The Fifth Elephant 2019

The eighth edition of India's best data conference

Participate

What's Machine Learning Bias?

Submitted by \AbdulMajedRaja (@amrrs) on Monday, 15 April 2019


Preview video

Session type: Full talk of 40 mins

Abstract

We have been constantly told this statement “Computers don’t lie”. Yes in fact Computers don’t lie, but neither does it speak the truth. A computer does what its Master programs it to do. Similarly, A model wouldn’t lie unless the Machine Learning Engineer doesn’t want it to lie.

Outline

The first step of finding solution to any problem is accepting The Problem exists. Let’s accept that fact and see how to use Kaggle Survey results and help the community tackle Machine Bias.

Speaker bio

Links

Slides

https://github.com/amrrs/under_dev/blob/master/Machine_Learning_Bias(3).pdf

Preview video

https://youtu.be/SdbyH-Z98vA

Comments

  • Anwesha Sarkar (@anweshaalt) Reviewer 2 months ago (edited 2 months ago)

    Thank you for your submission. Submit your preview video by 23rd March(latest). It helps us to provide a fair evaluation to the proposal and close the review process.

  • \AbdulMajedRaja (@amrrs) Proposer 2 months ago

    Hey @anweshaalt, Is it mandatory to have a video? And does the video have to be relatd to this or any other video that I’ve given talk about would do?

    • Zainab Bawa (@zainabbawa) Reviewer a month ago

      The video has to be two-minute about the talk you have proposed. Check https://hasgeek.tv for samples of preview videos.

  • Zainab Bawa (@zainabbawa) Reviewer a month ago

    AbdulMajedRaja, help us understand what is the goal of this presentation? Who is the target audience for this talk? What are the key takeaways for the audience from this proposed talk?

    • \AbdulMajedRaja (@amrrs) Proposer a month ago

      ML Bias is one the areas that companies are these days trying to tackle. Starting from People analytics to Government Projects, ML Bias can completely skew your outcome. So this talk is aiming to introduce this topic to the user, what implications it had so far, how companies like Google are trying to solve and also how Data scientists are seeing it based on the Kaggle Dev survey - https://www.kaggle.com/nulldata/ml-bias-iml-perspective-recommendation

      Target Audience: ML Practioners and Executives (like Managers) who take a call on implementing the ML Solution

      Key Takeways: Getting to know about ML Bias and its forms. How to skeptically approach the ML solution. Importance of IML (Interpretable Machine Learning) in understanding it.

      • Zainab Bawa (@zainabbawa) Reviewer a month ago

        Thank you for the clarification. We need to see the draft slides, and a two-min preview video making an elevator pitch for the talk. Without this, we cannot make a final decision on your proposal. This has to be done by or before 10 May to put your proposal under further reconsideration.

        • \AbdulMajedRaja (@amrrs) Proposer a month ago

          Added Video!

  • Zainab Bawa (@zainabbawa) Reviewer a month ago

    This proposal will be interesting if you can address the following questions:

    1. What is a defensible definition for “unfair bias”?
    2. How does one evaluate a model for unfair bias? Are there generic methods? Or can this only be done on a case-by-case basis? Explain with case studies and examples.
    3. What is a “Bias-free / Fair Dataset”? How to identify one? How to build one?
    4. What is Interpretable Machine Learning?
    5. How do you recognize and minimize bias, at various stages in the development of models and production?

    If you can address the above questions, upload your slides here by 27 May, latest, to close the decision your proposal.

    • \AbdulMajedRaja (@amrrs) Proposer 29 days ago

      In fact, this just goes on and on as the topic is vast and there’s no standard. Should I just cut down to a 20 min talk that just outlines things?

  • \AbdulMajedRaja (@amrrs) Proposer a month ago

    Hi Zainab, I couldn’t finish the slides due to a bunch of reasons and I hope to upload it by Tomorrow.i hope you’d accept it.

  • \AbdulMajedRaja (@amrrs) Proposer 29 days ago

    Hi Zainab, Here’s the deck : https://github.com/amrrs/under_dev/blob/master/Machine%20Learning%20Bias.pdf Please consider this deck as v0.1 for evaluation as I couldn’t finish it completely. There’s good room for improvement given the time is available.

    • Zainab Bawa (@zainabbawa) Reviewer 27 days ago

      Thanks for the updated slides, Majeed. We’d rather prefer a case study where you have direct experience dealing with the ML bias problem rather than a presentation on theory and a summary of works out there which participants can otherwise directly access online.

      If you have an experience/case study to talk about, then update your slides accordingly.

      • \AbdulMajedRaja (@amrrs) Proposer 27 days ago

        Hi Zainab, I understand your point. I’ve personally worked on an instance with this Bias-case but that’s a project of course I can’t talk about as it’ll be on record. As you know, This is just an emerging talk in the entire industry. Even at NIPS, This is all about driving awareness rather than actual case-study as only researchers might have such. And, I understand that most of these content are available online but I believe that’s how most of the emerging talks have always been and that’s how industry start adopting it which I see as a key purpose these Conferences like Fifth Elephant play.

        But anyways, I respect your decision of not picking this up as such because it doesn’t have Industry case-study. Thanks for going through it.

        • Zainab Bawa (@zainabbawa) Reviewer 27 days ago

          Thanks for the response, Majeed. Since 2017, we hold a discussion every year on bias and ethics, and get the community to think through this problem. This year, we have consciously decided that for a talk, we want case studies so that participants have examples from real-life to understand how to approach the complex problems of detecting bias in models, and deal with bias as models evolve. Hence the decision.

          • \AbdulMajedRaja (@amrrs) Proposer 3 days ago

            Hi Zainab, Hope you are good. I assume this proposal hasn’t made it. If any of my other talks are qualified for Anthill, Please consider them. I haven’t submitted again there.

          • Zainab Bawa (@zainabbawa) Reviewer 27 days ago

            Btw, we had this talk on bias at 50p, a conference on payments: https://www.youtube.com/watch?v=Zn7oWIhFffs

            • \AbdulMajedRaja (@amrrs) Proposer 27 days ago

              That’s nice i’ll watch it. In fact, I didn’t know about Fifth El and other Hasgeek conferences before. Seems some of my proposals I could’ve cross-posted on Anthill too but I missed doing that. But will keep a track from now.

          • \AbdulMajedRaja (@amrrs) Proposer 27 days ago

            Thanks Zainab, Worst-case if you don’t have anything else better in Bias this year and want to consider this, I would try to fit in like an anonymous case-study, could be vague, but I’ll try to work out.

            • Zainab Bawa (@zainabbawa) Reviewer 27 days ago

              If you can work on the presentation as an anonymous case study, then we can evaluate and decide on a slot.

Login with Twitter or Google to leave a comment