Universal Sharpness Dynamics in Neural Network Training: Fixed Point Analysis, Edge of Stability, and Route to Chaos

no code implementations • 3 Nov 2023 • Dayal Singh Kalra, Tianyu He, Maissam Barkeshli

In gradient descent dynamics of neural networks, the top eigenvalue of the Hessian of the loss (sharpness) displays a variety of robust phenomena throughout training.

Crystalline gauge fields and quantized discrete geometric response for Abelian topological phases with lattice symmetry

no code implementations • 20 May 2020 • Naren Manjunath, Maissam Barkeshli

The discrete torsion vector implies a type of crystal momentum fractionalization that is only non-trivial for $2$, $3$, and $4$-fold rotation symmetry.

Strongly Correlated Electrons High Energy Physics - Theory Quantum Physics

Measurement-induced topological entanglement transitions in symmetric random quantum circuits

1 code implementation • 15 Apr 2020 • Ali Lavasani, Yahya Alavirad, Maissam Barkeshli

We study a class of (1+1)D symmetric random quantum circuits with two competing types of measurements in addition to random unitary dynamics.

Quantum Physics Disordered Systems and Neural Networks Statistical Mechanics Strongly Correlated Electrons