Machine Learning applications in the Dark Matter Search SuperCDMS experiment.

What on Earth is Dark Matter? (It's actually everywhere in the universe).

Various scientific evidence has established that the universe is composed of mysterious entities like
Dark Matter and Dark Energy along with normal matter i.e. baryonic matter which makes us up, the sun, the earth, the starts, and the galaxies. Astonishingly, the normal matter comprises only 5 % of the total mass-energy density of the universe! The remaining part is composed of Dark
Energy which accounts for 69 % and Dark Matter which accounts for 26 %. Dark
Matter is a hypothetically proposed form of invisible matter that makes up about
85 % of the total matter in the universe. The nature of Dark Matter and the way
it interacts is still unknown and thus remains one of the major unsolved problems
in physics. Dark Matter interactions are rare because they do not interact with
electromagnetic radiation and hence are difficult to detect by direct observations.
Detecting this mysterious form of matter and studying its properties can open new
frontiers in physics.

Detection of Dark Matter: It has been hypothesized that dark matter particles called WIMPs are passing through Earth, millions of them every square centimeter per second! (Don't worry they don't bite). WIMPs are Weakly interacting massive particles and as the name suggests, they are weakly interacting with normal matter.

SuperCDMS:  The SuperCDMS collaboration aims to detect WIMPs. The underground detector can detect the rare signal of WIMP when it collides with the atomic nucleus of the detector. The experiment aims to measure recoil energy from the elastic scattering of dark matter particles with the target nucleus in the crystal. So the dark matter particles and the neutrons create a signal which we call "nuclear recoils" in the detector. All the other stuff which hits our detector is the background and they produce "electron recoils". So to clearly observed the dark matter signal, we should filter out the electron recoils background from the nuclear recoils. This is traditionally achieved by selecting signal from the "good" part of the detector i.e. signal which lies within the bulk of the detector material. This process is called Fiducial Volume optimization. I have used machine learning classification to complement the Fiducialization process to filter out the background events. Find out more in the link below.