Abstract: Provided are improved end-to-end self-supervised pre-training frameworks that leverage a combination of contrastive and masked modeling loss terms. In particular, the present disclosure provides framework that combines contrastive learning and masked modeling, where the former trains the model to discretize input data (e.g., continuous signals such as continuous speech signals) into a finite set of discriminative tokens, and the latter trains the model to learn contextualized representations via solving a masked prediction task consuming the discretized tokens. In contrast to certain existing masked modeling-based pre-training frameworks which rely on an iterative re-clustering and re-training process or other existing frameworks which concatenate two separately trained modules, the proposed framework can enable a model to be optimized in an end-to-end fashion by solving the two self-supervised tasks (the contrastive task and masked modeling) simultaneously.

Abstract: A high-efficiency non-collinearly phase matched parametric oscillator is provided, wherein a laser pumps a nonlinear optical material with a plural number of flat reflection surfaces that zigzag at least one parametrically generated off-axis radiation about the pump laser beam axis via multiple reflections from the surfaces. The off-axis zigzag oscillation of the radiation establishes parametric oscillation and improves energy coupling among mixing waves in a monolithic nonlinear optical material. Preferably the pump laser has a transverse beam size covering the area of the zigzagging parametrically generated radiation. To further enhance the performance of the off-axis zigzag parametric oscillator, the other parametrically generated radiation can be seeded by an external laser source or resonated in a cavity. The present invention also includes a double-side pumped off-axis zigzag parametric oscillator installed inside a standing-wave pump-laser cavity.