"Federated learning: out of reach no matter how close",Oleksandr Lapshyn
DataLive conference in Geneva 2018 - Bringing AI to the Data
1. 1
Bringing AI to the Data
Sasha Lazarevic, IBM Switzerland
https://www.linkedin.com/in/lzrvc/
LZRVC.com
IBM Watson
2. 2
Barriers to deliver business value from data and AI
1. Lack of skills (data science, AI, microservices, DevOps, management)
2. Outdated infrastructure and applications
3. Data quality problems
4. Governance or regulatory issues
Dev Ops
Sec
AI
3. 3
IBM Approach to removing these barriers
1. Engagement Methodology
2. Data Science Methodology
3. IBM Data and AI Platform
4. Support Model
4. 4
Engagement Methodology – Watson Value Framework
Value Definition
and Measurement
Step 1. Define what
success of the project
looks like to your
business and
stakeholders
Step 2. Select up to 3
KPIs for measuring
progree towards your
success definition
Step 3. Define how you
will measure each KPI
Step 4. Establish KPI
baselines and
benchmarks
Step 5. Observe,
measure and refine KPIs
Business Value Delivery Roadmap
5. 5
IBM Data Science Methodology
Source: https://www.ibmbigdatahub.com/blog/why-we-need-methodology-data-science
6. 6
IBM Data and AI Platform
ICpIBM
Cloud
Watson Studio & ICP for Data
NLU
NLC
Vision
WKS
Language
Speech
Watson
Assistant
Watson
Discovery
Watson
Compare and Comply
Customer Care
Risk and
Compliance
Knowledge
Worker
Process
Automation
Use Cases
Horizontal
Applications
Data
Science and
AI IDE
Operating
Infrastructure
IBM AI
OpenScale
Traceability
Explainability
Bias Mitigation
Prediction
Consistency
Continuous
Learning
IBM Q
WEX
SPSS
Developer
Tools
7. 7
IBM Cloud and ICp
1. IBM Cloud
Bare metal, virtual servers, storage
Containers, middleware, database services
Workload migration services
Global reach with more than 50 data centers
2. Swiss Banking Cloud
Highly automated, Swiss-managed IaaS environment
Modular and replicable
3. IBM Cloud private
On premise with/without managed services
Container platform based on Kubernetes
10. 10
Watson Studio
Watson Studio Live Demo (https://dataplatform.cloud.ibm.com/)
Watson Data Catalog
Community Assets
Data Refinery (Transformations, Visualizations)
Jupyter Notebooks
ML Flow Modeler (SPSS Modeler compatibility)
Neural Network Modeler
Team Collaboration and Access Controls
12. 12
ICP for Data – Data Exploration
Demonstration Video - https://ibm.biz/BdYG9n
13. 13
ICP for Data – Data Transformation
Demonstration Video - https://ibm.biz/BdYG9n
14. 14
ICP for Data – Data Visualizations
Demonstration Video - https://ibm.biz/BdYG9n
15. 15
ICP for Data – Machine Learning Modeling
Demonstration Video - https://ibm.biz/BdYG9n
16. 16
ICP for Data – Machine Learning Modeling
SPSS Modeler Streams - Demonstration Video - https://www.youtube.com/watch?v=sAnwvo6i3GU
17. 17
New Watson Features
Watson Assistant (fka Conversation)
Digressions – answer user’s question out of the context
Disambiguation ("Did you mean __ ?")
More response types : buttons, images, videos etc
Integrations with Salesforce, Avaya, ServiceNo
Log based chat builder (using human-to-human chats)
Bot asset exchange
Modularization of your assistant through Skills
19. 19
New Watson Features
Watson Discovery
Document segmentation
Connect to Sharepoint, Box, Salesforce …
Excel is now supported in addition to pdf, word, ppt, json and html
Smart document understanding
20. 20
New Watson Features
Compare and Comply
Learn the contract structure and language
Enable complex operations like comparison to other documents
Use cases like: “Find all payment terms in a contract”, “Identify differences in terms between two
similar contracts”, “Compare contract with invoice”
User Interface
Watson Compare
and Comply
Watson
Discovery
Watson
Knowledge Studio
Business
Owner
21. 21
New Watson Features
AI OpenScale
Data and model bias detection
Logging for traceability to business outcomes
Explainability of MK and DL models
Instrumentation for business insights
Business operation dashboard
22. 22
Watson on ICp
Watson Assistant
• GA Date: September 26, 2018
• Initial Capabilities:
• Classification (intents)
• Entities
• Dialog
• Full Functionality Planned for 1H-19
• Languages: All Public Cloud Languages
Watson Speech-to-Text: Customer Care
• GA Date: September 20, 2018
• Initial Capabilities:
• Speech Transcription
• Language/Acoustic Customization
• PCI Redaction
• Language Support:
• English, Japanese, Korean out-of-the-box
• Other currently supported languages may
require customization and/or account service
support to achieve desired quality levels.
Compare & Comply
• New Features: September 28, 2018
• Element Classification model
enhancements
• Comparison API
• Table Understanding
• OCR/scanned document intake
• Feedback API/Document Visualizer
Bringing AI to the Data
24. 24
Bringing AI to the Data
Sasha Lazarevic, IBM Switzerland
https://www.linkedin.com/in/lzrvc/
LZRVC.com
Thank you !
IBM Watson
Editor's Notes
Business understanding
Understanding lays the foundation for successful resolution of the business problem. The business sponsors define the problem, project objectives, value definition and measurement
2. Analytic approach
Data scientist defines the analytic approach to solving the problem and identifies techniques suitable for achieving the desired outcome.
3. Data requirements
Choice of analytic approach determines the data requirements
4. Data collectionThe data scientist identifies and gathers data resources—structured, unstructured and semi-structured—that are relevant to the problem domain
5. Data understandingDescriptive statistics and visualization techniques can help a data scientist understand data content, assess data quality and discover initial insights into the data
6. Data preparationdata cleaning, combining data from multiple sources, transforming data, feature engineering This takes 70 percent of overall project time. It can drop as low as 50 percent if data resources are well managed, Automating some steps of data preparation may reduce the percentage even farther:
7. Modelinguse a training set to develop predictive or descriptive models. The modeling process is highly iterative.
8. Evaluationcheck whether the model addresses the business problem appropriately, by computing various diagnostic measures
9. DeploymentDeploye into the production environment
10. Feedbackthe organization gets feedback on the model’s performance so data scientist can refine the model, increasing its accuracy and thus its usefulness
Integrated Collaboration Environment which supports full AI lifecycle
Integrated Collaboration Environment which supports full AI lifecycle
Integrated Collaboration Environment which supports full AI lifecycle
For 10 team members, 3 nodes are required, with 18 virtual cores and 64 GB RAM
For bigger team, 6 nodes with 24 virtual cores and 82 GB RAM
Smart Document Understanding that will be able to distinguish the document structure (research paper, financial statement, wide variety of document structures etc) and query the data across different parts of the document like it is SQL database
Customizations like earlier are possible through WKS, Custom UI