Do you want to build Africa's first machine learning model for particle classification? Here is an opportunity to engage with ATLAS researcher Sabrina Amrouche and a resident Zindi data scientist in a Q&A session to help you complete the TIC-HEAP Challenge. Don't miss out on this chance!
Please join us Thursday 16 Jan 19h00 CAT on Zoom (https://zoom.us/j/524505943) to hear about where the data comes from, see how our data scientists would approach the problem, and discuss the challenge with a physics researcher and your fellow data scientists. If you have any questions you would like us to address, please post them here before the event starts.
Looking forward to the session. There is a python starter code. Could you also provide one in R?
Good initiative, I am highly intrested, Thank you zindi community !
yes I would be very interested too. Also I second the request for starter pack in R
Highly interested and soon join the challenge
Looking forward attending
I look forward
looking forward to hearing from this interesting field and the highly knowledgeable staff
This is great. I look forward
Noted with thanks.
Could you explain what the image is actually showing? I am assuming that it is something like a bubble chamber trace with the X and Y on the image plane and the value of the pixel as the Z.
If this is correct what is your interpretation of detections in side by side pixels, would it be wrong to assume that the particle is somewhere between these pixels?
I read the Wikipedia article on these types of particles, I was hoping to find something like: some particles where negative charged and thus would tend to curve left, for example. However, I couldn't determine anything like this. Could you tell us what you would expect in an ideal situation for each particle? This would also be assisted by a description of how the labelled data was labelled.
Very excited to join the fun. Thanks
Hi ! Thanks for your interest.
The image is a 2D Histogram on the RZ components of a particle trajectory, R being X^2+Y^2. So the value of the pixels is the density (bins). The idea is to encode the track as an image and extract patterns relevant to the particle type.
So the particle passed through those pixels, leaving more hits in the brighter ones. There is in fact much more information available to characterize a particle but the idea here is to start simple and find patterns.
The labels were created using all the information from the detector and physics quantities (like the momentum) and this is how particle identification is done. Again the idea is to challenge machine learning into finding information in an earlier stage and simple or speed-up the process. The goal is then to model those missing correlations you mentioned :)