Revisions: Figure 1 -- New Figure 2

[jaybee84]

Please take a look at this figure from the point of view of an audience less familiar with ML. What are some of the points that are confusing ?

[dvenprasad]

Based on the review this is what I interpreted. The way we represent the samples and datasets is not uniform. In this we have used shapes to represent datasets but everywhere else we have used the shaped to represent samples. So we need to reiterate and find a different way to represent datasets, i.e., not a shape.

[jaclyn-taroni]

Per @jaybee84 - turn this into a representation learning figure, which is covered in the first section on dimensionality reduction and in the section on multi-task learning, etc.

[jaclyn-taroni]

In terms of turning this into a representation learning figure - what if we started by getting rid of panels a and d and then simplifying / stripping away some of the text? I'd say in that context we can probably remove the shape as a dataset concept from this figure. I do think b and c are a good start wrt the points we're trying to make.

[jaclyn-taroni]

Additional thoughts I had today:

• Add a panel between the current b and c that takes two original features that do not show an obvious pattern in the heatmap and make a scatter plot (that also doesn't have an obvious pattern)
• Then between new scatter plot ☝🏻 and the current scatter plot we could have arrows with graphics that represent different approaches (e.g., matrix factorization, neural network, etc.)

[dvenprasad]

Adding current figure for reference

[dvenprasad]

Updated figure based on feedback:

[dvenprasad]

New figure:

[jaybee84]

@dvenprasad can you please add in the updated figure here?

[dvenprasad]