UW EPE Symposium

Thursday 1 Jun 2023, 13:00 → 15:00 US/Pacific

PAB 6th floor (WRF Data Science Studio)

Shih-Chieh Hsu (University of Washington Seattle (US))

In this Symposium, UW EPE High School Interns will run a joint poser session with  Physics 417 and PHYS600 students. This event will be held in eScience Institute at PAB Tower. Each student group will present a poster at the event. 

The symposium is relevant to the broad UW data science community. You will be asked to spread the word about the awesomeness of Deep Learning Neural Network and your projects by inviting your friends/professors/fellow undergraduate students to this event. 

Poster specifications: Your poster should be a PDF file of 36'' x 24" with white background (portrait or landscape). We will adopt this guideline [here] about poster specifications set by UW CSSCR. Make sure that the motivation, methods, and results are clearly and visually communicated to a broad audience. There are two popular poster printing services on-campus ($30/poster, 2 working days, plan ahead): 

• https://posters.hsa.washington.edu/
• https://finance.uw.edu/c2/printing-copying/printing-copying 

See UW Research poster templates here

 

During the poster session: 
Your group is expected to shortly explain the poster (<5 min) to interested attendees. Please note that the whole group is expected to attend per poster. Students are encouraged to go around the posters and learn about other projects, just like in a regular conference. 

We are very much looking forward to this fun event! Please don't hesitate to reach out to instructor team if you need assistance/feedback or any questions while developing your poster.

photos

UWHighSchoolIntern2023.png

13:00 → 13:20 ML4GW Classification of BBH & BNS

Gravitational waves are a fundamental prediction of Einstein’s theory of General Relativity, in which the fabric of spacetime (x, y, z, t) is altered due to masses. These waves are generated by accelerating massive objects, such as two merging black holes (BBH) or neutron stars (BNS). When the objects merge they create disturbances in spacetime that propagate as gravitational waves. Our objective is to develop and optimize a 4-layer ResNet model that can quickly and accurately identify BBH and BNS signals from typical detector noise.

Speakers: Arif Chu (University of Washington), Xinyi Liu

Arif & Xinyi's ML4GW Poster Presentation.pdf

13:20 → 13:40 Machine Learning for Top Tagging

We survey top jet classification perforamcne in a variety of deep learning models, Convolutional Neural Network, Particle Flow Network, Dense Neural Network and ParticleNet.

Speakers: Tristan Kay, Yuuki Sawanoi (University of Washington (US))

Machine Learning for Top Tagging.pdf

13:40 → 14:00 Semi-Visible Jet Classification with Boosted Decision Tree

Particle colliders record large amounts of data on jets produced from proton- proton collisions. Some jets are described by the Standard Model, but some jets contain Dark Matter that cannot be described. Goal: Achieve higher ratio of signal to background by discriminating Semi- Visible Jets from Standard Model jets using the Boosted Decision Tree.

Speakers: Hebu Patil, Hebu Patil

EPE SVJ Poster Hebu.pdf

14:00 → 14:20 Sparse Point-Voxel CNN

Collisions in a collider create multiple types of particles, each of which creates a shower (in the EM calorimeter, Bremsstrahlung). These showers are measured in calorimeters. We want to associate each shower particle with a tau lepton source (or other noise). The [ast approaches. Point-cloud Network, wasted runtime. We studied combination of Voxel-based neural network and Sparse Convolution to optimize information loss with point-based branch.

Speakers: Lydia Maximovna Mazeeva (University of Washington (US)), Lydia Mazeeva, William Yao Ze Lai Feng, Yuuki Sawanoi (University of Washington (US))

SPVCNN.pdf

14:20 → 14:40 Using cVAE for Anomaly Detection

Wr are looking for model-independent way to search for new physics. Since we don’t know which laws new physics will follow. We are Interested in over-density anomalies. We use a model that tries to learn probability density of certain features to look for anomalies from the expected distribution. We will use LHC Olympics 2020 dataset to study this approach.

Speakers: Vincent Chen, Vincent Chen

Vincent Chen Poster.pdf

14:40 → 15:00 Cleaning gravitational waves - Old and New Data Analysis

The first gravitation wave (GW150914) was detected in 2015 by LIGO & Virgo interferometers. The GW detectors measure strain and auxiliary data to output time-series waves description disturbances/distortions in space-time. However, it encounters high background noise, low gain GW signal from interferometer strain output. We studied new machine learning based method, DeepClean, for denoising. We will show effectiveness of this method in comparison to conventional matched filtering method.

Speakers: Ahad Ather, Gaurang Pendharkar (Interlake High School)

Cleaning Gravitational Waves - Ahad and Gaurang.jpg