+ All Categories
Home > Documents > Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers...

Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers...

Date post: 16-Sep-2020
Category:
Upload: others
View: 0 times
Download: 0 times
Share this document with a friend
14
Artificial Intelligence Enablers for the Government of Canada Luc Gagnon Senior Assistant Deputy Minister, Chief Technology Officer Shared Services Canada / Government of Canada
Transcript
Page 1: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

Artificial Intelligence Enablers for the Government of Canada

Luc Gagnon

Senior Assistant Deputy Minister, Chief Technology Officer

Shared Services Canada / Government of Canada

Page 2: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

2

• Artifical Intelligence (AI) Democratization – a Journey

• Challenges

• Government of Canada (GC)-Wide AI Enablers

• Infrastructure AI Enablers of Interest to Shared Services Canada (SSC)

Topics

Page 3: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

3

Government of Canada and AI ….

AI Courses given by 2018 Turing prize winner Geoffrey Hinton to GC scientists

In 1994 & 1997

Page 4: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

4

A journey driven by algorithm research breakthroughs and Moore’s law

~25 years ago multi-layer perceptrons had 3 to 10 layers

• Everything written in procedural languages with some libraries

• Specialized AI researchers were the only users of the technology

• Limited data for training and testing

• Special purpose computers were scarce & not optimized for AI

Fast forward to 2019

• AI frameworks such as Tensorflow & Theano make Deep Neural Networks (DNN) and Machine Learning (ML) available without coding AI functions. With hundreds of layers.

• Cloud Infrastructure

• Tensor Processor Units (TPU) and Graphics Processor Units (GPU) are seamlessly available in the Cloud

• Data Lakes

• AI platforms and services

• IBM Watson, Amazon Web Service (AWS) ML, Google Cloud AutoML, customizable models such as You Only Look Once (YOLO), AWS Sagemaker, Microsoft Machine Learning,…

• Services without ML required : Natural Language Processing, computer vision, translation, forecasting, Bigdata.

Democratization of Artificial Intelligence

Page 5: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

5

1. Lack of specialized data science and AI skills

2. Insufficient volume of labeled data for ML

3. Need for widely available, elastic and kept-current AI platforms and tools

4. Data sovereignty and security management governance for AI applications

5. Selection of the right problems to solve with AI

6. Meaningful data sets and benchmarks to validate real-world performance

But what is preventing a faster adoption of AI?

Page 6: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

GC-wide AI Enablers

Page 7: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

7

• Comprehensive GC-wide Cloud program underway

Cloud–first program providing AI infrastructure & platforms

Supply Readiness Enablement Standardization

Responsibilities:

• Contracts for public

cloud

• Marketplace access

• Hybrid & private cloud

services

• Vendor management

• Broker service

• New Cloud product

features management

Responsibilities:

• Cloud strategy &

roadmap facilitation

• Cloud advice &

guidance from

cloud center of

excellence (CCoE)

• Reliability

engineering services

Responsibilities:

• Connectivity services

• Network and software

defined security and

data sovereignty

• Foundational services

& guardrails

• Cloud management:

reporting & auditing

• Cloud operations

• Migration planning

and execution services

Responsibilities:

• Cloud architectures

& enterprise

standards

• Software defined

infrastructure and

network standards

• Cloud deployment

templates

• Data shared services

• Provides best of class AI platforms and services

• Provides High performance computing options with GPU access

• Provides data sovereignty and security required for many GC applications

• Addresses challenges 3,4

Page 8: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

8

• Cover Natural Language Processing, computer vision, forecasting, translation, spoken speech understanding

• Some services leverage pre-trained models and perform retraining with user provided smaller data sets

• Guide efficient data labeling for future usage

• Designed for software developers with no deep algorithm knowledge

• Can be provisioned via the GC Cloud Broker Marketplace

• These services are often leveraging Application Program Interfaces (APIs) in compliance with the GC Digital Strategy

• Some use cases :

• Conversational spoken speech interfaces for government applications

• Bot services

• Face recognition

• These hosted AI services enablers address challenges 1,2,4,5*

Hosted AI Services

*Challenges

1. Lack of specialized data science and AI skills2. insufficient volume of labeled data for ML3. Need for widely available, elastic and kept-current AI platforms and tools 4. Data sovereignty and security management governance for AI applications 5. Selection of the right problems to solve with AI6. Meaningful data sets and benchmarks to validate real-world performance

Page 9: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

9

• Built around open source frameworks such as Theano, scikit-learn, Tensorflow, CNTK, MXNET

• AI frameworks are mainly used by GC data scientists and AI experts

• Enable easier collaboration in research and development

• Saves coding time and enables researchers to quickly experiment with different algorithms or DNN architectures

• Often provide already built implementations that run widely-available experimental benchmarks

• Some use cases :

• AI lab for scientists in science departments

• Advanced data analysis in data science

• Advanced AI frameworks enabler addresses challenges 3,6

Advanced AI Frameworks

*Challenges

1. Lack of specialized data science and AI skills

2. insufficient volume of labeled data for ML3. Need for widely available, elastic and kept-current AI platforms and tools 4. Data sovereignty and security management governance for AI applications 5. Selection of the right problems to solve with AI6. Meaningful data sets and benchmarks to validate real-world performance

Page 10: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

10

• Departmental data strategies have to be informed by near-term and longer-term AI needs:

• Machine learning requires high volume of labeled data

• 5K labeled examples per category for acceptable performance

• 10M labeled examples to match human performance

• An AI friendly data strategy needs a metadata strategy

• Labelling standards are not common

• Consider using synthetic data for creating AI capabilities

• lower cost, faster collection time

• If synthesis model is comprehensive data can be wider/deeper than real-world data

• Automatic labeling - perfect accuracy

• Can be used when sensitive data sets cannot be obtained

• AI-driven GC data strategy enabler addresses challenges 2,4,6

AI-Driven GC Data Strategy

*Challenges

1. Lack of specialized Data Science and AI skills2. insufficient volume of labeled data for ML3. Need for widely available, elastic and kept-current AI platforms and tools

4. Data sovereignty and security management governance for AI applications 5. Selection of the right problems to solve with AI6. Meaningful data sets and benchmarks to validate real-world performance

Page 11: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

Infrastructure AI Enablers

Page 12: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

12

• The most capable people can no longer cope with the volume of events being generated by their IT environments. AIOps is capable of processing millions of events and enable humans to focus on what matters.

• Adding diagnostic capabilities to advanced monitoring such as application performance monitoring (APM) and network performance monitoring and diagnostics (NPMD).

• A typical visualization and diagnostic machine learning use case.

• AIOps platforms enhance IT operation via machine learning and visualization.

• Faster diagnostic and even proactive prevention of technical issues by correlating different sources of data.

• Real-time diagnostic help for debugging teams leading to lower mean time to repair (MTTR).

• Can be used to automate actions to remedy problems.

Network and Data Center Operations with AI : Algorithmic IT Operations (AIOps)

Page 13: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

13

• Velocity and sophistication of the cyber threat can only be countered with intelligent automation to respond in cyber meaningful time (seconds not days/weeks/months)

• Investments in Data Lakes with vast amounts of relevant data for analysis

• Broad spectrum real time event correlation

• Queuing and prioritization of incidents

• Security Orchestration, Automation and Response (SOAR)

• User and Entity Behavior Analytics (UEBA)

• Risk and response predictive analysis from threat and defense posture analysis

AI in Cyber Security

Page 14: Artificial Intelligence Enablers for the Government of Canada...• Designed for software developers with no deep algorithm knowledge • Can be provisioned via the GC Cloud Broker

14

Questions?


Recommended