Bubble universes. Computer illustration of multiple bubble universes as predicted by the Eternal Inflation theory. The inflationary theory proposes that after the Big Bang, a condition known as a false vacuum created a repulsive force that caused an incredibly rapid expansion, much faster than the ordinary expansion observed today. Since this expansion is faster than the speed of light, areas of inflation would form bubbles that would be completely isolated from each other. This artwork could also represent the creation of separate parallel universes as fluctuations in a quantum foam.
April 25, 2019 - April 26, 2019

Physics ∩ ML

Location: Microsoft Research Redmond

Day 1 | Thursday, April 25

Time (PDT) Session Speaker
Session 1 Plenary talks
8:00 AM–9:00 AM Breakfast
9:00 AM–9:45 AM Gauge equivariant convolutional networks Taco Cohen
9:45 AM–10:30 AM Understanding overparameterized neural networks Jascha Sohl-Dickstein
10:30 AM–11:00 AM Break
11:00 AM–11:45 AM Mathematical landscapes and string theory Mike Douglas
11:45 AM–12:30 PM Holography, matter and deep learning Koji Hashimoto
12:30 PM–2:00 PM Lunch
2:00 PM–4:05 PM Session 2 Applying physical insights to ML
2:00 PM–2:45 PM Plenary: A picture of the energy landscape of deep neural networks Pratik Chaudhari
2:45 PM–4:05 PM Short talks
Neural tangent kernel and the dynamics of large neural nets Clement Hongler
On the global convergence of gradient descent for over-parameterized models using optimal transport Lénaïc Chizat
Pathological spectrum of the Fisher information matrix in deep neural networks Ryo Karakida
Q&A
Fluctuation-dissipation relation for stochastic gradient descent Sho Yaida
From optimization algorithms to continuous dynamical systems and back Rene Vidal
The effect of network width on stochastic gradient descent and generalization Daniel Park
Q&A
Short certificates for symmetric graph density inequalities Rekha Thomas
Geometric representation learning in hyperbolic space Maximilian Nickel
The fundamental equations of MNIST Cedric Beny
Q&A
Quantum states and Lyapunov functions reshape universal grammar Paul Smolensky
Multi-scale deep generative networks for Bayesian inverse problems Pengchuan Zhang
Variational quantum classifiers in the context of quantum machine learning Alex Bocharov
Q&A
4:05 PM–4:30 PM Break
4:30 PM–5:30 PM The intersect ∩

Day 2 | Friday, April 26

Time (PDT) Session Speaker
Session 3 Applying ML to physics
8:00 AM–9:00 AM Breakfast
9:00 AM–9:45 AM Plenary: Combinatorial Cosmology Liam McAllister
9:45 AM–10:15 AM Break
10:15 AM–11:35 AM Short talks
Bypassing expensive steps in computational geometry Yang-Hui He
Learning string theory at Large N Cody Long
Training machines to extrapolate reliably over astronomical scales Brent Nelson
Q&A
Breaking the tunnel vision with ML Sergei Gukov
Can machine learning give us new theoretical insights in physics and math? Washington Taylor
Brief overview of machine learning holography Yi-Zhuang You
Q&A
Applications of persistent homology to physics Alex Cole
Seeking a connection between the string landscape and particle physics Patrick Vaudrevange
PBs^-1 to science: novel approaches on real-time processing from LHCb at CERN Themis Bowcock
Q&A
From non-parametric to parametric: manifold coordinates with physical meaning Marina Meila
Machine learning in quantum many-body physics: A blitz Yichen Huang
Knot Machine Learning Vishnu Jejjala
11:35 AM–12:30 PM Panel discussion with panelists Michael Freedman, Clement Hongler, Gary Shiu, Paul Smolensky, Washington Taylor
12:30 PM–1:30 PM Lunch
Session 4 Breakout groups
1:30 PM–3:00 PM Physics breakout groups
Symmetries and their realisations in string theory Sergei Gukov, Yang-Hui He
String landscape Michael Douglas, Liam McAllister
Connections of holography and ML Koji Hashimoto, Yi-Zhuang You
3:00 PM–4:30 PM ML breakout groups
Geometric representations in deep learning Maximilian Nickel
Understanding deep learning Yasaman Bahri, Boris Hanin, Jaehoon Lee
Physics and optimization Rene Vidal