The need for foundation models in industry
Building an AI solution isn’t a simple task, especially when dealing with complex tasks and varied data types, as is common in industry. Up till very recently, the best approach was to create bespoke models for individual use cases, effective but resource intensive. Foundation models offer a way to help reduce the amount of work required to build AI systems that are both industry ready and functional.
In a recent podcast, Dr. James Loach, head of research at Senseye Predictive Maintenance, delved into what foundation models are, why they’re important and what that means for industry. To exemplify this, James describes a foundation model trained by Senseye to understand time series data, and the impact that has had.
Check out the full episode here or keep reading for a summary of that conversation.
What is a foundation model?
Foundation models are an important technology that has, over the last few years, come to the forefront of AI research. Just as the name implies, foundation models serve as a ‘foundation’ for building AI applications on, surpassing use case-specific models to allow generalization across a wide range of applications with little to no additional training.
For example, a classic AI application is a system controlling a robotic arm to pick up parts. Using traditional methods, the controlling model would have to be trained for each object the arm needs to pick up; one for screws, one for bottles, one for microcontrollers and so on. While far more functional than a hard coded pick-and-place algorithm, this system is still inflexible and costly to setup for new parts and products. This is where foundation models come in.
The key to a foundation model is its ability to generalize, in much the same way a human would. To continue the above example, a ‘robotic arm foundation model’ would be trained to pick up a wide range of items such that it understands how picking up objects of different size, material and weight work in general. From there, it could be deployed to pick up any objects with little to no additional training. This is the same principle which allows LLMs to perform novel tasks and learn new languages more easily.
Extending this type of functionality beyond something like language and into the realm of manufacturing and design isn’t a simple task though, and James notes the difficulties Senseye faced even creating a foundation model for something as constrained as time series data while at the same time highlighting the value achieved by that effort.
The value of applying a foundation model
The power of a foundation model is its ability to be deployed in any situation, even ones not covered explicitly by its training data. The time series foundation model trained by Senseye is a great example of this, where it brings intelligence into understanding of complex, arbitrary, time series data. While time series data is, in many ways alike across different applications, every instance is also unique. In order to bring human-like intelligence to machine understanding of this data, something as broad as a foundation model is required.
James explains how the use of their new time series foundation model was key in achieving human-like understanding of sensor data. The AI model is capable of learning and following trends, allowing it to more accurately judge when a sudden change in data is portent of a catastrophic issue, or simply a standard occurrence is key in achieving a robust, automated predictive maintenance solution.
Using AI for this application is important because understanding data requires a degree of intelligence; not every deviation is cause for alarm after all. With a robust foundation model in place, the process of deploying the model to a specific factory is simple and the results immediate, intuitive, and seamless.
As AI continues to be deployed across industries, foundation models like Senseye’s time series foundation model will be one of the key technologies to enable this growth. Enabling broader, general intelligence for industrial AI applications is key in realizing the promise of what AI can bring to industry, bringing flexible intelligence across data and systems to support humans with unique, data-driven insights.
To learn more check out the full podcast here.
Siemens Digital Industries Software helps organizations of all sizes digitally transform using software, hardware and services from the Siemens Xcelerator business platform. Siemens’ software and the comprehensive digital twin enable companies to optimize their design, engineering and manufacturing processes to turn today’s ideas into the sustainable products of the future. From chips to entire systems, from product to process, across all industries. Siemens Digital Industries Software – Accelerating transformation.
