Ecco
AI
model interpretability
SHOW HN2021
How HN took it
10 positive, 5 neutral or asking, 2 mixed, 0 critical, out of 17 comments about the project. Thread score +0.53.
Idea: 5 for, 0 againstDocs: 3 for, 0 againstDesign: 1 for, 0 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Visualizations are clear and helpful for learning transformers
From 3 comments
Wow, love the NNMF visualization. Like all great visualizations, it does a very good job of showing and not telling me what's going on. More of this, please. One question: how…
yowlingcat on HNThank you, Jay! I appreciate the addition of the colab notebooks with code examples. I can't wait to play around with this and investigate how language models _speak_.
anfal_alatawi on HNThis is fantastic, I used your earlier transformers article to first get a real grasp on the architecture. I hope you expand this to accommodate other modes of attention outside…
Grimm1 on HN
NMF factorization of activations is a good approach
From 1 comment
NMF for factorizing activations is brilliant!
shenberg on HN
Useful for organizations adopting AI
From 2 comments
Helping people understand "what the ai is thinking" is really important when you are trying to get organizations to adopt the technology. Great work.
indymike on HNFantastic work. This is the kind of stuff we need to get these models actually adopted and integrated into non-tech organizations.
ZeroCool2u on HN
Objections
Factorization results require interpretation and may hide structure
From 1 comment
Interesting. The non-negative matrix factorization on the first level kinda highlight some semantic groupings : paragraph, verbs, auxiliaries, commas, pronouns, nominal…
GistNoesis on HN
Requests
Support other attention modes beyond transformers
From 1 comment
This is fantastic, I used your earlier transformers article to first get a real grasp on the architecture. I hope you expand this to accommodate other modes of attention outside…
Grimm1 on HN
Add other dimensionality reduction methods such as UMAP
From 1 comment
This work is awesome! Are there theoretical reason to choose NMF over other dimensionality reduction algorithms, e.g. UMAP? Is it easy to add other DR algorithms? I may submit a…
Der_Einzige on HN
Questions
Novel insights obtained with the method
From 1 comment
Any examples of novel insights obtained with this method?
blackbear_ on HN
How it relates to explainable AI
From 1 comment
Wow, love the NNMF visualization. Like all great visualizations, it does a very good job of showing and not telling me what's going on. More of this, please. One question: how…
yowlingcat on HN
Alternatives named
LIME and Shapley values
From 1 comment
These are AI explanation methods. They belong to the toolbox which would include LIME, Shapley values...etc. Input saliency is a gradient-based explanation method.
jalammar on HN