ProgressReport 30 August 2024

Progress this fortnight has mainly been on getting new synthetic datasets for HH->4b. Most of the work is complete and the results are quite promising. The obvious next steps are to make 15 synthetic datasets (from different random seeds) and repeat the closure exercise on them. This will require getting a production-line for the creation and processing of datasets with different random seeds and should be a nice exercise of the new classifier code

AE has made nice progress on getting the low Pt analysis ready. I would expect first results on this front soon.

I have rescoped Sindhu to X->HH->4b. This should be a straightforward reuse of the ZZ/ZH->4b analysis this for Run2. I'd like her to get a serious first pass of this analysis an present in B2G. Once she has started we will see how quickly progress is made. Could then potentially move to less-charted 4b territory (X->YH / semi-supervised)

Chuyuan has promising results on fitting SvB with different lambda samples. Need to understand what he is working on now and what his next steps are.

A lot of nice progress in making 2 modules at a time. Need to better understand the quality of these modules, esp IV behavior, and determine our yields. Andrew has made some progress in understanding the IV curves. Seem to me that the modules are OK and the software used to make the measurements is buggy. Eva left. I think its critical that we replace her ASAP and potentially ramp up more. We will need to demonstrate throughput soon, Nov 4 modules per day, April 8/day, May 12/day. We should do a bottom-up estimate of personnel needs for 12/day.

Progress here is still slow. Aniket is close, but is missing a few key steps. I think we have the ttbar and HH reconstruction going, however some plots still look buggy to me. I expect progress here soon. Once we have well-understood Xtt and XHH variables, we should be in a position to argue for the low-pt analysis.

Soheun has been very productive and we a couple of promising ides: importance of latent space / smearing in context of imperfect background model / clustering. Seems like we are not getting closure without a signal, this is being followed up on now. The other big questions are 1) how will this compare to the other data-driven anomaly detection schemes 2) how can we interpret a null result. (eg: how to interpret the smearing)

Out of touch here. Need to chat with Wes about big picture next steps.