Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Automating Fraud Detection - Continuous Model D...
Search
techsessions
February 14, 2018
Technology
0
6.7k
Automating Fraud Detection - Continuous Model Deployment
Stephen Whitworth, Co-Founder & Machine Learning Engineer, Ravelin
techsessions
February 14, 2018
Tweet
Share
More Decks by techsessions
See All by techsessions
Building Multilingual Recommendations Systems for BBC News
techsessions
1
6.6k
Bayesian Online Change-Point Detection at Scale
techsessions
2
7.3k
Modeling the Importance of Flight Partners at Skyscanner
techsessions
0
6.6k
Constructing Flight Itineraries with Machine Learning
techsessions
0
6.6k
Natural Language Processing in Media: Challenges and Opportunities
techsessions
0
14k
The Impact of Automation at Scale
techsessions
0
8k
Machine Learning at Zopa
techsessions
0
8.1k
The Inner Workings of Monzo’s Help Search Algorithm
techsessions
3
14k
Modern Techniques for Dimensional Reduction
techsessions
1
14k
Other Decks in Technology
See All in Technology
実践アプリケーション設計 ③ドメイン駆動設計
recruitengineers
PRO
13
4.1k
実践データベース設計 ①データベース設計概論
recruitengineers
PRO
4
2k
kubellが考える戦略と実行を繋ぐ活用ファーストのデータ分析基盤
kubell_hr
0
120
PRDの正しい使い方 ~AI時代にも効く思考・対話・成長ツールとして~
techtekt
PRO
0
380
allow_retry と Arel.sql / allow_retry and Arel.sql
euglena1215
0
150
カミナシ社の『ID管理基盤』製品内製 - その意思決定背景と2年間の進化 #AWSUnicornDay / Kaminashi ID - The Big Whys
kaminashi
3
720
AI時代に非連続な成長を実現するエンジニアリング戦略
sansantech
PRO
3
930
ソフトウェア エンジニアとしての 姿勢と心構え
recruitengineers
PRO
26
12k
Flutterでキャッチしないエラーはどこに行く
taiju59
0
210
AI時代にPdMとPMMはどう連携すべきか / PdM–PMM-collaboration-in-AI-era
rakus_dev
0
250
大「個人開発サービス」時代に僕たちはどう生きるか
sotarok
9
5k
トヨタ生産方式(TPS)入門
recruitengineers
PRO
6
1.4k
Featured
See All Featured
Producing Creativity
orderedlist
PRO
347
40k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
667
120k
Site-Speed That Sticks
csswizardry
10
800
How STYLIGHT went responsive
nonsquared
100
5.8k
jQuery: Nuts, Bolts and Bling
dougneiner
64
7.9k
Building Applications with DynamoDB
mza
96
6.6k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
51
5.6k
Gamification - CAS2011
davidbonilla
81
5.4k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
48
9.7k
Making Projects Easy
brettharned
117
6.4k
The Power of CSS Pseudo Elements
geoffreycrofte
77
5.9k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
34
3.1k
Transcript
Stephen Whitworth | 08.02.18 ravelin.com Continuous model deployment
ravelin.com ravelin.com Credit card fraud detection platform for merchants
ravelin.com ravelin.com Score customers in real time for likelihood of
fraud
ravelin.com ravelin.com Machine learning sits at the core of our
detection strategy
ravelin.com ravelin.com Normal ML deployment cycle: release few times a
quarter
ravelin.com ravelin.com Ravelin deployment cycle: deploy new models many times
a week
ravelin.com ravelin.com Frequency reduces difficulty: if something is hard, do
it more often. (Martin Fowler)
Training infrastructure • Python / Go hybrid pipeline • Packaged/distributed
through Docker • On demand compute on big machines • One line to build a new model, run experiments
Pipeline output • New model, trained from scratch • All
output archived to Google Cloud Storage • Performance metrics posted to internal registry • Model deployed to asynchronous live cluster • HTML report of performance for team
Summary report
Comparing two models
• Summarisation over raw details • Minimise manual toil at
all costs • Automation reigns king • Unit test output of models • Make model deployment ‘boring’ Principles for high-performing ML teams
• Data Scientists - join my team! • Head of
Product • Product Managers • Javascript Engineer • Investigations Analyst • Full Stack Engineers • Backend Engineers • Devops Engineer We’re hiring - www.angel.co/ravelin