A Collaborative Approach for Algorithm Operationalization Alexander - - PowerPoint PPT Presentation

▶
a collaborative approach for algorithm operationalization
SMART_READER_LITE
LIVE PREVIEW

A Collaborative Approach for Algorithm Operationalization Alexander - - PowerPoint PPT Presentation

Jan 11, 2023 242 likes •411 views

A Collaborative Approach for Algorithm Operationalization Alexander Werbos, L.E. Dafoe, S. Marley, T.S. Zaccheo Outline The Problem: Getting Novel, Robust Science Into Operations Tightening the O2R2O Loop Eliminating Rewriting

slide-1
SLIDE 1

A Collaborative Approach for Algorithm Operationalization

Alexander Werbos, L.E. Dafoe, S. Marley, T.S. Zaccheo

slide-2
SLIDE 2

Atmospheric and Environmental Research and Jefferies Technology Solutions

2

Outline

  • The Problem: Getting Novel, Robust Science Into

Operations

  • Tightening the O2R2O Loop

– Eliminating Rewriting – Reducing Manual Configuration – Providing a Unified Testing Framework

  • Building an Open Standard

– Multi-Mission Enterprise Algorithm Model

  • Implementing Advanced O2R2O Principles

– Algorithm WorkBench Component Model – Applicability to JPSS Modeling

slide-3
SLIDE 3

Atmospheric and Environmental Research and Jefferies Technology Solutions

3

Outline

  • The Problem: Getting Novel, Robust Science Into

Operations

  • Tightening the O2R2O Loop

– Eliminating Rewriting – Reducing Manual Configuration – Providing a Unified Testing Framework

  • Building an Open Standard

– Multi-Mission Enterprise Algorithm Model

  • Implementing Advanced O2R2O Principles

– Algorithm WorkBench Component Model – Applicability to JPSS Modeling

slide-4
SLIDE 4

Atmospheric and Environmental Research and Jefferies Technology Solutions

4

The Problem: Getting Novel, Robust Science Into Operations

  • Desire is for cutting-edge

data products built from the newest data streams

  • Existing process is effective,

but requires a long timeline between science and

  • perations
  • Strive for a more efficient

process that moves algorithm developers closer to operational systems

slide-5
SLIDE 5

Atmospheric and Environmental Research and Jefferies Technology Solutions

5

Outline

  • The Problem: Getting Novel, Robust Science Into

Operations

  • Tightening the O2R2O Loop

– Eliminating Rewriting – Reducing Manual Configuration – Providing a Unified Testing Framework

  • Building an Open Standard

– Multi-Mission Enterprise Algorithm Model

  • Implementing Advanced O2R2O Principles

– Algorithm WorkBench Component Model – Applicability to JPSS Modeling

slide-6
SLIDE 6

Atmospheric and Environmental Research and Jefferies Technology Solutions

6

Tightening the O2R2O Loop

Develop Science Code Operate Algorithm Scientific Testing Identify Defects and Upgrades

Develop Science Code Operate Algorithm Scientific Testing Operationalize Algorithm Identify Defects and Upgrades Adapt Algorithm Configuration Document changes for migration Test Operational Algorithm

slide-7
SLIDE 7

Atmospheric and Environmental Research and Jefferies Technology Solutions

7

Eliminating Rewriting

  • Strive for a single codebase that

is shared between science and

  • perational environments
  • Algorithms must use common

data interfaces

– Allow data to be retrieved in different ways in test vs. operations

  • Algorithms must be independent
  • f block size and parallelization

– Different operational systems will invoke algorithms on data with different coverage and resolution

Develop Science Code Operate Algorithm Scientific Testing Operationalize Algorithm Identify Defects and Upgrades Adapt Algorithm Configuration Document changes for migration Test Operational Algorithm

slide-8
SLIDE 8

Atmospheric and Environmental Research and Jefferies Technology Solutions

8

Reducing Manual Configuration

  • Develop a system-independent

way to express algorithm configuration

– Represent data flows between algorithms – Allow different system configurations to substitute data from different sources – Flexible model that can be read and modified by a variety of tools

