12

DATA SCIENCE AT DNB

WORLD BANK - FINSAC CONFERENCE ON FINTECH

IMAN VAN LELYVELD 22 MAY 2019

1

Agenda

Process How did it go? What have we accomplished so far?

Technology Challenges & Solutions

Showcase Explanation of PoCs: what are the new possibilities?

What’s next? What have we learned?

A dot on the horizon …

IT and tools PeopleData

A dot on the horizon …

Data PeopleIT and tools

The dream … … and reality

Which elements do we need?

• Research Area Network (RAN), Data Platform + AnalyticalWorkspaces/Datalabs/Data scienceToolkit, memory/cpu/storage

• Cloud deployment; Data(platform) connectivity, other connectivity (open data, etc..), quick scaling of datalabs

• Open source tooling (e.g. R,python, Git, Neo4J, MongoDB, SQLlight, MySQL, …..)

Tool

ing • Appreciation of the scientific method

• Knowledge of statistics (descriptive, explorative, predictive, causal, ...)

• Knowledge of coding in ‘interpreter’ languages (Python, R, Julia, ...) andsupport (Anaconda, JupyterNotebooks, Git, ...)

Peop

le

• Informal: knowledge networks, lunches, seminars

• Creating a community, many already do ‘something’ with datascience: Get-togethers, what do people need?, datascience 101 sessions, seminar withexternals, deep-dive sessies (R, Python, GITLab, Neo4J, MySQL, MongoDB, etc..), show first results

Cul

ture• Decentral vs. Central

• Governance (!!!) – data protection, deployment of analysis (KIVKII)

• Agile, pilots, data science as a brand• FTE’s

Org

anis

atio

n

Process: venturing down unbeaten paths…

• Technology

• Legal• tension between experiments and a complete contract

• Means and projects

Agenda

Process How did it go? What have we accomplished so far?

Technology Challenges & Solutions

Showcase Explanation of PoCs: what are the new possibilities?

What’s next? What have we learned?

ModernFlexibleScalable

SecureManageableTraceable

DNB Surf Sara

DNB Net.RAN High Secure

X10

X∞

Internet

VPNVPN Tunnel

SSH TunnelHTTPS

workplace workplace

Log

Repo.GIT

Ansible

workplace

ProjectProject

ProjectProject

DNBproxy

DNB Surf Sara

DNB Net.RAN High Secure

X10

X∞

Internet

VPNVPN Tunnel

SSH TunnelHTTPS

workplace workplace

Log

Repo.GIT

Ansible

workplace

Ansible

ProjectProject

ProjectProject

DNBproxy

Infinitenumber of projects

Makes governancepossible through a case-orientedapproach

Per VM- 100TB- 80cores- 512GB

X.. VM’sper project

Isolationper project

Agenda

Process How did it go? What have we accomplished so far?

Technology Challenges & Solutions

Showcase Explanation of PoCs: what are the new possibilities?

What’s next? What have we learned?

PoC 1: Credit Risk 90

PoC 2: CCP Risk Indicators 50

PoC 3: CDS Contagion 100

PoC 4: IRS Margin Requirements 60

PoC 5: Pattern recognition in Solvency II reports 20

PoC 6: Residential Real Estate 10

PoC 7: AnaCredit 10

PoC 8: Future securities statistics 5

PoC 5: Pattern recognition in Solvency II reports

Plausibility of group reporting in Solvency II Sizable reports (> 4.000 dimensions per group)

Domain knowledge required

Specifying DQ checks labour intensive

Applying machine learning algorithms Checking causal relations within reports

Comparing group reporting and solo reporting

On going work Further DQ checks

Linking with other financial data

Other algorithms (# dimensions >> # groups)

User-interface for visualising and feedback

Agenda

Process How did it go? What have we accomplished so far?

Technology Challenges & Solutions

Showcase Explanation of PoCs: what are the new possibilities?

What’s next? What have we learned?

Lessons?

• The hard core

• Hackathon

• Cross-pollination• Lelione• STAT ML

Community• Python & R lunches

• A whiff of Data Science

• Hands on case study with Jupyter Notebook for DNB board and management

• Manifest

• What is responsible data science?

• Datapreneur

• 22 participants from across the whole bank starting with open source and tackling their own business problems

• Python, R, GIT, Agile, co-coding

• Training

• Data Science 101, Joint with DNB Academy