Distributed Ledger Audit Trail for Explainable ML Results
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models lack explainability and reproducibility, leading to interpretability issues and the inability to validate decisions over time, especially in complex AI systems, which can result in undesirable classification pathways and non-compliance with anti-discrimination regulations.
Innovation Solution
Implementing a distributed ledger system to track and audit machine learning model results by creating blocks that include input data, initial results, model data structures, and training data links, enabling reproducible and explainable outcomes through blockchain technology.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models are used to process input data and generate outputs, then the model can learn from training corpus and improve performance day-by-day, but the model becomes opaque and loses interpretability
Solution Approach 1:
The system performs preliminary actions by capturing and storing metadata mapper objects, input data, and model outputs at the time of processing. This creates an immutable audit trail that preserves the state of the model and data at each decision point, enabling later explanation and validation without affecting the model's continuous learning capability.
Solution Approach 2:
The distributed ledger acts as an intermediary between the machine learning model and the stakeholders. It stores and provides access to metadata mapper objects, input data, and outputs, serving as an explainable interface that allows humans to understand model decisions while the model continues to operate autonomously.
2Productivity
If the ML model matures by appending metadata to training corpus, then the model learns from historical data, but the same input data generates different outputs over time making validation difficult
Solution Approach 1:
The system captures the complete state of the model including metadata mapper objects, training corpus references, and processing parameters at the time of each prediction. This preliminary recording enables exact reproduction of past outputs by retrieving and re-executing with the same input data and captured model state.
Solution Approach 2:
The distributed ledger creates immutable copies of the model state, input data, and outputs at each point in time. These copies serve as reproducible snapshots that can be retrieved and validated later, ensuring that the same input data produces the same output when processed with the captured model state.
3Reliability
If intermediate outcomes are saved for validation proofs, then past results can be audited, but the size of model outcomes and training corpus becomes unmanageable
Solution Approach 1:
The system extracts only the essential information needed for validation and storage in the distributed ledger: metadata mapper objects, key input data references, and model outputs. The full training corpus and intermediate processing details are referenced rather than stored, reducing data volume while maintaining auditability.
Solution Approach 2:
The distributed ledger serves multiple functions simultaneously: it stores audit trails, provides reproducible model states, enables validation proofs, and maintains references to training data. This multi-functionality consolidates what would otherwise require separate storage systems into a single manageable structure.
Data Source
AI summary
Providing reproducible machine learning model results by receiving input data for a machine learning (ML) model, processing the input data using the ML model, yielding an initial result, adding a first block to a distributed ledger, the block comprising the input data, the initial result, an ML model data structure, and a link to training data for the ML model, wherein the training data resides in previous distributed ledger blocks, and providing an output including the initial result.