Heroic-Faith is a leading medical firm that provides a respiratory monitoring system for medical professionals. With the integration of artificial intelligence, Heroic-Faith’s system examines a patient’s respiratory rates and detects abnormal breath sounds, offering a real-time alert on atypical respiratory signals for clinicians.
Challenges in the healthcare
Accurate detection of abnormal breath sounds is one of the core functions of Heroic-Faith’s product. To achieve a precise prediction, data scientists must repeatedly test different algorithms and adjust the hyperparameters of the AI model. This model training process is highly time-consuming, as it requires a long waiting time for computing. The lengthy model training process vastly increases the development cycle of Heroic-Faith’s product. As more training data comes in, a longer model training process is needed, further delaying the progress of making deep learning models for production.
Adding to the challenge, it is complicated to convert the deep learning model to a proper format, given the need to make the model applicable to run on edge devices, causing further troubles to the engineers. Moreover, it is difficult to compress the model into a tiny size while maintaining the accuracy of predictive results.
Results
Heroic-Faith applied artificial intelligence software, in order to shorten the model training process, and to deploy models on different devices in an easy way.
Through Avalanche Computing’s AI tool, hAIsten, Heroic-Faith trained their model distributedly on multiple GPUs. They accelerate the model training process by 300% in the optimized environment, which is built-in Avalanche Computing’s artificial intelligence tool. Thus Heroic-Faith can test more algorithms within a shorter period, without wasting time waiting for long-time computing. They are able to work much more efficiently and productively, quickly moving from model training to real-time model inference.
In near future, when deploying models on mobile devices, cloud servers, and other different edge devices, Heroic-Faith is able to transfer their model into various formats ⸺ depends on the operating system of the devices and the computing resources⸺in one command. Heroic-Faith can utilize computing resources in the most economic way. Most importantly, though the model is compressed on edge devices, real-time predictions with high accuracy are still available, which support medical professionals to make better clinical decisions.
With the artificial intelligence tools hAIsten, Heroic-Faith shortened the overall development cycle of deep learning projects, accelerate AI to market. Their respiratory monitoring system is expected to pass FDA this year.