The ACM seminar provides a regular venue for presenting and discussing topics such as numerical analysis, scientific computing, modeling, simulation and visualization. The seminar also hosts regular group meetings and special presentations associated with the NSF-funded RTG program Computation- and Data-Enabled Science (CADES).
Faculty, students and researchers are welcome to attend and present.
Meetings will be held in FHM 462, the large conference room in our department, unless specifically indicated otherwise. Unless specifically stated otherwise, the seminar meeting time will be Fridays at 10:30am.
Notification of upcoming presentations, including Zoom links when relevant, will be sent as calendar invitations to those on our mailing list. To subscribe, click here.
Organizer: Prof. Jeffrey Ovall (jovall@pdx.edu)
Current Seminar Schedule: Fall 2026
October 2 (10:30am): RTG Group Meeting
Research updates:
October 9 (10:30am): Mark Lowell, Clarifai Inc.
Title: Progressive Sharpening is Caused by Layerwise Jacobian Alignment
Abstract: During neural network training, the sharpness of the Hessian matrix of the training loss rises until training is on the edge of stability. As a result, even nonstochastic gradient descent does not accurately model the underlying dynamical system defined by the gradient flow of the training loss. We use an exponential Euler solver to train the network without entering the edge of stability, so that we accurately approximate the true gradient descent dynamics. We demonstrate experimentally that the increase in the sharpness of the Hessian matrix is caused by the layerwise Jacobian matrices of the network becoming aligned, so that a small change in the network preactivations near the inputs of the network can cause a large change in the outputs of the network. We further demonstrate that the degree of alignment scales with the size of the dataset by a power law with a coefficient of determination between 0.74 and 0.98.
October 17 (10am -- 5:45pm): Cascade RAIN Meeting (Washington State University, Vancouver)
This is the latest in a fun series of meetings dating back to 2014. The website is:
https://sites.google.com/view/rain2026/
There is no registration fee, but you are encouraged to register (on the meeting website) for planning purposes.
A history and description/philosophy of the meetings can be found here:
October 23 (10:30am): No meeting today
October 30 (10:30am): No meeting today
Title:
Abstract:
November 6 (10:30am): Marko Hajba, Vitrovitica University of Applied Sciences
Title: Residual-Based Error Control in Scientific Machine Learning
Abstract: In this talk, we present a residual-based framework for error control of scientific machine learning models for parametrized elliptic partial differential equations. The aim is to build adaptively expandable surrogate models, in which the residual of the governing equation guides both the refinement and the verification of the approximation. As a benchmark, we consider the localization landscape of Schrödinger Hamiltonians with four-parameter potentials, formed as sums of Gaussian bumps on a periodic square. Numerical experiments confirm that the residual norm, computable without a reference solution, is a reliable and efficient error estimator at every parameter value and that its local contributions indicate where the error concentrates in space.
November 13 (10:30am): Luka Marohnic, Zagreb University of Applied Sciences
Title: Adaptive rational quasi-Hermite approximation and applications in acoustic transmission problems
Abstract: In this talk we present an adaptive rational approximation method for real and matrix-valued functions of real parameter that uses barycentric form of the interpolant with weights optimized from the derivative data. Using derivatives at sampled points eliminates the need for test points commonly present in classical AAA methods. Furthermore, the approximation residual can be estimated from the Hermite condition slacks which are nonzero because the interpolant itself does not accomodate for the derivative data. The continuous residual allows for estimating the approximation error and guessing the next sampling point. Finally, we demonstrate the application of the method to the problem of finding quasiresonances in acoustic transmission problems.
November 20 (10:30am): Will Pazner, Portland State University
Title: Squaring the Triangle: the High-Order Duffy de Rham Complex and Low-Order Preconditioning
Abstract: The efficient solution of high-order finite element problems necessitates the use of matrix-free solvers. On meshes consisting of tensor-product elements, this technology is well-established. Sum factorization and partial assembly techniques allow for the efficient operator evaluation, and solvers such as p-multigrid or low-order-refined (LOR) result in uniformly well-conditioned systems. On simplicial meshes, matrix-free solver techniques are much less established.
In this talk, we develop low-order-refined preconditioners for high-order finite elements on triangles. We construct new finite element spaces that are defined by means of a triangular lattice on each element. These spaces form a discrete de Rham complex we call the Duffy de Rham complex. The resulting high-order elements give rise to systems that are spectrally equivalent, independent of polynomial degree, to their lowest-order counterparts on the refined mesh defined by the same lattice. The new spaces are supersets of the standard Lagrange, Nédélec, and L² spaces, and LOR solvers for the Duffy spaces can be used to precondition the standard systems using a fictitious space approach.
This talk will describe the construction of the spaces, the theory of spectral equivalence, practical AMG algorithms for the low-order system, and demonstrate numerical results on a variety of high-order test cases.
December 4 (10:30am): RTG Group Meeting
Research updates:
Dates for Winter 2027, Spring 2027
January 8
January 15 RTG Group Meeting
January 22
January 29
February 5 RTG Group Meeting
February 12
February 19
February 26
March 5 RTG Group Meeting
March 12
__________________________________________________________________________________________________________________
April 2 RTG Group Meeting
April 9
April 16
April 23
April 30
May 7 RTG Group Meeting
May 14
May 21
May 28
June 4 RTG Group Meeting