IEEE 3303-2023
IEEE Standard Adoption of Moving Picture, Audio and Data Coding by Artificial Intelligence (MPAI) Technical Specification Compression and Understanding of Industrial Data 1.1

Standard No.

IEEE 3303-2023

Release Date

2023

Published By

Institute of Electrical and Electronics Engineers (IEEE)  US  /  IEEE

Latest

IEEE 3303-2023

Introduction

Standard Overview and Technical Background

IEEE Std 3303-2023 formally adopts the MPAI (Moving Picture, Audio and Data Coding by Artificial Intelligence) technical specification "Compression and Understanding of Industrial Data 1.1" (MPAI-CUI). This standard aims to utilize artificial intelligence technology to extract key information from the governance, financial, and risk data of industrial enterprises to predict the performance of enterprises within a specified prediction period. Its core outputs include Default Probability, Organisational Model Index, and Business Discontinuity Probability.

---

AI Framework and Interoperability Levels

The MPAI standard system is based on the AI Framework (AIF), defining applications as AI Workflows (AIW) composed of multiple AI Modules (AIM). MPAI defines three interoperability levels: Level 1 (implementer-specific, compliant with AIF standards), Level 2 (compliant with application standards), and Level 3 (certified through performance evaluation). The higher the level, the stronger the transparency, credibility, and reliability.

Interoperability Level	Requirements	Characteristics
Level 1	Implementer-specific, only needs to comply with AIF standards	Lowest interoperability, implementation details not disclosed
Level 2	Compliant with MPAI application standards, passed conformance testing	Standardized input/output, enhanced replaceability
Level 3	Passed performance evaluation, certified by authorized evaluation bodies	Highest level, ensuring reliable, robust, fair, and reproducible performance

---

Corporate Performance Prediction Use Case Architecture

The core use case of MPAI-CUI is "AI-based Corporate Performance Prediction." Its function is: receiving Prediction Horizon, Governance, Financial Statement, and Risk Assessment as inputs, processing them through five AIMs, and outputting Default Probability, Organisational Model Index, and Business Discontinuity Probability.

AI Module Functions

AIM	Function	Input Data	Output Data
Financial Assessment	Calculate financial features from financial statements	Financial Statements	Financial Features (e.g., 20 items including EBITDA, Current Ratio, etc.)
Governance Assessment	Calculate governance features from governance data	Governance Data	Governance Features (e.g., 10 items including number of shareholders, gender of decision-makers, etc.)
Risk Matrix Generation	Construct risk matrix from risk assessment	Risk Assessment	Risk Matrix (including 4 features for cyber risk and seismic risk)
Prediction	Calculate default probability and organisational model index based on financial and governance features	Financial Features, Governance Features, Prediction Horizon	Default Probability, Organisational Model Index
Perturbation	Calculate business discontinuity probability combining default probability and risk matrix	Default Probability, Risk Matrix	Business Discontinuity Probability

---

Data Formats and Feature Definitions

The standard strictly defines the JSON format for input and output data. For example, financial statements include 20 financial features, such as EBITDA Margin (EBITDA/Revenue) and Quick Ratio ((Current Assets - Inventory)/Current Liabilities), all calculated according to IFRS. Governance features include Number of Shareholders, Gender of Decision-Makers, etc. The risk matrix consists of 2 types of risks (cyber risk, seismic risk) and 4 features (probability of occurrence, business impact, severity, risk retention).

---

Terminology Explanation

• AI Framework (AIF): The runtime environment for executing AI workflows, containing 7 components including access, communication, and controller.
• AI Module (AIM): A data processing unit that receives specific inputs and produces outputs; its internal architecture can be AI or traditional algorithms.
• AI Workflow (AIW): A structured aggregation of multiple AIMs that implements a complete use case.
• Interoperability: The ability to replace AIMs or AIWs with implementations of equivalent functionality.
• Performance: A set of attributes characterizing the reliability, robustness, fairness, and reproducibility of an implementation.

---

Practical Application Case

Suppose a manufacturing enterprise wishes to assess its bankruptcy risk over the next 12 months. The enterprise can provide its financial statements, shareholder structure, board composition, and self-assessment data on cyber risk and seismic risk in accordance with the MPAI-CUI standard. After processing by standardized AI modules, the system will output: Default Probability (e.g., 3.2%), Organisational Model Index (e.g., 0.78, indicating a relatively sound governance structure), and Business Discontinuity Probability (e.g., 1.5%). These indicators can help enterprises identify financial vulnerabilities in advance, optimize governance decisions, or purchase insurance.

---

Standard Development Background and Technical Evolution

MPAI was established in 2019, focusing on the standardization of AI data coding. MPAI-CUI is the first AI application standard targeting industrial data compression and understanding, released as V1.1 in 2021, and subsequently adopted by IEEE as IEEE 3303-2023. This standard fills the gap in the standardization of AI prediction performance in the industrial sector, solving the "black box" problem of previous corporate performance models. By standardizing inputs, outputs, and module functions, it achieves replaceable and auditable AI systems. Future versions may expand to include more vertical risk types or incorporate supply chain data.

---

Implementation Recommendations

When implementing this standard, enterprises are advised to follow these steps:

1. Data Preparation: Collect governance, financial, and risk data according to the JSON schema defined by the standard, ensuring data quality and integrity.
2. Select AIF Implementation: Choose a framework compliant with MPAI-AIF standards and download AIM implementations that have passed conformance testing.
3. System Integration: Embed the AIF into existing IT architecture to ensure smooth data flow.
4. Performance Evaluation: Commission an MPAI-authorized performance evaluation body for Level 3 certification to ensure the reliability of prediction results.
5. Continuous Monitoring: Regularly update data and models, and monitor revisions to the standard.

Notes: The standard provides no guarantees; users must assess compliance and legal risks independently. Data privacy and ownership must be handled in accordance with local regulations.

IEEE 3303-2023 Referenced Document

• ISO 31000 ISO 31000:2018 Risk management A Practical guide

IEEE 3303-2023 history

• 2023 IEEE 3303-2023 IEEE Standard Adoption of Moving Picture, Audio and Data Coding by Artificial Intelligence (MPAI) Technical Specification Compression and Understanding of Industrial Data 1.1

Topics on standards and specifications

Standard and Specification