Introduction to Explainable AI (ML Tech Talks)
Ғылым және технология
This talk introduces the field of Explainable AI, outlines a taxonomy of ML interpretability methods, walks through an implementation deepdive of Integrated Gradients, and concludes with discussion on picking attribution baselines and future research directions.
Chapters:
00:00 - Intro
2:31 - What is Explainable AI?
8:40 - Interpretable ML methods
14:52 - Deepdive: Integrated Gradients (IG)
39:13 - Picking baselines and future research directions
Resources:
Integrated gradients → goo.gle/2PxfRtq
Vertex AI → goo.gle/3ifu7S5
What-if-tool → goo.gle/3ehZWbZ
Catch more ML Tech Talks → goo.gle/ml-tech-talks
Subscribe to TensorFlow → goo.gle/TensorFlow
product: TensorFlow - General; re_ty: Publish;
Пікірлер: 25
This talk is out of this world! Thank you for the affort.
Thank you so much for the video. it's 2 years old but it's really gold and helpful. Again thank you so much for such great content
Wow! What an enjoyable and informative talk! Hope to see more on XRAI. Thank You!
Great to see this talk!
Amazing explanation, Thank you!!
Important subject and great presentation.
Thankyou for this, the explanations were really fluid, also the notebook is very helpful.
Interesting insights into Explainable AI. Thank you for making it available.
Thank you for an interesting talk! Hope you make dive deep explain IG in the next video further.
Good explanation thank you Sir !
Great presentation!
Good to know where the focus is on .. where improvements will come over in time
@xt3rm1nat0r8
Жыл бұрын
Totally agreed
Great talk, many thanks for putting it up. Seems basically you can only use it to go back and check your data inputs when things go wrong. You still have no idea why certain groups of pixels are being highlighted and others not. It would help to see some adversarial attack images to see how trustworthy the model is and what could be done to avoid those types of errors.
interesting. thank you
تحياتي الخالصة من الجزاءرthank you very mutch
تحياتي الخالصة شكرا جزيلا
Hello how did you get this type of role ml solutions engineer?
Just watching Doug's hands at the beginning 🙌
@marvinalbert
Жыл бұрын
Certainly his hands at the beginning would contain high attribution intensities for all kinds of interpretability tests of this great talk 😄
Is PDP model specific?
@wangjason8400
2 жыл бұрын
PDP is model-agnostic. That slide is wrong.
Appreciate the effort. I felt the presentation and explanation could have been better.
16:06
I still feel it cannot explain things well....