SAP Leonardo Machine Learning - Making Business Applications Intelligent

CUSTOMER
Nazanin Zaker, Lead Data Scientist, SAP Machine Learning Business Network
Frank Wu, Head of SAP Machine Learning Business Network
SAP Leonardo Machine Learning
Making Business Applications Intelligent

2CUSTOMER© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ
The information in this presentation is confidential and proprietary to SAP and may not be disclosed without the permission of
SAP. This presentation is not subject to your license agreement or any other service or subscription agreement with SAP.
SAP has no obligation to pursue any course of business outlined in this document or any related presentation, or to develop
or release any functionality mentioned therein. This document, or any related presentation and SAP's strategy and possible
future developments, products and or platforms directions and functionality are all subject to change and may be changed by
SAP at any time for any reason without notice. The information in this document is not a commitment, promise or legal
obligation to deliver any material, code or functionality. This document is provided without a warranty of any kind, either
express or implied, including but not limited to, the implied warranties of merchantability, fitness for a particular purpose, or
non-infringement. This document is for informational purposes and may not be incorporated into a contract. SAP assumes no
responsibility for errors or omissions in this document, except if such damages were caused by SAP´s willful misconduct or
gross negligence.
All forward-looking statements are subject to various risks and uncertainties that could cause actual results to differ materially
from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements, which speak
only as of their dates, and they should not be relied upon in making purchasing decisions.
Legal disclaimer

Machine learning is the reality behind artificial intelligence
§ Big Data (for example, business networks,
cloud applications, the Internet of Things,
and SAP S/4HANA)
§ Massive improvements in hardware
(graphics processing unit [GPU] and
multicore)
§ Deep learning algorithms
§ Computers learn from data without
being explicitly programmed.
§ Machines can see, read, listen,
understand, and interact.
What is machine learning?
Why now?

4CUSTOMER© 2017 SAP SE or an SAP affiliate company. All rights reserved. ǀ Source: SAP CSG analysis, McKinesy Quarterly Report July 2016, Google PR, Microsoft PR, SAP Market Model
60%
Of companies see ML
as critical for
competitive
advantage
$18B
Enterprise Machine
Learning Market
by 2020
94%
Transactional
Enterprise
Digital
Enterprise
Intelligent
Enterprise
Of human tasks
will be automated
by 2025
97%
Image recognition
accuracy
(human: 95%)
95%
Speech recognition
accuracy
(human: 94%)
Productivity
Human repetitive
tasks
Enterprise system
Human high value
tasks (augmented
by AI)
The Automation of Repetitive Tasks is Allowing Humans to be More
Productive and Focus on Higher Value Tasks

76% of the world’s
transaction revenue
touches an SAP system
25 industries
12 lines of business
Data
Science
Platform
Intelligent
Services
Intelligent Apps
Conversational Interfaces
Machine Learning
SAP Leonardo Business Outcomes
Increase
revenue
Re-imagine
processes
Quality time
at work
Customer
satisfaction
Enabling
innovations
SAP Leonardo Enables the Intelligent Enterprise
Intelligent
S/4HANA
Intelligent
Cloud
The world’s largest
business network

SAP Leonardo Machine Learning: transforming enterprise data into business
value
Input Machine Learning Output
Train
model
Prepare
data
Apply
model
Capture
feedback
Text
Image
Video
Speech
… and more Services
(such as invoice processing,
profile matching)
…and more
Applications
(such as cash application)

SAP Leonardo Machine Learning: Portfolio of Capabilities
Data Science Platform & Tools
Developer
DataScientist
Text/ Document
Services
(e.g. Sentiment
Analysis)
Image/Video
Services
(e.g. Image
Classification)
Speech/ Audio
Services
(e.g. Voice
Recognition)
Structured
Data Services
(e.g. Time Series
Analysis)
Business
Services
(e.g. Service
Ticket Intelligence)
Graph
Services
(e.g. Link
Recommender)
Predictive
Services
(e.g.
Forecasting)
Intelligent Services
Data
Exploration
Data Integration
Data Preparation
End to End
Automation
In-Application
Deployment
Lifecycle
Management
ML Model
Creation
Model Storage
Production
Readiness
TensorFlow
Integration
Integration of Machine Learning into existing applications
(e.g. SAP Analytics Cloud, SAP Business Integrity Screening, SAP Cash Application)
Standalone Machine Learning Applications
(e.g. SAP Brand Impact)
Intelligent Apps
Conversational Interfaces
SAP Cloud Platform SAP HANA Platform
End-User

