Events During the Week of March 20th through March 27th, 2022
Sunday, March 20th, 2022
- Academic Calendar
- Spring recess
- Abstract: *Note: actual end time may vary.*
Monday, March 21st, 2022
- Academic Calendar
- Classes resume
- Abstract: *Note: actual end time may vary.*
- Department Meeting
- CLOSED Department Meeting
- Time: 12:15 pm - 1:45 pm
- Place: Virtual - link to be sent later
- Speaker: Mark Eriksson, UW-Madison, Department of Physics
- Closed Department meeting. Zoom link to be sent later.
- Host: Mark Eriksson
Tuesday, March 22nd, 2022
- Department Meeting
- CLOSED Department Meeting
- Time: 12:15 pm - 1:45 pm
- Place: Virtual - link to be sent later
- Speaker: Mark Eriksson, UW-Madison, Department of Physics
- Closed Department meeting. Zoom link to be sent later.
- Host: Mark Eriksson
- Council Meeting
- Physics Council Meeting
- Time: 4:00 pm - 5:00 pm
- Place: 2314 Chamberlin
- Speaker: Eriksson, UW-Madison, Department of Physics
- Host: Eriksson
Wednesday, March 23rd, 2022
- Physics ∩ ML Seminar
- Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer
- Time: 11:00 am - 12:15 pm
- Place: Online Seminar: Please sign up for our mailing list at www.physicsmeetsml.org for zoom link. We will also livestream the talk in Chamberlin 5280.
- Speaker: Greg Yang, Microsoft Research
- Abstract: You can't train GPT-3 on a single GPU, much less tune its hyperparameters (HPs)...or so it seems. I'm here to tell you this is not true: you *can* tune its HPs on a single GPU even if you can't train it that way! In the first half of this talk, I'll describe how, in the so-call maximal update parametrization (abbreviated µP), narrow and wide neural networks share the same set of optimal HPs. This lets us tune any large model by just tuning a small version of it — we call this *µTransfer*. In particular, this allowed us to tune the 6.7 billion parameter version of GPT-3 using only 7% of its pretraining compute budget, and, with some asterisks, we get a performance comparable to the original GPT-3 model with twice the parameter count. In the second half of this talk, I'll discuss the theoretical reason µP has this special property and the connection to the study of infinite-width neural networks and, more generally, the theory of Tensor Programs. The first half will target general practitioners or empirical researchers in machine learning, while the second half targets those who are more theoretically curious. This talk is based on
- Department Meeting
- CANCELLED - Department Meeting
- Time: 12:15 pm - 1:15 pm
- Place: CANCELLED
- Speaker: Mark Eriksson, UW-Madison, Department of Physics
- This meeting is CANCELLED due to lack of urgent business.
- Host: Mark Eriksson
Thursday, March 24th, 2022
- Astronomy Colloquium
- Whitford Lecture - Massive Star Envelopes and Explosions: 3D Simulations and Observational Consequences
- Time: 3:30 pm - 5:00 pm
- Place: 4421 Sterling Hall, Coffee and Cookies at 3:30 pm, Talk starts at 3:45 pm
- Speaker: Lars Bildsten, KITP, UCSB
- Abstract: Some stars unambiguously reveal their 3D nature to the observer (e.g. Betelgeuse),whereas other stars exhibit phenomena which we attribute to physics that is intrinsically multi-dimensional (e.g. convection). It has been a theoretical challenge to realistically model stars in 3D due to the obvious computational hurdles. This is especially true when radiative transfer must also be simultaneously solved through a highly turbulent medium with large density variations. However, computational capabilities have now reached the level where physically realistic 3D RHD models, in our case using Athena++, can be calculated to levels worthy of comparisons to observations in at least two distinct stellar contexts. The first is the outer envelopes of massive stars on the main sequence, where TESS observations are finding surprising temporal variability. The second case we are studying is whole star convective models of luminous red supergiants at the end of their lives. The structure of their outer envelopes is manifest when observed prior to explosion and impacts the early part of the supernovae light-curves, especially the initial shock breakout.
Web Link - Host: Professor Ellen Zweibel
Friday, March 25th, 2022
- Wisconsin Quantum Institute
- HQAN Quantum Research Colloquium
- Time: 9:00 am - 10:00 am
- Place: 4274 Chamberlin
- Speaker: Hannes Bernien and Kevin Singh (UChicago), Alex Levchenko (UW-Madison)
- Abstract: Join the hybrid meeting for HQAN's biweekly research colloquium. Coffee and pastries will be provided.
- Theory Seminar (High Energy/Cosmology)
- Wormholes and Saddles in Axion Gravity
- Time: 1:00 pm
- Place: Chamberlin 5280
- Speaker: Gregory Loges, UW Madison
- Abstract: The saddle-point approximation often used to make sense of path integrals is especially subtle when gravity is dynamical, most notably because of the conformal factor problem which renders the action unbounded below. In this talk I will discuss two aspects of saddle points in axion gravity, making use of the dual description in terms of a 3-form flux: (i) how Picard-Lefschetz theory can be used to identify in a democratic way which Lorentzian, complex and Euclidean saddle points contribute to Lorentzian path integrals, and (ii) the perturbative stability of the Giddings-Strominger Euclidean wormhole. Based on [2203.01956] with Gary Shiu and Nidhi Sudhir.
- Host: George Wojcik
- Physics Department Colloquium
- Exploring cosmology with the South Pole Telescope
- Time: 3:30 pm
- Place: 2103 Chamberlin Hall
- Speaker: Amy Bender, Argonne National Laboratory
- Abstract: Observations of the cosmic microwave background (CMB) probe a diverse set of fundamental physics in both the early and late universe. CMB measurements are a cornerstone of our now exceptionally precise Lambda-CDM model of cosmology, however, many questions remain. Did an inflationary epoch occur a fraction of a second after the Big Bang? What is the nature of dark energy? Is the Hubble constant controversy new physics or simply measurement systematics? I will describe the SPT-3G receiver that is currently installed on the South Pole Telescope and how it is measuring the CMB to answer these questions. Now starting the fourth year of survey observations, SPT-3G data is already providing tantalizing hints about what scientific goalposts will be possible in the very near future. I'll highlight a few of these results as well as providing perspectives on the upcoming CMB-S4 experiment.
- Host: Albrecht Karle