EXplainable ALgorithmic Tools

Información del proyecto

EXALT

Identificador del acuerdo de subvención: 101082299

DOI

10.3030/101082299

Proyecto cerrado

Fecha de la firma de la CE 31 Agosto 2023

Fecha de inicio 1 Septiembre 2023

Fecha de finalización 28 Febrero 2025

Financiado con arreglo a

• European Research Council (ERC)

Coste total

Sin datos

Aportación de la UE

€ 150 000,00

Coordinado por

MIM.AI SPOLKA Z OGRANICZONA ODPOWIEDZIALNOSCIA
Poland

Periodic Reporting for period 1 - EXALT (EXplainable ALgorithmic Tools)

Período documentado: 2023-09-01 hasta 2025-02-28

Resumen del contexto y de los objetivos generales del proyecto

Deploying algorithmic solutions in real-world applications raises two challenges. First, we need easy-to-use and universal algorithms. Second, we need to guarantee that algorithmic solutions can be understood by people using them. We address the first of these challenges in TUgbOAT project, which aims to deliver unified algorithmic tools. Here, we propose to develop tools that would address the second of these challenges. In many use scenarios, algorithms propose a solution to a human operator. The main challenge in such cases is to convince him to use the returned solution. Traditionally we think of algorithms in a black-box manner, i.e. as a tool to find a good solution.
We do not expect algorithms to give a human-understandable explanation of why this is the best solution, or what alternatives exist or what are the bottlenecks. Nevertheless, we humans still tend to ask these questions even if we understand the algorithms that are used. Currently, we lack good tools that could explain the results of optimization algorithms, e.g. for the assignment problem. For practitioners, like ourselves, that work together with companies to deploy algorithmic solutions in real-world cases, the need to provide explainable algorithms becomes immanent. Here we will test and implement results developed in TUgbOAT that can be used to complement the algorithms with human explanations. In particular, we plan to: - enrich algorithms to give meaningful alternative solutions, - apply Shapley value methods to determine key solution elements, - work with perturbed inputs to create robust and more concise solutions, - generate concise decision trees that would explain steps taken by algorithms. This project aims to deliver the base parts of a software library that would give explainable algorithms. We plan to concentrate on the task assignment problem (i.e. matchings) where we already cooperate with companies.

Trabajo realizado desde el comienzo del proyecto hasta el final del período abarcado por el informe y los principales resultados hasta la fecha

Task A. Creation of EXALT Library
Goal A.1 Design of the library. - The EXALT Library bridges the gap between optimization and explainability, enhancing transparency and trust in algorithmic decisions. By focusing on alternative solutions, game-theoretic insights, robustness, and interpretability, EXALT is positioned to become a powerful tool for modern businesses. This milestone report details a practical path toward realizing these goals.
Goal A.2 Implementation of the library. - The successful implementation of these explainable optimization algorithms, along with the robust explanation module, demonstrates the readiness of the EXALT Library to address real-world challenges. This milestone significantly enhances the library's utility and adoption potential across various industries.
Task B. Validating our Ideas on Synthetic Datasets
Goal B.1 Preparation of Synthetic Datasets - The datasets are structured to reflect diverse application scenarios:
• Variability in Features: Examples include different feature sets tailored to specific analytical goals, such as time-series clustering or variable-length sequence analysis.
• Controlled Noise and Outliers: To evaluate robustness, datasets include controlled levels of noise and anomalies, ensuring that models can handle real-world imperfections.
• Balanced Representation: Examples are balanced to avoid biases and ensure fair evaluation of algorithms.
This structured approach to dataset creation and processing ensures that every module in the library, from preprocessing to explainability, is thoroughly validated. It also provides a reproducible and extensible foundation for scaling the project to new datasets and applications.
Goal B.2 Validating Our Ideas - The milestone has been successfully achieved by executing each optimization algorithm and testing them on their respective datasets. Each algorithm was implemented and validated to ensure they perform as expected. Finding the best solution, including a choice of algorithm and its best parameters will be performed by a partner for a real dataset.

Avances que van más allá del estado de la técnica e impacto potencial esperado (incluida la repercusión socioeconómica y las implicaciones sociales más amplias del proyecto hasta la fecha)

Key results achieved include:

Advanced Explainability for Clustering Assignments: The Explainability Module integrates SHAP, LIME, and decision tree-based models, offering both global and instance-level interpretability of clustering outcomes.
Comprehensive Clustering Quality Assessment: The Clustering Evaluation Module within EXALT provides an automated, structured framework for assessing cluster validity using statistical measures and visualization techniques.
Scientific Contribution: The research findings were consolidated into a scientific paper, contributing to the broader AI and machine learning community by establishing best practices for clustering explainability and evaluation.

1.Scientific and Technical Impact
Enhanced Model Transparency & Trust: The integration of explainability techniques ensures that clustering models can be understood by both technical and non-technical stakeholders, fostering greater trust in AI-driven segmentation.
Improved Clustering Performance Assessment: By embedding a rigorous evaluation framework within EXALT, organizations can quantitatively assess clustering quality, leading to better data-driven decision-making.

2.Industry and Business Impact
Informed Business Decision-Making: By making clustering results interpretable and integrating expert-driven validation, businesses can use the framework for customer segmentation, market analysis, and personalized recommendations.
Increased Adoption of AI in Business Processes: The EXALT framework’s interpretability and quality assessment capabilities make AI-driven clustering more accessible to companies hesitant to adopt black-box models.

3.Societal and Ethical Impact
Fairness and Accountability in AI Systems: Explainable AI helps mitigate biases in clustering algorithms by allowing stakeholders to understand and challenge model decisions.
Regulatory Compliance and Ethical AI: By integrating explainability and evaluation mechanisms, EXALT aligns with emerging AI regulations and ethical guidelines in Europe (e.g. EU AI Act).

The project improved clustering transparency, dependability, and usability by effectively delivering a suite of explainability and assessment tools within the EXALT framework. Robust performance evaluation was made possible by automated clustering evaluation measures, and interpretability was enhanced by the combination of SHAP, LIME, and decision tree-based approaches. The use of AI-based clustering solutions in commercial and scientific applications is anticipated to increase as a result of these developments. Key initiatives include ongoing research, practical demonstrations, market expansion, commercialization initiatives, and regulatory alignment will be necessary for wider adoption in order to optimize the impact and guarantee long-term success.