Reimagine your value chain with SAP Leonardo Machine Learning
• Trend Analysis
(Face, Age, Gender,
Emotion, Apparel)
• Personalized Design
Design
• Learning Recommender
• Synchronous Translation
of training content
• Career Path Recommender
Human Resources
• Predictive Maintenance
• Quality Inspection
• Optimal Planning &
Scheduling
Operations
• Cash Application
• Accounts Payable
• Remittance Advices
• Predictive Accounting
• SAP Business Integrity
Screening
Finance
• Image-based Purchasing
• Goods & Services Classification
• Supplier Risk Assessment
• Catalog Enrichment
Inbound Logistics
• Routing Optimization
• Supply Chain Resilience
• Last-mile Delivery
Outbound Logistics
• Brand Impact
• Social Media Analysis
• Customer Behavior
Segmentation
Marketing
• Conversational AI
• Service Ticketing
• Customer Support
• Solution Recommender
Sales & Service

SAP Leonardo Machine Learning Strategic Partnerships
§ Study & formulate best practices on AI tech,
§ Advance the public’s understanding of AI,
§ Serve as an open platform for discussion
and engagement about AI,
§ and its influences on people and society
§ SAP accepted as partner
§ Enables one global answer
to ML & AI ethics
§ First Enterprise offering to use
NVIDIA's Volta AI Platform
§ Running Kubernetes on NVIDIA
GPUs in SAP Data Center
§ Open-source software library for
Machine Intelligence
§ Our standard ML framework
(ease of training, enablement)
Partners Focus Areas Achievements

SAP Cash Application
Next-generation intelligent invoice-matching powered by machine-learning
History
Payments
Invoices
Matching
proposals
Improves days sales
outstanding
Allows shared services
to scale as the
business grows
Integrated with SAP
S/4HANA for reduced
TCO and time to value
Empowers finance to
focus on strategic tasks
and service quality
SAP Cash Application intelligently learns matching criteria from
your history and automatically clears payments.
SAP Leonardo
Machine Learning

SAP Cash
Application
Next-generation intelligent
invoice matching powered
by machine learning

SAP Leonardo Machine Learning Foundation
Enabling customers and partners to build the intelligent enterprise
Applications
Ready to use
Training
Inference
Ready to use Services
Bring your own Model
Customize Model
Create Training
SAP Cloud Platform

Core capabilities - description of features
• Deploy and run your own TensorFlow Model on ML foundation
• Manage your model’s status and monitor its resource consumption
• Leverage and benefit from the platform capabilities of ML foundation like authentication and scalability
Bring your own model
• Use your existing data assets to retrain ML foundation’s image or text classifier
• Simply access ML foundation’s API for retraining – no extensive machine learning knowledge required
• Leverage ML foundation’s capabilities to serve your training jobs
Customize model
Ready to use services
• provides readily consumable pre-trained models that can be used as a web service by calling simple
REST APIs
• Explore the functional services such as image classification, product image classification, topic detection,
time series changepoint detection

Broken product similarity search use case
Image Feature Extraction
Similarity Scoring
Service Ticket, e-mail incl.
image of broken product
Product Identification and
automatic classification
SAP’s Machine Learning automatically classifies product images and enables faster customer
interaction with precise information on potential product repair cost or item substitution.
SAP Leonardo
Machine Learning

The combination of Functional Services
Broken product similarity search use case
Images DB
Image Feature
Extraction Service
Vectors DB
Image Feature
Extraction Service
Similarity Scoring
Result

