FIGURE 5: SDC extends the principles of SDN’s primary focus on Ethernet and
packet layers to the optical transport layer. SDC is a key enabler of automation
throughout the network and across all operational levels and accelerates time-torevenue
from a month to minutes among other benefits.
7-LAYER ENTERPRISE AI SYSTEMS ARCHITECTURE
TABLE 1: Shows how the 7 layers of the ISO-OSI model map to the multiple layers of any enterprise AI system.13
March/April 2019 I 45
spectrum waste due to guard bands
while facilitating seamless
capacity growth without network
re-engineering or major disruption
to existing operating processes.
Another interrelated innovation
model designed to retire current
methods of optical capacity
planning, engineering, hardwarebased
deployments requiring
numerous truck rolls, extensive
manual labor, and human
intervention involves leveraging
software defined capacity (SDC)
as shown in Figure 5.
These frameworks can be
considered the key building blocks
toward the creation and deployment
of cognitive networks. Sophisticated
AI algorithms can be used, for
example, to build a microservicesbased
path computation engine
(PCE) that can replace manual
offline route and capacity processes
and can ultimately overcome
multiple and often challenging
optical fiber impairments.
The models presented by Masoud
are some of many as IoT and 5G
continue to evolve. For example,
Tapati Bandopadhyay offers another
OSI model architecture integrating
AI (see Table 1).
It is advisable for ICT designers
and professionals to embark on
becoming more acquainted with
various new architectures from
IT, especially those that advocate
increased collaboration between
ICT, OT, and IT.
API Layer: Containerized, easy to consume, easy to train and build on, easy to deploy, APIfied, modularized, AI in Lego forms.
Presentation
Layer:
UI/ UX of AI use cases: Presentation and visualization of output such as classification models/clusters/
anomalies | Chat- text, voice, video | AR/VR | HMI
Use-case
Layer:
Combining/ leveraging multiple core algorithms to solve business use-cases: e.g. Tensorflow API-> image processing>
image classification/ clustering -> damaged cars vs. normal cars basis whatsapp/ mobile images from field- for
automotive insurance- remote damage evaluation
Algorithms
Layer:
Selection of most apt algorithms basis the use-case/ problem e.g. image/ text/ missed, temporal, transfer learning,
reinforcement, NN's with memory- LSTM, HTM | Problems: classifier, clustering, fraud, profiling, CLV, churn, predictor,
synthetic data gen, approximation, autoML, meta-learning
Data Processing
Layer: Data prep: quality checks, fitment for ML/training, assumptions testing, cleaning, sparse/lossy/noisy/blurred
data handling, Data security and governance, privacy, access control, regulatory compliance e.g. PII, GDPR
Data Integration
Layer:
Data search, identification of relevant & trusted sources, integration, data lakes, connectors, mixed data
structured- unstructured
Physical
Layer: AI-Optimized Chips | Infra: GPU, TPU, Neuromorphic, Optical, Quantum computing