How to cite?

[DrAndiLowe]

Hi Luke, what's the appropriate way to cite your thesis? I was unable to find it in inspirehep or CERN CDS. Thanks!

[DrAndiLowe]

P.S. I can haz PDF? I can't LaTeX this. :-(

[DrAndiLowe]

OK, I found an old paper copy in my files. Just let me know how I can give credit for your work!

[lukedeo]

Figuring out where to put it! I'll let you know where I end up putting it :)

[DrAndiLowe]

Thanks! I dumped mine on arXiv and on the CERN document server.

I'm interested in your online jet reweighting method. I understand the rationale, but was wondering if anyone has done something similar elsewhere. If so, can you point me to papers?

Probably naive questions ahead... ahem...

If I understand the method correctly, in equation 3.43 it looks like you're dividing a 2D histogram corresponding to u,d,s flavour jets by a 2D histogram corresponding to the heavy flavour of interest, and the weight for a jet with that specific heavy flavour is the result of that division for the bin that matches the pT and eta of that jet. That is to say, the result of the division is a "reweighting" 2D histogram, and we just have to locate which bin a jet lives in to find the weight. Is that a correct interpretation of the method?

You're then using these weights as instance weights during training, right? Can I use these values as instance weights in other ML algorithms that support instance weights? I want to achieve the same result; to "flatten" the pT and eta distributions across jet flavour so that I don't introduce bias (that is dependent on the jet pT and eta) for a particular flavour. So can I treat these weights as regular case weights?

If my intuition about equation 3.43 is correct, what happens when you don't have a lot of data and there's quite a bit of variation between bins? Did you consider smoothing using a 2D KDE? Did you use any particular scheme to choose the binning? Freedman-Diaconis, something like that?

Thanks!