, the Director of ML Infrastructure at Neflix said, Metaflow, removes the barriers of research and product by turning the code into a flow. What Machine Learning work flow needs is to firstly prepare the data, then train the model and then score the model. With Metaflow, you could turn your Python code into workflow nodes.
Michelle Ufford — Powering ML with notebook-based infrastructure
Nowadays, either its in Berkeley or MIT’s lecture room or in Google or Uber’s office, the engineers, analysts and scientists use Jupyter Notebook to process data. I remember, 2 years ago, when I was taking the first Data Science course, Applied Data Science with Venture Applications
IEOR 135, at UC Berkeley as an undergrad, our professor taught us step-by-step on how to load data frame in Jupyter Notebook and how to apply different Machine Learning packages on it. During the event, Michelle, Manager of Data Platform Architecture Core said, using Interact, Jupyter Notebook becomes even simpler. A cleaner UX with multiple buttons of choices of execution, who wouldn’t like it?
Sonia Bhaskar — Enhancing the streaming experience with ML and Metaflow
So how does the technology they just talked about relate to Netflix’s data science application? Sonia, a senior Data Scientist, told us about how Netflix is making the decision of the trade-off between the best possible video quality and the minimized bad events. Do users want a higher quality or a less interruption (low rebuffers) or less wait time (low play delay)? They use data science to figure out the balancing point in this triangle.
Sui Huang — Applying ML to non-linear phenomena
Aside from the linear programming, what else does Netflix care about for the users? Clearly, the joy, happiness and entertainments are important, so Netflix applies ML into social learning. For example, examining tweets topics flagging streaming quality issues, Netflix uses Natural Language Processing technique to identify problems from a social perspective.
Meghana Bhatt — Interpretable ML with human inputs as signals
Last but not least, Meghana, Director of Content Demand and Valuation Science, shared how Machine Learning are linked to human. Though called Artificial Intelligence, the content selections are informed by Machine Learning but led by humans. The outcome of content selection is the effort of both machine and human’s input.
Hope you found this event summary helpful. Good luck to all the data science enthusiasts out there! And thanks again to Netflix team for organizing the great event and inviting HasBrain team as one of the guests!
The HasBrain team has been going around interviewing 30+ data scientists from top tech companies in the Silicon Valley. I am planning to continue to share valuable insights from these conversations. Feel free to subscribe to HasBrain Platform to keep yourself updated. In the meanwhile, feel free to drop my an email if you have any questions. Stay tuned!
