version2

[timm]

while this paper is clearly perfect and wonderful in every way, uit is also somewhat rushed paper and there will probably, hopefully, be a better v2.

what goes in there?

[timm]

here, its more about problem types. we also have scripts that could be useful

[timm]

FSE NEIR is 4+1 pages, due Friday, June 15, 2018

[timm]

ICSE'18 is long gone

[timm]

ASE workshops are feb26 http://www.ase2018.com/?p=calls#workshops

[timm]

FSE workshops are March7. but maybe we can do something special with swan+promise there

https://2018.fseconference.org/track/fse-2018-Workshops

[timm]

stuff where one hand shakes the other

• https://github.com/ai-se/ResourcesDataDrivenSBSE/issues/7#issuecomment-360922519

[minkull]

It's a pity we're too late for ICSE's TBs, but it would definitely be good for us to target ICSE 2019. FSE NEIR sounds very interesting, and it would be nice to target it.

In terms of workshops, I'm having an idea here. What if there were two workshops -- one on MSR and another one on SBSE -- but in the same venue? This could really help for something special around DSE to be done, if the two workshops agree on collaborating, and specially if one or both the workshops are well established. It may be easier to get the two communities together in this way than if we had a single new workshop accepting both MSR and SBSE papers, or a single new workshop that accepts only DSE papers (this latter may attract too few submissions). Hey, SSBSE is co-located with FSE every two years. And this year it's with ASE. There may be some opportunities here.

[timm]

New dection3. After “what”. Argue that SBSE is the machine hiding underneath MSR anyway. It’s sly optimization. The Ric n rule paper

[timm]

Paper needs another section on algorithms .

[timm]

There are various research prototypes lying around we should be able to Make ouclbkic. Need code samples for using deep, jmetal, Sway etc

[timm]

Paper needs pseudo code idom

[timm]

Paper needs testing examples from evosuite

[timm]

All the case studies here are NC state. What can we get from uk? From Aus?

[timm]

Bibliography. The 25 papers u have to read

[minkull]

ML vs optimisation: From an algorithmic perspective, many of the ML algorithms are indeed optimisation algorithms. From the problem perspective, the main difference between ML and optimisation is (in my opinion) that ML is interested in generalisation, whereas "generalisation" is only considered to be important in optimisation when we talk about robust optimisation, i.e., a subfield of optimisation.

MSR vs SBSE: The objective functions that one comes up with in SBSE are anyway only "surrogates" of the actual (unknown) objective function to be optimised. So, similar to MSR (and ML) where we can't really compute the exact error on unseen data (the generalisation objective), we can't have the exact function to be optimised in SBSE either. This could lead to an argument that MSR and SBSE should be the same thing, both from the algorithmic and the problem perspectives. This further strengthen the argument that the two areas should be working together.

[markuswagnergithub]

What can we get from uk? From Aus? ==> 1) We are just bringing MSR-SBSE together just these days (paper + grant to be submitted). 2) Phones: happy to talk about this in the updated version, especially since we had not talked much about in-vivo optimisation involving hardware and noisy (b*tchy) functions. There are many dragons - and the challanges require different algorithms.

[markuswagnergithub]

"The Ric n rule paper": Ric?

[vivekaxl]

ROC ‘n’ Rule Learning—Towards a Better Understanding of Covering Algorithms

[minkull]

What can we get from uk? From Aus? ==>

1. PPSN 2017 and Information Sciences 2018 work on learning optimisers (hyperheuristics for the software project scheduling problem)
2. TOSEM'13 paper on MOEAs for SEE, even though I guess there are newer stuff that could be used as case studies now
3. I have some other work under review. Let's see how long it takes for us to hear the outcome