Spring/Summer 2024

Group meetings are for students and faculty at CMU and UPitt.
The STAMPS webinars are open to everyone.

These meetings are Fridays 1:30-2:30 PM. Meetings will be in-person in Baker Hall (BH) 232M (second floor of the StatDS Department in Baker Hall). If that plan is to change, we will send out an alert via the mailing list.

January 19 - No Meeting #

January 26 - Maggie Hansen (CMU Robotics Institute) #

NOTE: We will start this meeting exceptionally at 2pm.
Title: Hypothesis Mapping for Autonomous Scientific Exploration
Description: This talk will first cover previous work on hypothesis mapping for autonomous exploration by a mobile robot, or Bayesian updating of a spatial model paired with active sampling constrained to cost and distance budgets. Following an overview of this prior work, Maggie will propose updates and extensions that would incorporate physical process modeling into such a hypothesis mapping framework and enable learning about the relevance of each individual process model, along with spatial variations in these measures.

February 2 - James Carzon (CMU Stats&DS) #

Title: Extending Normalizing Flows to Propagate Uncertainty in Neural ODEs
Description: Predictions for continuous-time dynamical systems may be subject to several interacting sources of uncertainty. In a system where dynamical noise is assumed to be present, a predictive model must account for both the randomness of the initial state of an input and the sensitivity of the employed differential equation solver. We propose a prediction pipeline which efficiently quantifies predictive uncertainty for real dynamical data. The pipeline firstly entails training a neural ODE to learn the underlying dynamics of the generating system. Secondly, a flow-based model which extends the normalizing flow model is developed to learn a time series of densities which propagates the assumed initial state noise information under the learned dynamics. We demonstrate these methods with some numerical examples.

February 9 - Arthur Wu (Pitt Physics & Astronomy) #

Related paper: https://arxiv.org/pdf/2310.17696.pdf
NOTE: We will start this meeting exceptionally at 2pm.
Title: Quantum Entanglement and Bell Inequality Violation in Semi-Leptonic Top Decays
Description: Quantum entanglement is a fundamental property of quantum mechanics. Recently, studies have explored entanglement in the t ̄t system at the Large Hadron Collider (LHC) when both the top quark and anti-top quark decay leptonically. Entanglement is detected via correlations between the polarizations of the top and anti-top and these polarizations are measured through the angles of the decay products of the top and anti-top. In this work, we propose searching for evidence of quantum entanglement in the semi-leptonic decay channel where the final state includes one lepton, one neutrino, two b-flavor tagged jets, and two light jets from the W decay. We find that this channel is both easier to reconstruct and has a larger effective quantity of data than the fully leptonic channel. I will also talk about how we compared the performance of several algorithms (OSB, SVD, Iterative Bayesian) to unfold the data.

February 16, 1:30PM EST - Public Webinar : (Ashley Villar, Harvard)

February 23 - No Meeting #

March 1 - No Meeting #

March 8 - No Meeting (Spring Break) #

March 15 - No Meeting (PhD Open House) #

March 22, 1:30PM EDT - Public Webinar : (Matthias Katzfuss, University of Wisconsin, Madison)

March 29 - No Meeting #

April 5, 1:00PM EDT - Public Webinar : Wouter Verkerke (University of Amsterdam / Nikhef)

April 12 - John Alison (CMU Department of Physics) #

April 19, 1:30PM EDT - Public Webinar : (Mark Risser, LBNL)

April 26 - Konstantin Malanchev (CMU Physics, SNAD Team) #

Title: Active anomaly detection with isolation forests
Description: Konstantin will present the work he are doing in the SNAD team on anomaly detection tools for time-domain astronomical data. Their algorithms for active anomaly detection are based on the Isolation Forest method. They have also developed “anomaly signatures” to delve into the importance of features for specific outlier samples. Additionally, he will share some of the astronomical results they have achieved.

May 3 - No Meeting #

May 10 - No Meeting #

May 17 - No Meeting #

May 24 - Pau Batlle (Computing and Mathematical Sciences, Caltech) #

Title: Confidence intervals for functionals using optimization: from strict bounds to the Burrus conjecture disproof
Description: Confidence intervals defined as optimization programs date back to the 1960s, initially as tools to address the ill-posedness of inverse problems without introducing prior regularization. I will discuss the recently uncovered connection between those techniques and classical test inversion constructions and how it can help calibrate intervals for the original goal of ill-posed inverse problems. This insight has been vital in disproving the Burrus conjecture (1965) and opening the door for future improvements of optimization-based constructions.