Explainable AI Cheat Sheet - Five Key Categories
Vložit
- čas přidán 5. 06. 2024
- Explainable AI is a set of tools and methods that allow us to understand Artificial Intelligence and Machine Learning models and their predictions. This video is an accessible and gentle introduction to Explainable AI methods. It is an introduction to the Explainable AI Cheat Sheet (ex.pegg.io).
Intro (0:00)
One Explainability Motivation (0:58)
Explainable AI Cheat Sheet (2:39)
1. Interpretable Models (3:25)
2. Model-Agnostic Methods (4:26)
3. Model-Specific Methods (6:40)
4. Example-Based Methods (7:52)
5. Neural Representations (9:35)
Conclusion & Resources (12:13)
-------------
The Explainable AI Cheat Sheet:
ex.pegg.io
Mirror: github.com/arpeggiohq/explain...
Resources:
Interpretable Machine Learning Book (Christoph Molnar): christophm.github.io/interpre...
Explainability for NLP (Isabelle Augenstein): • Practical Talk 1: Expl...
NLP Highlights: Interpreting NLP Model Predictions (Sameer Singh): / 117-interpreting-nlp-m...
Please Stop Doing "Explainable" ML (Cynthia Rudin): • Please Stop Doing "Exp...
I am a data scientist & really appreciate your work ! Keep good work going !
Really appreciate these resources! Thanks for always explaining things so clearly!
Thank you for what you have provided here! The links are providing tremendous added value to your videos.
Thank you so much for your contribution, i really appreciate it.
Thank you very much for the video.
9:57 yeah a dedicated video for each of the methods will be really great!
Very informative and easily understandable. Thank you for making this video
Now I think it would be great if Jay can make video on SHAP explanability and usage...hope you have time to accept this request.
+1
I'll also be looking forward to it.
I'd certainly love to learn more about it at some point
@@arp_ai will wait for your video for this topic
Very important topic covered in good details. Thank you
Best blog and guidance video for AI
Thank You for nice and easy information. I was looking for such information 👍
Very interesting ! Thank you for sharing
Explainable AI is not only based on neural networks. Everyone wants to make neural networks explainable which is not the case by design. You also have to consider other types of models like rule based models (expert systems) or even probabilistic models which are explainable by design.
Thanks man!
Found this channel via Linkedin post. It was a good find!
Thank you for ur video it's really well organized and easy to understand
great video. highly informative !
Thanks for the information. It was helpful for me :)
Nice video! But even better the resources which were referenced! Thank you...
Interesting content. I am looking forward to the paper reading videos
just fascinating
Thanks for this video (love the john coltrane )
Thanks Jay. Amazing video. I have a question though: why the pretraining in gpt-3 is called unsupervised learning?it works on labelled data, so I think it should be a supervised learning task.
It's better called self-supervised learning nowadays. It's unsupervised in the same way that word2vec is unsupervised -- it is not trained on an explicitly labeled dataset, but rather on on examples generated from free text.
Thanks for knowledge sharing. Where decision trees lie in cheat sheet?
Jay, I was exploring SHAP for explainable deep learning, and it didn´t work. The git mentioned an update in TensorFlow that conflicts with SHAP. What do you know about that?
Hello! Can you please make a video on interpretability of unsupervised ML models
Hi sir, Shap gives you the info same like feature importance results?
SHAP is a method of obtaining feature importance, yes.
Why logistics or linear regression are considered interpretable? If you have 1000x of features, how can you interpret a non-sparse weight matrix?
that John Coltrane cover in the back though!
The example based part was kind of shallow, you didn't talk about how they figure out the specific flaws in the model
That tends to be a different problem which can either arise from model but potentially also from the dataset. The problem becomes more model debugging. XAI is one debugging tool, but there are many others, especially deep examinations of the data.
Some day you should do a video about the background in your videos. Please.
Thanks a lot. It was a great introduction and really helped me🙏🫀