  • System Configurations must be

editable by algorithm developers

– Test subsets of operational systems for small-scale integrations

Develop Science Code Operate Algorithm Scientific Testing Operationalize Algorithm Identify Defects and Upgrades Adapt Algorithm Configuration Document changes for migration Test Operational Algorithm

slide-9
SLIDE 9

Atmospheric and Environmental Research and Jefferies Technology Solutions

9

Providing a Unified Testing Framework

  • Scientific and Operational

configurations must be testable

  • n the same data

– Test infrastructure must use same data interfaces as algorithms, to ensure portability

  • Testing mechanisms must use

algorithm configuration model

– Facilitate automated tracking of data as system is tested – Verify complete system coverage

Develop Science Code Operate Algorithm Scientific Testing Operationalize Algorithm Identify Defects and Upgrades Adapt Algorithm Configuration Document changes for migration Test Operational Algorithm

slide-10
SLIDE 10

Atmospheric and Environmental Research and Jefferies Technology Solutions

10

Outline

  • The Problem: Getting Novel, Robust Science Into

Operations

  • Tightening the O2R2O Loop

– Eliminating Rewriting – Reducing Manual Configuration – Providing a Unified Testing Framework

  • Building an Open Standard

– Multi-Mission Enterprise Algorithm Model

  • Implementing Advanced O2R2O Principles

– Algorithm WorkBench Component Model – Applicability to JPSS Modeling

slide-11
SLIDE 11

Atmospheric and Environmental Research and Jefferies Technology Solutions

11

Building an Open Standard

  • Genuine Multi-Mission sharing of algorithms and data

requires collaboration

– Democratization of developing systems and algorithms that can run within them

  • No single organization should serve as authority

– Must enable distributed management of algorithms – Algorithms must be encapsulated as components

  • Standards must address the needs of diverse missions

and systems

– Facilitate smooth data flow between weather models, LEO, and GEO observing platforms – Allow new and upgraded data streams to be migrated to existing algorithms

slide-12
SLIDE 12

Atmospheric and Environmental Research and Jefferies Technology Solutions

12

Multi-Mission Enterprise Algorithm Model

  • Describe algorithm inputs and outputs in abstract terms

– Allow algorithms to be run at different grid resolutions – Automated systems to track algorithm interdependencies – Generate processing trees algorithmically

  • Represent outputs in semantically-useful form

– Users can apply reusable tools to export data in a variety of formats

  • Allow user-driven modification of metadata

– Users can experiment with configuration changes and visualize their results on the entire processing chain

slide-13
SLIDE 13

Atmospheric and Environmental Research and Jefferies Technology Solutions

13

Outline

  • The Problem: Getting Novel, Robust Science Into

Operations

  • Tightening the O2R2O Loop

– Eliminating Rewriting – Reducing Manual Configuration – Providing a Unified Testing Framework

  • Building an Open Standard

– Multi-Mission Enterprise Algorithm Model

  • Implementing Advanced O2R2O Principles

– Algorithm WorkBench Component Model – Applicability to JPSS Modeling

slide-14
SLIDE 14

Atmospheric and Environmental Research and Jefferies Technology Solutions

14

Algorithm Workbench Component Model

  • AER Algorithm WorkBench is a complete ground

processing system, evolved from the GOES-R testing infrastructure

– Runs multiple algorithm blocks in parallel – Allows users to automatically generate execution trees – Can be run on user workstations, servers, or cloud systems

  • Initial Effort Implements Open-Standard principles

– Algorithm data are stored as freely-editable XML files – Fragment-based storage architecture is designed to be extended by multiple users – Shared algorithm data model can be implemented in different environments

slide-15
SLIDE 15

Atmospheric and Environmental Research and Jefferies Technology Solutions

15

Applicability to JPSS Modeling

  • Prototype effort using MagicDraw

and the AER Algorithm WorkBench to work with JPSS algorithm model

– Users can model algorithm components in MagicDraw – Allows desktop visualization and editing of system design – Algorithm WorkBench imports XML model and can immediately use algorithms in its generated trees

slide-16
SLIDE 16

Atmospheric and Environmental Research and Jefferies Technology Solutions

16

Summary

  • Current Research to Operations loop is effective but

high-overhead

  • Using abstract design principles, algorithms can

standardize on a single code base shared between research and operations

  • Algorithm component metadata can allow system

engineering models to directly control algorithm execution

  • Prototype effort using MagicDraw and AER Algorithm

WorkBench has demonstrated these principles