Changes to the manuscript

cbg-ethz / bmi

Mutual information estimators and benchmark

https://cbg-ethz.github.io/bmi/

MIT License

33 stars 5 forks source link

Closed pawel-czyz closed 4 months ago

pawel-czyz commented 6 months ago

[x] Use the NeurIPS template.
[x] Cite the variational estimators literature (Poole et al., Song and Ermon, McAllester and Stratos)
[x] Benchmark with tasks created using the BMM models to have a table in the appendix.
[x] Make Section 4 a Subsection 3.4.
[x] Revisit the introduction.
[x] Revisit the discussion and link to the appendix for more results.
[x] Answer NeurIPS checklist questions.
[x] Update the arXiv version.

a-marx commented 6 months ago

From the extended review that we got, a few additional points might be good to address:

[x] Double check the reference the reviewer provided for Prop 10 (Haussler & Opper, 1997, Kolchinsky & Tracey, 2017)
[x] Can we provide more experiments with the Bayesian estimator? Perhaps run it on the benchmark with BMM models---just to showcase its limitations and advantages?
[x] Clarify the takeaway from Figure 4

pawel-czyz commented 4 months ago

Great work @grfrederic and @a-marx! The new version will appear on arXiv tomorrow morning, so I'm closing this issue :slightly_smiling_face: