Abstract
This disclosure relates to the field of sensor measurements. Data from a variety of sensors, like temperature, audio, color or multi-axis acceleration is a key pre-requisite for understanding the status of a system and correcting for departures from its reference state. A challenge in this context is the cost of acquiring data either in terms of software sensing time and storage resources or also because the system may be inoperable or have operation reduced during sensor use.
A solution is disclosed that takes advantage of the predictive powers of generative pre-trained transformers (GPTs) used in large language models (LLMs) to take a small set of sensor data and generate from it the full set required for system analysis and control. The central idea revolves around the insight that when sensor data is tokenized in a custom way, GPTs can be successfully trained to make predictions not about language but about measured data. This enables minimizing the use of sensors to deliver control mechanisms that previously were more or even prohibitively time consuming and that may also have resulted in waste, which impacted the sustainability of their use.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
Recommended Citation
INC, HP, "Generating sensor data", Technical Disclosure Commons, (October 03, 2024)
https://www.tdcommons.org/dpubs_series/7396