<-References Software  Go to ToC      AI Modules ->

1. Technical Specification 2. Reference Software 3. Conformance Testing 4. Performance Assessment

1. Technical Specification

Technical Specification: Neural Network Watermarking (MPAI-NNW) – Neural Network Traceability (NNW-NNT) V1.1 assumes that Workflow implementations will be based on Technical Specification: AI Framework (MPAI-AIF) V2.1 that specifies an AI Framework (AIF) where AI Workflows (AIW) composed of interconnected AI Modules (AIM) are executed.

Table 1 provides the full list of AIWs specified by NNW-NNT V1.1 with links to the pages dedicated to each AI Workflow which includes its function, reference model, Input/Output Data, Functions of AIMs, Input/Output Data of AIMs, and links to the AIW-related AIW, AIMs, and JSON metadata.

All NNW-NNT V1.0 specified AI-Workflows are superseded by those specified by V1.1. NNW-NNR V1.0 specification can still be used if it version is explicitly indicated.

Table 1 – AIWs of NNW-NNT V1.1

Acronym. Title JSON
NNW-NTI No Training Imperceptibility X
NNW-WTI With Training Imperceptibility X
NNW-NIR No-Inference Robustness X
NNW-WIR With Inference Robustness X
MMC-AMQ Answer to Multimodal Question X

3. Conformance Testing

An implementation of an AI Workflow conforms with MPAI-MMC if it accepts as input and produces as output Data and/or Data Objects (the combination of Data of a Data Type and its Qualifier) conforming with those specified by MPAI-MMC.

The Conformance of an instance of a Data is to be expressed by a sentence like “Data validates against the Data Type Schema”. This means that:

  • Any Data Sub-Type is as indicated in the Qualifier.
  • The Data Format is indicated by the Qualifier.
  • Any File and/or Stream have the Formats indicated by the Qualifier.
  • Any Attribute of the Data is of the type or validates against the Schema specified in the Qualifier.

The method to Test the Conformance of a Data or Data Object instance is specified in the Data Types chapter.

4. Performance Assessment

Performance is a multidimensional entity because it can have various connotations, and the Performance Assessment Specification should provide methods to measure how well an AIW performs its function, using a metric that depends on the nature of the function, such as:

  1. Quality: the Performance of an Answer to Question Module AIW can measure how well the AIW answers a question related to an image.
  2. Bias: Performance of an Answer to Question Module AIW can measure the quality of responses in dependence of the type of images.
  3. Legal compliance: the Performance of an AIW can measure the compliance of the AIW to a regulation, e.g., the European AI Act.
  4. Ethical compliance: the Performance Assessment of an AIW can measure the compliance of an AIW to a target ethical standard.

The current MPAI-MMC V2.3 Standard does not provide AIW Performance Assessment methods.

<-References Software  Go to ToC      AI Modules ->