Rethinking Supervisory Technology
Dr. Li Xuchun
Head of Supervisory Technology Office
Data Analytics Group
Monetary Authority of Singapore
AGENDA
Value of AI & DA
Big Data to Smart DataApplying Smart Data
Machine Learning & NLP
MAS SupTech
How MAS uses AI & DA to enhance supervision
(SupTech) – Initiatives & Infrastructure
SupTech Use Case
Market Surveillance
WAR IS 90%
INFORMATION
-Napoleon Bonaparte
Why the hype today?
Availability of Data
The 5Vs of Data
Value
Volume
Veracity Variety
Velocity
Big Data
Smart Data
DIFFERENCE OF SMART DATA & BIG DATA
The 5Vs of Data
SMART
DATABIG
DATA
ACTION
INSIGHT
DIFFERENCE OF DATA
BUSINESS
OBJECTIVES
DATA
FROM BIG DATA TO SMART DATA
Which should come first, business objectives or data?
What should I do differently as a senior manager?
Anthony ScriffignanoChief Data Scientist, Dun & Bradstreet
You lead with the problem.
You never lead with the data,
and you never lead with the technology.
Business Understanding
Data Understanding
Data Preparation
Modelling
Evaluation
Deployment
DATA
CONFIDENTIAL
FROM BIG DATA TO SMART DATAStart with the BUSINESS OBJECTIVES
CONFIDENTIAL
FROM BIG DATA TO SMART DATA
Enforcing data QUALITY
FROM BIG DATA TO SMART DATA
Using suitable METHODS
Supervised
Unsupervised?
+1
-1
Algorithm familyData Labels
Semi-supervised
Number of
samples
Threshold
%10 %20 %30
TP/TN
TP/TN
TP/TN
FROM BIG DATA TO SMART DATA
Checking for ACCURACY
FROM BIG DATA TO SMART DATA
Training (80%) Testing (20%)
Model
Build (“Train”)
Model
Test (“Validate”)
Model
Operationalise
Model
Historical Data Live Data
Achieving GENERALISABILITY
Easily understood. Who is the audience? What
is their level of technical expertise? Do they
understand the model?
•
Yield simple and direct
insights. Do the results make
sense?
Serve the business objective.
Does the model answer the
problem statement?
Look into the
black box!
FROM BIG DATA TO SMART DATA
•
•
Ensuring INTERPRETABILITY
SMART USE OF DATA: 7 Habits of the New Paradigm
Know the question you are trying to answer
Understand the data inside out
Find the right algorithm for the job.
Complexity may not be better
Be aware of limitations in the data,
algorithms, conclusions Data-driven insights
Competency
EffortProcess
1
Be skeptical. Question the data, assumptions, results
Automate to free up time to contextualise the results
Experiment. Don’t be afraid of trying and failing, and trying
again. Go for a “fail-fast” approach
2
3
4
5
6
7
• Lack of project management
• Over interpretation/extrapolation of results
• Absence of culture for converting insights into
actionable outcomes
• Perceiving data as technology projects
PITFALLS TO AVOID
How is your organization:
collecting, storing and using your data?
using data to drive insights and decision-making?
making analysis repeatable and shareable?
WHERE ARE YOU ON YOUR DIGITAL JOURNEY?
Computer programs that automatically improve their performance through experience (data)
Subfield of artificial intelligence
Building blocks
MACHINE LEARNING
Named entity recognition
Machine translation
Syntactic parsing & tagging
Automatic question answering
Sentiment analysis
NATURAL LANGUAGE PROCESSING
Automatically adapts and customizes to individual users
Discovers new knowledge from large datasets
Mimics humans and replaces certain monotonous tasks
Develops systems that are too difficult or expensive to construct manually
MACHINE LEARNING & NLP
Source Prepare Analyze Use
• What data do I
need?
• Where do I find
it?
• In what format
can I get it?
• Where do we
store it?
• Who can
access it?
• Who cleans and
validates it?
• How do we know
if it is valuable?
• How do we
extract insights
from it?
• What tools can
we use for the
analysis?
• How do we turn
insights into action?
• How do we
visualize data and
present insights?
• How do we best
distribute insights?
DATA VALUE CHAIN
MAS SupTechMission
Enhance financial supervision through the use of Artificial Intelligence and Data Analytics
Shape the SupTech landscape through collaboration and mutual exchange
Collaborate with RegTech ecosystem to overcome current challenges from compliance and supervision.
Big Data
Automation
Visualization
AI & ML
Usage of granular data
Data cleaning
Data consolidation
Data quality check
Interactive dashboards
Network graphs
Supervision of FI
Market surveillance
AML detection
Usage of financial data and alternative data
OUR INITIATIVES
MAS Supervisors
DataCollectionGateway
3rd Party Datasets
Enterprise Systems
Centralised Data Storage
Collection Storage Analytics Usage
Experimentation& development
EnterpriseApplications
Data Processing:
Data Cleaning
Data Storage
DATA ANALYTICS INFRASTRUCTURE
Automation Trade Analysis Prediction
Objectives
APOLLO
Expert Reports
Feature Engineering
Create prediction
engine
Export the predicted outcome
Human expert review
Improve model using new
training data
Build model with created features
Add trading data
Process
APOLLO
Feature Evaluation
Thank You!
Top Related