Abstract
Server cooling management frameworks that utilize neural networks are trained with a stream of multimodal sensor data, However, model predictions from such models lack explainability. This disclosure describes a fine-tuned model conditioned on multimodal sensor data to perform cooling schedule prediction. The model can also provide explainability by responding to natural language queries. The approach utilizes a transformer decoder architecture, with the model conditioned on multimodal sensor data from a natural server environment. The conditioning enables localizing the understanding of a language model to the specific context of use for server cooling scheduling decisions.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Shin, D, "Explainable Server Cooling Schedule Prediction Using Machine Learned Model Conditioned on Multimodal Data", Technical Disclosure Commons, (August 25, 2023)
https://www.tdcommons.org/dpubs_series/6182