← Back to Sessions

Using the University of Utah Messaging Passing System to Help Realize Real-Time Machine-Learning Modules in Network Operations

Description: 

Machine learning (ML) and, in particular, deep-learning (DL) models have demonstrated their improved utility over traditional (near) real-time seismological methodologies during periods of enhanced seismicity. As such, University of Utah Seismograph Stations is eager to begin leveraging these tools so as to improve its monitoring capabilities. However, operationalizing ML models still faces significant challenges. These challenges include, but are not limited to, models that are underlain by substantially more complex software stacks and models with computational demands much greater than that of their traditional seismological algrorithm counterparts. Both problems can simultaneously be addressed by way of distributed computing. Regardless of whether the computational hardware exists in the cloud or on-premises, distributed computing challenges are ultimately solved by the movement of data. To that end, we will provide a status update on the University of Utah Message Passing System (UMPS) as a pathway to utilize ML models in an operational setting. Currently, UMPS provides low-overhead, secure mechanisms that allow computers connected via a network to communicate through broadcast and request-reply patterns. Of particular importance is that the request-reply mechanisms allow for not only many requestors but many repliers. Consequently, computationally demanding DL algorithms are now inherently scalable and will be able to keep pace with the ever-increasing rate of data acquisition. Additionally, we introduce a preliminary many-to-many communication strategy that enables interaction with remotely running daemon processes.

Session: Emerging Developments in Operational Monitoring Systems and Products

Type: Oral

Date: 4/18/2023

Presentation Time: 05:00 PM (local time)

Presenting Author: Ben Baker

Student Presenter: No

Invited Presentation: 

Authors