XAI - The Genesis | #herAIstory Session Recap

Session Theme

Second session of herAIstory -XAI-The Genesis

Can a self driven car ever make a moral decision like humans do?

Can we trust a robot judging our legal case and making fair & ethical decision?

Why can’t we manipulate a model to respond the way we want it to?

Have you wondered what fairness for an algorithm should mean?

Would you rely on a decision that an algorithm makes and be okay if it can’t be explained?

Do questions like these keep you awake all night? Just like you, even we, at RoundSqr, are very passionate about AI and its explainability.

Then, lock your date on 8th November, 2019! Let us gather and together explore the paper on XAI – The Mythos of the Model Interpretability

Distinguished Speakers

Arya Vedabrata
Head – Cyber Security Data Science
Sriram Venkat
Lead – Analytics
Malini Vyakaranam
Associate Data Scientist
Srinivas Atreya
Chief Data Scientist

Session Highlights

Last Friday evening, the atmosphere in RoundSqr’s brand new office was buzzing with excitement. A group of accomplished AI / ML practitioners and enthusiasts had gathered to discuss the hottest topic in their field – Explainability (XAI). They were keen to explore answers to questions being asked around the world, through the analysis of real-life case studies.

In the second session of RoundSqr’s #herAIstory, we dissected a seminal paper – The Mythos of the Model Interpretability. On the whole, it turned out to be a power-packed informative session, to say the least.

In fact, we hosted over 50+ participants, all the way from practicing data scientists, enthusiasts, evangelists, and students, to renowned cardiologists beginning to explore AI.

Speaker at herAIstory session - XAI - Arya
Arya (Head of Cyber Security Data Science – Qualcomm)

kicked off the proceedings. He walked us through some of the innovative (albeit manual) ways in which he had to explain many of his models (around predicting malicious URLs, reporting incident severity, etc.) in order to get approval. In his opinion, three aspects drive the need for explainability – impact, strategy, and stakeholders.

Sriram Venkat (Lead Analytics – Google Consumer Hardware)

described some of the tenets of the model interpretability (like trust, causality, transferability, informativeness, fairness, and ethics). Subsequently, he emphasised the importance of transparency and post-hoc explainability of models (especially in healthcare, while making predictions around tumours and other high-risk ailments).

Speaker at herAIstory session 2 - XAI - Sriram Venkat
Speaker at herAIstory session 2 - XAI - Malini Vyakaranam, RoundSqr
Malini (our In-house Data Scientist)

bust a few myths on XAI. Then she explained why linear models and decision trees are not interpretable as assumed by many. Furthermore, she explained why it is essential for humans to be unbiased themselves to be able to build ethical and fair models.

The session concluded with some very interesting questions on XAI to the panel (moderated by our Chief Data Scientist, Srinivas Atreya). Below are some of the questions that stood out:

  • Isn’t the model only as good as the data that it is trained on?
  • If the data is generated by humans, who are inherently biased, then will the models not be biased?
  • Explainability vs Accuracy – Should we sacrifice one for the other?
  • Will diplomacy be introduced into models?
  • What are the sensitive features? and lastly,
  • Why bias is good?
Speaker at herAIstory session 2 - XAI - Srinivas Atreya, RoundSqr
Audience at herAIstory Session 2 - XAI
And the day came to an end with an engaging networking session and yummy food????

A BIG THANK YOU to all the speakers and guests. We are overwhelmed by the number of people that came, the selfless manner in which knowledge was shared, and how engaged everyone was.

Also, we are extremely humbled by the thought that we can make a difference to this world. However small that might be. Truly, such support motivates us and makes our resolve stronger to invest more time in this event.

Do watch out for another edition in January…

Don’t want to miss the upcoming session? Then register now to receive a session update.

Want to contribute to herAIstory as speaker / volunteer? Write to us. Contact Lalitha Chamarty | lalitha.chamarty@roundsqr.com

#herAIstory – let us change the Industry, one woman at a time 