Broken product similarity search demo

Ready-to-use Services: Roadmap
Tabular Image & Video Text Speech & Audio Business Services
General
availability
§ Time series change point
detection
§ Similarity scoring
§ Image classification
§ Customizable image
classification
§ Image feature extraction
§ Topic detection
§ Text classification
§ Text feature extraction
§ Customizable text
classification
§ Intelligent Financing API
§ Ticket Intelligence -
Classification
§ Ticket Intelligence -
Recommendation
Alpha § Multi-dimensional time
series forecasting
§ Product image
classification
§ Human detection service
§ Object detection service
§ Machine translation
§ Language detection
§ Product text classification
§ Document clustering
§ Speech-to-text*
Road map § Time-to-failure
forecasting
§ Association rule learning
§ Customizable
recommender
§ Multi-dimensional data
clustering
§ Generic classification
(tabular and text)
§ Image segmentation
§ Face detection
§ Document optical
character recognition
§ Image text extraction
§ Image NER/extraction
§ Apparel detection
§ Sentiment analysis
§ Named entity recognition
§ Hate speech detection
§ File-to-text conversion
§ Voice recognition
(speaker identification)
§ Text-to-speech*
§ CV Matching
§ Customer Retention
§ Brand Impact
§ Accounts Payable
*Internal release only, not yet available for externals
Status as of October 2017

Ready-to-use Services: Easy Consumption
Calling REST APIS through the API Business Hub

Release Plan
___________Preparation___________ ____________Training_____________
available
newly released
roadmap
____________Inference____________ _____Usage_____
Training
Execution
Model
Publishing
Service
Consumption
Integrated ML
Capabilities
Configure
existing models
with your own
data.
Deploy your
model and make it
available as a
service.
Consume
scalable and
secure ML
Services on SAP
Cloud Platform.
Use ML
capabilities
integrated in SAP
solutions.
Customize Model
Ready to use
Apps
Ready to use
Services
Bring your own
Model
Training Creation
Train your own model
Upload your
Training-Script
and your data to
create your own
model.
Data PreparationData Exploration
Clean and label
your data.
Explore and
analyse your
data.
Prepare your data

Clean & Prepare Train Model
Test Model
Get Data
Part ID
Supplier
Name
Description
LaserJet Laser Printer - Plain Paper Print
M506DN - Plain paper LaserJet Printer -
Multi-level device security helps protect
from threats -Original HP Toner cartridges
with JetIntelligence and this printer
produce more high-quality pages. -11.7? H
x 16.5? W x 14.8? D -Media Feeder -1 x
automatic - 100 sheets - Legal (8.5 in x 14
in) weight: 60 g/m2 - 199 g/m2 - 1 x
automatic - 550 sheets - Legal (8.5 in x 14
in) weight: 60 g/m2 - 120 g/m2
Import
Export
Brand
Model
Technology
Dimension
Output
HP
M506d
Laser
11.7x16.5x14.8
Color
SAP Catalog Enrichment
Released last week

Catalog Enrichment Demo

Integration
SAP Cloud Platform
Catalog Management Application
System integration
Product
Description
Normalized
Attributes / Items
SAP Catalog Enrichment Service on SAP Leonardo ML Foundation

Information Extraction in SAP Catalogs
What are Information extraction systems:
• Collect information from many parts of text, and understand limited relevant pieces.
• Create a structured representation of relevant information.
Ÿ Organize information and make it practical for the users: as an example, Table
format catalogs
Ÿ Put information in a new form that allows further functions to be made by computer
algorithms on top of them: as an example, Make catalogs searchable
• Named Entity Recognition (NER): It is a sub task to find and classify names in text.
Goals
Solution

Three standard approaches to NER
q Rule Based NER
q Supervised Sequence models
q Unsupervised models
q Semi-supervised learning

Rule Based NER
q Create regular expressions to extract entities.
q Provide a flexible way to match strings of text.
Example: Suppose you are looking for a word that:
1. starts with a capital letter “N”
2. is the first word on a line
3. the second 2 letters are lower case letter
4. is exactly 5 letters long
5. the 4th letter is a vowel
6. The last letter is lower case
the regular expression would be “^N[a-z][a-z][aeiou][a-z]” where

