DISTRIBUTED ENSEMBLE LEARNING FOR PROVISIONING INDICES, ACTIONS, AND INPUTS FOR ASSURANCE AND PERFORMANCE

Copyright 2018 Cisco Systems, Inc. 1

DISTRIBUTED ENSEMBLE LEARNING FOR PROVISIONING INDICES, ACTIONS, AND INPUTS FOR ASSURANCE AND PERFORMANCE

AUTHORS: J.D. Stanley

Carlos M. Pignataro Plamen Nedeltchev Victoria Blaylock

Andrea Gold Prabhat Bhattarai

CISCO SYSTEMS, INC.

ABSTRACT

Self-adapting machine learning software provides actionable insights to proactively

upgrade other elements of the system and environments that are ancillary or adjacent to the

main system. These suggested actions occur based on software threshold calculation

matrices distributed throughout network nodes. Outputs and reports are distributed to

humans, machines, or other math-based cognitive decision support systems to operate

based on the actions or suggestions. The math-based learning approach to combining

multiple machine learning engines and models produces higher accuracy levels and

statistical outcomes than traditional one-calculation approaches.

DETAILED DESCRIPTION

The emerging nature of distributed machine learning on edge devices, edge nodes,

and across distributed control points in infrastructures faces many challenges in ensuring

automation and orchestration. As those solutions evolve in the market, large scale

provisioning systems (e.g., Software Defined Networking (SDN) networks, Network

Functions Virtualization Infrastructure (NFVI), or even Operations Support Systems (OSS)

/ Business Support Systems (BSS) variants) will focus heavily on the service chains of the

infrastructures and workflows of the resources or services. The gap in this model is a

purpose-built flow technique on distributed Machine Learning (ML) engines.

Abstracting out calculations enables users to implement them as inputs into other

automation and infrastructure management solutions and/or as indices generated on the

ebbs and flows of the system environments. Users seek more insights and more data to help

Copyright 2018 Cisco Systems, Inc. 2

them make informed decisions across many variables to optimize resource deployments,

mitigate risks, or optimize infrastructure elements.

Most of the existing ML approaches in the market are siloed and standalone, and

are applied to a specific use case with a desired result. For example, cybersecurity software

calculations seek out security and threat risks. The same interrogated dataset may be

leveraged and examined from a different lens, such as inventory attributes (e.g., mixing the

cyber data with inventory data). This generates new datasets and insights for other purposes

and uses (e.g., insurance rates based on multi-factor risk assessments).

In distributed ML engine environments, using the system calculations and derived

data has value outside of the system. Where one user may see a set of readings as noise,

the other user may see value or may combine that data into their own adjacent system

calculations.

Thus, the problem relates to siloed calculations between systems, and especially to

systems that touch and flow over the same infrastructu...