This week we will have 15 minute talks by three undergraduates on the research experiences they were involved in  this summer. 

Joshua Gammage

Asymptotic Loss of Overparameterized Machine Learning Models

The bias-variance tradeoff seems to suggest that highly overparameterized models will perform progressively worse after a point. However, there is a 'double descent' phenomenon, which has been exploited by state of the art models.

Jerry Luckenbaugh

Coupled Oscillators and Hysteresis in Sparse Networks

The Kuramoto model of coupled oscillation is an important framework for understanding synchronization and self-organization in biology, physics, and engineering. Several consequences follow from varying the topology in networks of coupled oscillators. The behavior of the order parameter is analyzed as the number of connections in a network and the global coupling parameters are varied over time. We show numerical evidence of a hysteresis operator for the order parameter as a function of the global coupling.

Nuno Mendes Castanheira

Novel Solutions for Third Harmonic Generation Using Quasi-Phase Matching

Despite the ubiquity of nth-harmonic generation in nonlinear optics, there remains a fundamental concern with achieving high efficiencies using Third-Harmonic Generation (THG) with three-wave mixing methods. This is especially true in broadband applications, where n = 3 THG methods tend to reach a few percent in efficiency at the highest. We investigate a novel, coupled, nonlinear system numerically, which rises from four different photon interactions associated with Sum-Frequency-Generation (SFG), and nth-harmonic generation. Particularly, we investigate a system that mimics THG outputs. One of this system’s photon interactions is suppressed by means of lacking in phase-matching, and the remaining processes are matched artificially by the use of Quasi-Phase Matching (QPM) to achieve ideal functional behavior through inexpensive and replicable means.