115 | Human-Driven Machine Learning with Saleema Amershi

Data Stories

[This podcast is fully supported by our listeners. If you enjoy listening to Data Stories, consider supporting us on Patreon!]

How should we interact with machine learning? Is there a way to better understand what machine learning systems do and how? What role can visualization play in machine learning? These are some of the questions we discuss with Saleema Amershi on the podcast this week.

Saleema is a Researcher at Microsoft Research AI (MSR AI) where she works on technologies for helping people build and use machine learning systems.

On the show we explore the need for interaction in machine learning and discuss some of the fascinating work Saleema has done in this space, as well as what visualization can do to make machine learning more transparent for its users.

Enjoy the show!

Links

• Saleema Amershi’s home page
• Saleema’s paper ‘Revolt: Collaborative Crowdsourcing for Labeling Machine Learning Datasets’
• Saleema’s paper on ModelTracker ‘ModelTracker: Redesigning Performance Analysis Tools for Machine Learning’

---

http://datastori.es/wp-content/uploads/2018/02/saleema.m4v

Chapters

1. Welcome to Data Stories! (00:00:27)

2. Introducing Saleema Amershi from Microsoft Research (00:00:53)

http://saleemaamershi.com/

3. What is Interactive Machine Learning? (00:02:47)

4. Saleema's research on the data that goes into Machine Learning (00:10:34)

https://www.microsoft.com/en-us/research/wp-content/uploads/2017/04/pn4864-changA.pdf

5. Saleema's project ModelTracker (00:19:04)

http://saleemaamershi.com/papers/amershi.CHI2015.ModelTracker.pdf

6. Reactions to Saleema's research (00:24:22)

7. The problem of transparency in Machine Learning (00:28:02)

8. What's gonna happen in the future in Machine Learning? (00:32:57)

9. Get in touch with us and support us on Patreon (00:36:47)

https://www.patreon.com/datastories