This presentation covers some of the major data science and AI announcements from the May 2020 Microsoft Build conference. Included in this talk are 1) Azure Synapse Link, 2) Responsible AI, 3) Project Bonsai & Project Moab, and 4) AI Models at Scale (deep learning with billions of parameters).
1. Data Science Recap
Mark Tabladillo Ph.D.
May 21, 2020
Founder, PASS Data Science Virtual Chapter
2020
2. Recap of Main News
from Microsoft
Build 2020
Cloud Solution Architect
Microsoft United States
Connect on LinkedIn
Twitter @marktabnet
3. Topics
Azure Synapse Link
Responsible AI
Project Bonsai & Project Moab
AI Models at Scale
4. What if I want to run analytics in near real-
time on my operational data at scale?
5. Azure Synapse Link
Microsoft is announcing Azure Synapse Link, a cloud-native
implementation of HTAP (hybrid transactional analytical processing),
which is an architecture for enabling analytics on live operational
data. With Azure Synapse Link, Azure is the first cloud service to
deliver on the promises of HTAP, without the costs, complexities and
trade-offs associated with implementations on-premises.
Azure Synapse Link is now available with Azure Cosmos DB and will
soon be available with Azure SQL, Azure Database for PostgreSQL
and Azure Database for MySQL.
6. Azure Synapse Link:
Building real-time HTAP solutions with Azure
Cosmos DB & Azure Synapse Analytics
https://azure.microsoft.com/en-us/blog/azure-analytics-clarity-in-an-instant/
7. Azure Cosmos DB is optimized for
operational workloads with single-digit
millisecond read and write latency
99.999% high availability, guaranteed
throughput and consistency
Turnkey global data replication across all
Azure regions
Fast NoSQL database with open APIs for any scale
What is Azure Cosmos DB
Real-time
Applications
& Services
Azure
Cosmos DB
8. If you have large amounts of data,
analytical queries will take a long time
to run and will be resource intensive
HUGE performance impact on the
OLTP workloads
Running OLTP and OLAP workloads on the same
database
Real-time
Applications &
Services
Azure
Cosmos DB
Reporting &
Dashboards
Azure Cosmos
DB
Spark connector
10. Analytical Store
Column store optimized for
analytical queries
Transactional Store
Row store optimized for
transactional operations
Azure Cosmos DB Azure Synapse Analytics
Container
Cloud-Native HTAP
Azure
Synapse Link
SQL
Auto-Sync
Machine learning
Big data analytics
BI Dashboards
Operational
Data
Generate near real-time insights on your operational data
Azure Synapse Link: How it works?
13. Responsible AI in Three Areas
Understand New model interpretability and fairness assessment capabilities enable the
development of more accurate and fair models.
Protect New differential privacy computing capabilities enable customers to build
machine learning models using sensitive data while safeguarding the privacy of
individuals. This is a result of the partnership between Microsoft and Harvard’s
Institute for Quantitative School Science, which was announced last September.
Additionally, new confidential machine learning capabilities provide a secure and
trusted environment for machine learning.
Control New capabilities for fine-grained traceability, lineage, and access control of data,
models and experiments enable organizations to meet strict regulatory
requirements. Additionally, new workflow documentation capabilities to enforce
accountability in the machine learning process will be made available to
customers shortly after the Build conference.
17. Project Bonsai Public Preview
Create and optimize intelligence for
industrial control systems with simulations
and machine teaching
Project Moab
Open-source machine teaching robotics
hardware kit
New Technical Demos and Customer Stories
featuring SCG and partner simulations using
Project Bonsai
18. AI models at scale
Massive, multi-purpose
AI models
Infrastructure at scale
The AI Supercomputer
Development at scale
Empowering every
developer
19. ▪ Microsoft Turing: Largest AI
model ever built (17B
parameters)
▪ Changing how AI is
developed: from narrow,
custom models to multi-
purpose, customized, massive
models
▪ Turing language: Most
powerful model for multi-task
natural language processing
▪ The road of generalization:
Multi-modality text / images /
video
20. AI models and
development at scale
▪ Announcing Open Source frameworks & optimizers for massive
model training
▪ Future release of Microsoft Turing language model
AI computing at scale ▪ Announcing one of the top five publicly disclosed supercomputers
in the world