ALMA project is a collaborative initiative centered around Algebraic Machine Learning (AML), a completely new machine learning approach rooted in algebraic data representations. It offers an alternative to traditional statistical learning by not using parameters and not relying on fitting, regression, backtracking, constraint satisfiability, logical rules, production rules or error minimization.

The primary goal of the EU-funded ALMA project is to harness the unique properties of AML to usher in a new era of interactive, human-centric machine learning systems. These innovative systems are designed to tackle multiple challenges, including reducing bias and preventing discrimination, retaining existing knowledge when learning new information and fostering trust, reliability, and explainability in human-AI interactions. Additionally, they aim to promote distributed, collaborative, and parallel training of newly developed ML systems.

Comprising 8 interdependent Work Packages (WPs), the project aims to pioneer a new generation of human-centric machine learning systems by investigating how to use AML to benefit society. This is done by developing new ML use cases (e.g. medical imaging), applications to robotics, applications for Human-Computer Interaction (HCI), and measuring the impact on ethics and world models.

• WP1 handles administrative management and technical coordination, influencing all WPs with legal, contractual, financial, and administrative activities. It also encompasses progress monitoring, quality control, risk mitigation analysis, and action plan design. WP1's leadership is divided between eProsima, overseeing administrative management, and DFKI, responsible for technical direction.

• WP2 extends research on AML foundations, focusing on generalization, comparison with traditional statistical machine learning, human-AML interaction, and collective learning of multiple AML systems.

• WP6 acts as middleware for efficient data sharing of AML Description Language (AML-DL) compliant data. It produces the AML-IP, a framework for accommodating AML components with intercommunication capabilities over LAN or the Internet. The AML-IP integrates tasks from WPs 2, 3, 4, and 5 and targets open-source tools to work with AML and edge/cloud environments. It also seeks to expand compatibility with robotics and constrained device platforms.

• WP5 and WP3 are closely aligned. WP5 designs techniques for learning world models, ethics, and culture, integrating data-driven learning with expert knowledge and human input. WP3 creates a human-readable language and tools for encoding knowledge, constraints, and task goals in the required AML format.

• WP4 leverages collective learning and human-AML interaction to design new interfaces and evaluation methods, ensuring effective interaction with intelligent systems. It explores the user-system relationship to develop a working and interactive interface.

• WP7 serves as an overarching layer, creating a testbed to demonstrate human-machine interaction with AML across various use cases, such as image classification, intelligent, supportive tools, and higher-level cognition for domestic assistance. 

• WP8 is responsible for disseminating and exploiting AML, including designing a communication strategy, planning scientific workshops, and contributing to standards development. 

In summary, the ALMA project is poised to advance AML-based machine learning systems that prioritize the explainability of how Machine Learning systems learn and evolve through its diverse Work Packages.

Find all the information on the AML-IP v0.1.0 in the release notes.

MORE INFORMATION ABOUT ALMA:

For any questions, please get in touch with This email address is being protected from spambots. You need JavaScript enabled to view it..