Simple methods will not always work!
q Capitalization is a strong indicator for capturing proper names, but it can be
tricky:
§ First word of a sentence, titles, nested named entities are capitalized
q New proper names constantly emerge
movie titles, books, singers, restaurants, and etc.
q The same entity can have multiple variants of the same proper name
q Proper names are ambiguous

Learning System
q Supervised learning
§ labeled training examples
§ methods: Hidden Markov Models, k-Nearest Neighbors, Decision Trees, AdaBoost, SVM, RNNs (LSTM,
BiLSTM)
§ examples: NE recognition, POS tagging, Parsing
q Unsupervised learning
§ labels must be automatically discovered
§ method: clustering
§ examples: NE disambiguation, text classification
q Semi-supervised learning
§ small percentage of training examples are labeled, the rest is unlabeled
§ methods: bootstrapping, active learning, co-training, self-training
§ examples: NE recognition, POS tagging, Parsing, …

The ML sequence model approach to NER
Training
1. Collect a set of representative training documents
2. Label each token for its entity class or other (NA)
3. Design feature extractors appropriate to the text and classes
4. Train a sequence classifier to predict the labels from the data
Testing
1. Receive a set of testing documents
2. Run sequence model inference to label each token
3. Appropriately output the recognized entities

Conditional Random Fields
- sequence model: Conditional Random Fields (CRFs)
- It is a complete sequence conditional model, and not only a chaining of local models.
Training is slower comparing to hidden Markov models (HMM).
CRFs are very similar to maximum entropy Markov models (MEMMs).

Named Entity Recognition using multi-layered bidirectional LSTMs
Sentences are used as inputs for the recurrent neural network. Representation of words in the
sentence is via the form of embedding.
§ Possible embedding: word2vec, Glove, fasttext
Bidirectional LSTM network are used to classify the named entities.
§ 2 layers of bidirectional network
§ Softmax as the last layer to produce the final classification outputs.
§ AdamOptimzer for optimization
Evaluation:
§ F1 Scores, Prediction Accuracy and Recall

Named Entity Recognition using multi-layered bidirectional LSTMs (Cont.)
Tokenizing
Stemming
Word2vec/Glove
Training (BiLSTM)
Test and get
accuracy
64gb microsdxc card class
10Word2ve
c
model
Bi-
LSTM
Softmax
Word vectors
Embeddin
g
For words
capacity type type transmission
transmission
speed
speed
Bi-
LSTM

Examples and Detection Accuracy
Ø Sample text inputs and the classification results for SAP Catalogs.
hd capacity: 2.5tb hdd 32gb ssd
memory (ram): 6gb ddr3
operational system: windows 7
type: c560 ultraslim notebook
processor: 3.5 ghz intel core
hd capacity: 256gb
memory (ram): 8gb 2133mhz lpddr3 sdram
color: space gray
video card: iris plus graphics 640
type: macbook pro
processor:
2.3ghz dual core intel core i5 turbo boost up
to 3.6ghz intel
Sample product description #1:
macbook pro 13in - space gray: 2.3ghz 256gb
(mpxt2ll/a + s6202ll/a) sea # 1735383 quote #
2204065820 - 2.3ghz dual-core intel core i5
turbo boost up to 3.6ghz / intel iris plus graphics
640 / 8gb 2133mhz lpddr3 sdram / 256gb pcie-
based ssd / force touch trackpad / two
thunderbolt 3 ports / backlit keyboard (english) /
user's guide (english) / applecare+ for 13-inch
macbook pro mpxt2ll/a + s6202ll/a
Sample product description #2:
c560 23-inch ultraslim notebook c560 ultraslim
notebook - 3.5 ghz intel core 6gb ddr3 2.5tb
hdd 32gb ssd windows 7
Positive Negative
True 344 63
False 35 276
Results (accuracy: 86%)

SAP Leonardo
Machine Learning
Thank you

SAP Leonardo Machine Learning - Making Business Applications Intelligent

More Related Content

What's hot (13)

Similar to SAP Leonardo Machine Learning - Making Business Applications Intelligent (20)

More from NVIDIA (20)

Recently uploaded (20)

SAP Leonardo Machine Learning - Making Business Applications Intelligent