1. Bridging the Gap between Current Operation
System and AI-Integrated System
Dr. Dickson Lukose
3 – 4 April, 2018
Kuala Lumpur, Malaysia
2. Market Pull
2
Transformation
from Process Driven Enterprise
to Data Driven Enterprise
PROCESS-
DRIVE
ENTERPRISE
Slow Reaction to Market Dynamics Fast Reaction to Market Dynamics
Evolution of Business Paradigm
The AI Financial Summit APAC
3. Artificial Intelligence for Data Science
3
Dr. Marcell Vollmer
Chief Digital Officer @ SAP Ariba
The AI Financial Summit APAC
4. Artificial Intelligence for Data Science
4
Source: https://whatsthebigdata.com/2016/10/17/visually-linking-ai-machine-learning-deep-learning-big-data-and-data-science/
The AI Financial Summit APAC
5. Challenges in Big Data Science Initiatives
5
Overcoming Silo
Mentality
Overcoming
Data Sharing
The AI Financial Summit APAC
6. Data Lake (Data Swamp) vs Semantic Data Lake
6
DATA LAKE
Enterprise Data
Sensor Web
Structured, Semi-Structured
& Unstructured
Unstructured (Structured) Structured & Semi-
Structured
Structured
Linked Open Data
CRAWLERS DATA HARVESTERS
WEB
KNOWLEDGE
HARVESTER
DATA
INGESTION ENGINE
Social Media
?DATA
SWAMP
SEMANTIC
DATA LAKE
HARMONIZATION
CLEANSING
FUSION
DEDUPLICATION
The AI Financial Summit APAC
7. When do we need Semantic Data
Lake (Enterprise Knowledge Graph)?
7The AI Financial Summit APAC
8. Semantic Data Lake Design Goals
8
Data Ingestion
- User configurable tools to map, ingest and link
data from any source.
- Structured & Unstructured data processing,
NLP, Text Analytics, Image Analytics, Video
Analytics.
Data Representation
- Flexible, standard-based graph model,
accessible by informaticians and business
users.
- RDF & OWL
Data Consumption
- Semantic Search, Dashboards, Network
Analytics, Semantic Analytics
- Natural Language Question Answering
The AI Financial Summit APAC
9. Data Analytics (Building Models)
9
Classification & Regression
Clustering
Anomaly Detection
Topic Modeling
DeepNet
Timeseries
Association Discovery
The AI Financial Summit APAC
10. Case Studies: Applications of Artificial
Intelligence
Case Study One:
Subjective Analytics on Social Media and
Social Network
Case Study Two:
Chatbots
10
18. Chatbot (Software Agent) Landscape
Chatbot is not new!!!!
ELIZA is an early natural language
processing computer
program created from 1964 to 1966
at the MIT Artificial Intelligence
Laboratory by Joseph
Weizenbaum.
The AI Financial Summit APAC
19. AIML: Artificial Intelligence Markup
Language
2001: AIML, or Artificial Intelligence Markup Language, is an XML dialect for creating
natural language software agents released.
Dr. Richard Wallace (Chief
Science Office of Pandorabots)
A.L.I.C.E. (Artificial Linguistic
Internet Computer Entity) is a free
software chatbot created in AIML
(Artificial Intelligence Markup
Language).
The Loebner Prize is an annual
competition in artificial
intelligence that awards prizes to
the computer
programs considered by the
judges to be the most human-
like.
Steve Worswick
is the creator
of Mitsuku.
Mitsuku chatbot
wins Loebner
Prize for most
humanlike A.I.,
yet again.
The AI Financial Summit APAC
20. Chatbot Conversation Framework
20The AI Financial Summit APAC
AIML
Machine Learning
Deep Learning
Natural Language Processing
Question-Answering
Semantic Technology
23. 23
Data Science for Managers
16-18 October 2017
Operationalisation
“A repeatable, efficient process for creating and effectively deploying
predictive analytics models into production”
Step 1: Move from cottage industry
to an industrial process for
building analytics models.
Step 2: Access to data is standardised.
Step 3: Definitions of this data are
shared and analytical
datasets are generated in a
repeatable (and where possible)
automated.
Step 4: Use BDA Workbench
(BDAW) to implement
(realize) systematic approach
to data management feeds
and defining modelling
workflow.
Step 5: Use BDAW to enable
analytic team to perform
ongoing management and monitoring
of models that are deployed in
production.
23The AI Financial Summit APAC
24. 24
Data Science for Managers
16-18 October 2017
Key Operationalisation Challenges
• Analytic Model built by the Modelling
Group and Deployed by IT.
• Time required to deploy models and
to integrate models with other
applications can be long.
• Models are deployed in proprietary
formats.
• Models are application dependent.
• Models are system dependent.
• Models are architecture dependent.
• Model should be independent of the
environment (Dev/Stg/Prod).
• Model should be independent of the
application coding language.
Solution:
Predictive Model Markup Language
(PMML) or JSON-PML or CUSTOM.
Move deployment responsibility from
IT to Operations/Product-Team.
Use BDAW.
Analytic Infrastructure
IT Organization
Storage
Compute
Network
Analytic Operation,
Security and Compliance
Deployed Models
Operations, Product Team
24The AI Financial Summit APAC
25. 25
Data Science for Managers
16-18 October 2017
Predictive Model Markup Language (PMML)
• The Predictive Model Markup
Language (PMML) is an XML-
based predictive model interchange
format.
• PMML provides a way for analytic
applications to describe and
exchange predictive models
produced by data mining and
machine learning algorithms.
Ref: http://dmg.org/pmml/v4-1/GeneralStructure.html
• Since PMML is an XML-based
standard, the specification comes
in the form of an XML schema.
• The PMML XSD contains required
elements and attributes that must
be present for the PMML to be
valid
25The AI Financial Summit APAC
26. 26
Data Science for Managers
16-18 October 2017
(IT/Data Engineer) (Data Science Team) (Product Owner)
Model Building and Deployment
26The AI Financial Summit APAC
27. 27
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Revision Complexity
27The AI Financial Summit APAC
28. 28
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Performance
Q: How to measure
Business Impact?
A: Measure in terms of:
- Economic Impact
- Social Impact
- KPI
28The AI Financial Summit APAC
29. 29
Data Science for Managers
16-18 October 2017
Model Life Cycle – Acceptable Model Error
29The AI Financial Summit APAC
30. 30
Data Science for Managers
16-18 October 2017
Model Life Cycle – When is the Right Time for Model Revision
30The AI Financial Summit APAC
31. 31
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Revision
31The AI Financial Summit APAC
32. 32
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Revision Complexity
32The AI Financial Summit APAC
33. 33
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Revision Complexity
33The AI Financial Summit APAC
34. 34
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Revision Complexity
34The AI Financial Summit APAC
35. 35
Data Science for Managers
16-18 October 2017
Model Life Cycle – Model Revision Complexity
Process Data
Exploratory Data Analysis
Build Models in
Development/Modeling
Environment
Deploy Models in
Operational System
PMML
Model
Revision
Retire Old Model
and Deploy
Revised Model
PMMLDATA LOG
DATA LOG
35The AI Financial Summit APAC
36. 36
Data Science for Managers
16-18 October 2017
Model Building, Deployment and Revision
(IT/Data Engineer) (Data Science Team) (Product Owner)
36The AI Financial Summit APAC
37. Dr. Dickson Lukose (PhD)
GCS Agile Pty. Ltd.
Level 10, 461 Bourke Street
Melbourne, VIC 3000
Australia
Email: dlukose@gcsagile.com.au
Phone: +61408510817