1 of 19

BinaryTransparency Toolkit Development

​

​

​

​

Mentee: Dwight Ross (Tougaloo College)

Mentor: Christopher Palmer (University of Maryland)

​

1

2 of 19

Overview

  • Abstract
  • Test Simulation
  • Variables
  • Data Distributions
  • Recommendations
  • Sobel Operators
  • Future Work
  • Acknowledgements

2

3 of 19

Abstract

​

​

Abstract: The BinaryTransperancy Toolkit intends to create a functional understanding of our machine learning outputs. The goal is to understand how our input variables are correlated to the output of the machine learning algorithm. The toolkit works with variables that are loosely directly correlated. However, there are some types of correlations that are not well covered with the toolkit and that is what we will work on. As a test case we will further the study of separating the ttbb and zhh decays.

​

3

4 of 19

Test Simulation

4

We have simulation for these two processes and they have already been analyzed in a preliminary version of the toolkit with a paper draft. These two final states have four b quarks and two opposite-signed, same-flavor leptons.

5 of 19

Variables

We did our simulation with nine variables to check and find recommendations.

​

5

Variables

M_ll

hh_dR

hz_dR

hh_dphi

hh_deta

hz_dphi

hz_deta

bb1_dR

bb2_dR

6 of 19

Quantiles

6

The black bars show the median and the solid red lines shows 1𝞂 from the black line which is 34% of data in between each each red line to black line. The red dashed region shows 99.8% of simulation.

Threshold on Transformed BDT

7 of 19

Quantiles

7

8 of 19

Distributions between Signal and Background

8

9 of 19

Goals

  • Improve cut suggestion algorithm to will remove background without reduction of sensitivity
  • Improve treatment when correlation is directly correlated

9

10 of 19

Recommendations from the BDT

10

11 of 19

Examples of cuts

11

12 of 19

Cuts as we keep increasing significance

12

13 of 19

Fitting Nonlinear Edges

  • Since most graphs create nonlinear shapes then we want to be able look through all the data that wouldn’t go through a rectangular and spherical cut.
  • The way we thought about doing this is by using Sobel Operators to find edges along our histograms.

13

14 of 19

Sobel Operators

  • The Sobel Filter is used for edge detection. This is done by calculating the gradient of image intensity at each pixel in each image.
  • The result shows how abruptly or smoothly the image changes at each pixel, and therefore how likely it is that that pixel represents an edge.
  • We use a 2d array from our 2d histograms as an image to perform convolutional operations on each image with a cutoff point to find edges.

14

15 of 19

Example of Sobel Operators

15

16 of 19

Edges Showcase

16

17 of 19

Future Works/Summary

  • Future Works
    • Parameterizing non-linear edges with our histograms by using polynomial regression corner by corner.
    • Use BinaryTransparency with other decays as well as flipping signal and background with our current simulation
  • Summary
    • Square and circular edges already working in the toolkit.
    • We worked on non-linear 2d shapes and edges.
    • I implemented Sorel imaging for finding edges without any prior modeling.

17

18 of 19

Resources

Edge Detection, www.cs.auckland.ac.nz/compsci373s1c/PatricesLectures/Edge%20detection-Sobel_2up.pdf. Accessed 30 July 2024.

​

18

19 of 19

Acknowledgements

University of Maryland College Park

Dr. Chris Palmer

PURSUE

"This work was supported by the DOE grant RENEW-HEP: U.S. CMS SPRINT - "A Scholars Program for Research INTernship" at Tougaloo College, MS (DE-SC0023681), Brown University (DE-SC0023651), RI, University of Puerto Rico-Mayaguez, PR (DE-SC0023680), and University of Wisconsin – Madison, WI (DE-SC0023643) and U.S. CMS Operations at the Large Hadron Collider (NSF-2121686)"

​

19