For years, interoperability — the ability for two systems to communicate effectively — has been an important aspect of our increasingly digital world. For banking, healthcare, and other everyday industries, we’ve come to expect that the platforms we use to exchange information can communicate seamlessly whenever we need them to.
Because we all have hundreds of thousands of data points associated with our lives as they relate to our health, our finances, and other major facets of life as we know it, it makes sense that recent developments in machine learning and artificial intelligence (AI) could be used to make all this data work together for our benefit.
Interoperability in action: Healthcare
Let’s use healthcare as an example of how interoperable machine learning technology can enhance our lives. Consider high-tech medical procedures like CT scans that automatically generate large volumes of sensor data for a single patient as opposed to health information your doctor manually enters into a proprietary database during a routine check-up. Without a way to quickly and automatically integrate these disparate data types for analysis, there is lost potential for fast diagnosis of critical illnesses. This has created a demand for optimization across different information models. Current methods and legacy systems simply aren’t friendly in terms of interoperability — but recent developments in machine learning are opening the door for the possibility of stronger, faster translation between information platforms. The result could be vastly improved medical care and optimized research practices.
The role of neural networks
Modeled after the human brain, neural networks are comprised of a set of algorithms that are designed to recognize patterns. They interpret sensory data through a sort of machine perception, labeling or clustering raw input. The patterns they recognize are numerical, contained in vectors, into which all real-world data, be it images, sound, text or time series, must be translated. According to a 2017 article in MIT News, neural networks were first proposed in 1944 by Warren McCullough and Walter Pitts, two University of Chicago researchers who moved to MIT in 1952 as founding members of what’s sometimes called the first cognitive science department. Since that time, the approach has fallen in and out of favor, but today it’s making a serious comeback.
In 2017, the Open Neural Network eXchange (ONNX) format was created as a community-driven open-source standard for deep learning and traditional machine learning models. The goal of this project was to tackle the limitations created within a disjointed AI ecosystem by creating a standardized framework. With support from numerous companies in the AI community, ONNX has gained steam in terms of adoption for both software and hardware industries. ONNX allows developers and data science teams to minimize future performance and compatibility challenges while opening doors for massive innovations in a variety of tech-centric fields. Simply put, ONNX allows developers to use their machine learning framework of choice and removes compatibility barriers so hardware and software products can work together more easily, now and in future iterations.
In terms of compatibility, according to a Medium article by Microsoft’s Faith Xu, “…the ONNX community has contributed many different tools to convert and performantly run models. Models trained on various frameworks can be converted to the ONNX format using tools such as TensorFlow-ONNX and ONNXMLTools (Keras, Scikit-Learn, CoreML, and more). Native ONNX export capabilities are already supported in PyTorch 1.2. Additionally, the ONNX model zoo provides popular, ready-to-use models.”
Semantic interoperability – a requirement for successful AI
While the ONNX format has already helped to unify the AI and machine learning efforts of many large companies, it has become apparent that simply having all your data in the same format does not automatically mean success. This has spurred an ongoing focus on semantic interoperability for future projects. The idea is that you can’t necessarily learn the patterns, predictions, or anomalies of data when the data is a mash-up of sources that do not mean the same thing. Semantic interoperability is the ability of computer systems to exchange data with unambiguous, shared meaning. To this end, high-quality, human-annotated data sets are needed to accurately train machine learning models, regardless of whether your data has been aggregated from a single source or heterogenous sources and converted via an ONNX-style format or not.