Explainable AI
Understand the logic behind Juniper’s Mist AI
Explainable AI is the ability for humans to understand the decisions, predictions, or actions made by an AI. This explainability is key to building the trust and confidence needed for broad adoption of AI and AIOps, in order to reap its benefits.
Learn how Juniper Mist AI solves common networking challenges with the following set of Explainable AI examples.
Finding faulty cables
A faulty cable is a prime example of a needle-in-a-haystack networking issue — it’s both time-consuming and difficult to manually identify a bad cable.
Decision tree algorithm
Using a decision tree algorithm, cable data such as frame errors and one-way traffic are analyzed to determine if a cable, whether copper or optical, is exhibiting poor quality that’s likely impacting end-user experiences.
Auto RRM (radio resources management)
Despite initial planning, Wi-Fi performance changes over time due to shifts in a site’s RF (radio frequency) characteristics. Manually adjusting radio resources can be cumbersome and difficult.
Reinforcement learning
The reinforcement learning algorithm intelligently and dynamically optimizes RF in real time for the best possible Wi-Fi coverage, capacity, and connectivity. This approach, which can be customized on a per-site basis, far surpasses relying on manual settings or traditional fixed algorithms.
Service level expectations (SLE) metrics
It can be challenging to gain deep actionable insights about a network’s state and behavior and how it impacts end-user experiences— let alone identify anomalous conditions like client connection issues following, for example, an Android software OTA update.
Mutual information
SLEs are a key tool that represent how your users experience network service, whether connected wirelessly, wired, or even out of site over the WAN. The mutual information algorithm helps you figure out which network features are having the most impact on the failure or success of your SLEs.
Marvis’s conversation interface
With the proliferation of users, devices, applications, and cloud, all coupled with the increasing complexity of connecting and securing sites, it’s becoming impossible to operate networks using traditional approaches. Manual operations that require logging into individual device CLI (command line interface) or digging through log files cannot keep up. The key to scaling is to shift to AIOps and takign advantage of a virtual network assistant (VNA). Hence, a VNA must translate human language to network operations.
Natural language processing (NLP)
NLP powers Marvis to interpret language. Simply ask Marvis about network health, unhappy users, or troubleshooting a site, and you’ll receive actionable information.
