Authors: Wendy Chong, Dinesh Verma, Mathews Thomas, Mudhakar Srivatsa, Satishkumar Sadagopan, Utpal Mangla
Today, a 911 emergency response time can take anywhere from 37.5 minutes to 43 minutes according to a study conducted by the University of California, San Francisco based on 63,000 cardiac arrest cases, a life-threatening condition, occurring outside the hospital premises. With the advent of 5G, an end-to-end 5G network slice can be established between the 911 caller, EMT (Emergency Medical Technician), and the hospital alongside an optimal driving route that supports rich multimedia communication between EMT and the attending physician, it is possible to make critical real-time decisions on behalf of the patient.
With the confluence of 5G, MEC (Mobile / Multi-access Edge Computing), and AI/ML (Artificial Intelligence and Machine Learning), such use cases are closer to reality. To achieve that, the network should support a common architecture from Core to Edge, for both Network and IT workloads, with the flexibility to move workloads across the network, as per the ebbs and flows of network and customer traffic supported by dynamic network slicing. It should also support the flexibility of build once and deploy anywhere on the Core to Edge spectrum, to provide the best experience even with bandwidth intensive use cases. This should be complemented with DevSecOps methodology and robust security framework to ensure 5G & Edge network security, cloud & container security, devices, application, and data security.
IBM’s implementation of the hybrid Cloud in support of 5G & Edge is depicted below.
Researchers at IBM, in collaboration with several Universities and Government labs, have developed a number of techniques and algorithms that solve AI/ML challenges in Edge Computing through the DAIS ITA program (a joint US/UK Government funded program for advancing the basic science of distributed analytics in multi-domain coalition environments). IBM Researchers will soon make it available as an Edge SDK (Software Developer Kit) on top of the IBM Hybrid Cloud platform.
IBM has already launched AIOps, providing IT infrastructure management for the cloud which can also be applied to network infrastructure. There is a high expectation that the net new value will come from industry use cases, creating new applications and services in the 5G world, in conjunction with MEC & AI/ML. The SDK will provide tools for the application developer community enabling the creation of these use cases. Amongst the core primitives in the SDK are, Federated Learning, Coresets, Model Fingerprinting, and Neural Tomography.
Medical and healthcare data is personal and private but learning about the conditions that were experienced by others that are similar is an enormous opportunity. How do we learn about the cohort of people that has a similar condition without sharing the raw data? Federated Learning algorithms form distributed cohorts and create a global model without revealing raw personal data.
Coresets are machine learning algorithms that provide mechanisms to transfer a succinct summary of the data from the ambulance to servers along the network edge or cloud as appropriate without losing data quality, while enabling real-time decisions with pre-trained AI models at the point of action, at the speed of 5G.
But how do we know these pre-trained AI models are accurate for this edge environment? Model Fingerprinting can solve this by selecting the right model from the model repository depending on the dynamic characteristics of the environment at the edge, ranking them for accuracy.
Equally challenging is to ensure that the application performance is maintained across varying network conditions. In the case of moving ambulance, Neural Tomography can monitor and configure the network slice dynamically instructing the driver in real-time to avoid a particular route (e.g., a congested tunnel).
IBM recently announced the collaboration effort with AT&T to prove out these algorithms in 5G enterprise settings. The IBM research campus is equipped with 5G infrastructure, handsets, and MEC servers, as well as taking advantage of full Edge computing and Hybrid Cloud stack to exercise the Edge SDK underpinned by cloud-like capabilities to develop, build, run and manage in a DevOps fashion.
Stay tuned for more in this series as we dive deeper into each of the capabilities, core primitives in the SDK in future blogs.
I want to give a HUGE THANKS to all my IBM colleagues who contributed to this blog.