RESOURCES
DISSEMINATION MATERIALS:
4. NEWSLETTER #4 (December 2022)
- XAI in Manufacturing
- Hybrid ML Explainability
- Draft Catalogue of XMANAI AI and Graph Machine Learning Models
- Dissemination and Project Activities
Available for download here.
3. NEWSLETTER #3 (July 2022)
- The Value of Meaningful Data
- Towards a Common Data Model
- XMANAI Graph Data Model
Available for download here.
5. NEWSLETTER #2 (November 2021)
Driven by our key message “Moving from ‘black box’ to ‘glass box’ Artificial Intelligence in Manufacturing”, the project has produced some interesting progress during the second semester. Check out our second newsletter with sections on:
- Key Stakeholders
- A Look into our Pilots
- Requirements Elicitation
- XMANAI Minimum Viable Product (MVP)
- Dissemination and Collaboration
Available for download here.
4. XMANAI ROLL-UP
XMANAI roll-up to present the overall project concept and objectives. In 2022 it is expected that the covid-19 pandemic allows, at least in a certain degree, the return to the physical/hybrid interaction with the community where the roll-up can be used.
XMANAI focuses on explainable AI, a concept that contradicts the idea of the ‘black box’ in machine learning. Carving out a ‘human-centric’ trustful approach tested in industrial demonstrators, the project aims to transform the manufacturing value chain with ‘glass box’ models explainable to a ‘human in the loop’.
Available for download here.
3. XMANAI TRIFOLD
The trifold is an efective means to disseminate the project both on physical and online events, summarising not only the concept, factsheet, and objectives but also the main results available to the moment. Within the trifold you will find a brief presentation of the pilots and the high level architecture, highlighting the core XMANAI services.
Available for download here.
2. NEWSLETTER #1 (May 2021)
Driven by our key message “Moving from ‘black box’ to ‘glass box’ Artificial Intelligence in Manufacturing”, the project has produced some interesting progress during the first six months. Check out our first newsletter with sections on:
- Project Objective
- Meet Our Team
- XMANAI Concept and Approach
- XMANAI Scientific Workshop
- Collaboration
Available for download here.
1. XMANAI FLYER
XMANAI consists of 15 partners from 7 countries Italy, Germany, Spain, Estonia, Portugal, Greece and Cyprus. The consortium is very well balanced in terms of research-industry collaboration as the next table depicts, containing a very well though constructed mixture of collective expertise from industry, research, academia, technology providers with solid background on manufacturing, data science and AI, and big data, and human-machine ethics sectors.
Available for download here.
PUBLICATIONS:
“Moving from “”black box”” to “”glass box”” Artificial Intelligence in Manufacturing with XMANAI”
Artificial Intelligence (AI) is finding its way into a broad range of industries, including manufacturing. The decisions and predictions that can be potentially derived from AI-enabled systems are becoming much more profound, and in many cases, critical to success and profitability. However, despite the indisputable benefits that AI can bring in society and in any industrial activity, humans typically have little insight about AI itself and even less concerning the knowledge on how AI systems make any decisions or predictions due to the so-called “black-box” effect. This paper presents the XMANAI approach, that focuses on explainable AI models and processes, to mitigate such an effect and reinforce trust. The aim is to transform the manufacturing value chain with ‘glass box’ models that are explainable to a ‘human in the loop’ and produce value-based explanations for data scientists, data engineers and business experts.
Available for download here.
gLIME: A NEW GRAPHICAL METHODOLOGY FOR INTERPRETABLE MODEL-AGNOSTIC EXPLANATIONS
This paper contributes to the development of a novel graphical explainability tool that not only indicates the significant features of the model, but also reveals the conditional relationships between features and the inference capturing both the direct and indirect impact of features to the models’ decision. The proposed XAI methodology, termed as gLIME, provides graphical model-agnostic explanations either at the global (for the entire dataset) or the
local scale (for specific data points). It relies on a combination of local interpretable model-agnostic explanations (LIME) with graphical least absolute shrinkage and selection operator (GLASSO) producing undirected Gaussian graphical models. Regularization is adopted to shrink small partial correlation coefficients to zero providing sparser and more interpretable graphical explanations. Two well-known classification datasets (BIOPSY and OAI) were selected to confirm the superiority of gLIME over LIME in terms of both robustness and consistency/sensitivity over multiple permutations.
Specifically, gLIME accomplished increased stability over the two datasets with respect to features’ importance (76%-96% compared to 52%-77% using LIME). gLIME demonstrates a unique potential to extend the functionality of the current state-of-the-art in XAI by providing informative graphically given explanations that could unlock black boxes.
Available for download here.
PUBLIC DELIVERABLES:
D1.1: State of the Art Review in XMANAI Research Domains
D1.1 provides an overview over the Explainable Artificial Intelligence (XAI) domain and it is structured as a collection of existing methods and algorithms, that can be used as inspiration for developing the XMANAI methodology.
It presents the state-of-the-art on XAI and machine learning from a theoretical perspective, comparing different methods and tools. Furthermore, it includes the analysis of open-source solutions for Artificial Intelligence (AI) implementation and a look at XAI applications in the industry.
To provide a complete overview, the analysis is complemented by the exploration of human aspects in decision making and AI.
Available for download here.
D1.2: XMANAI Concept Detailing, Initial Requirements, Usage Scenarios and Draft MVP
This deliverable aims at bringing together the XMANAI concept by: (a) brainstorming on different user journeys in Explainable AI for business users, data scientists and data engineers, (b) eliciting the backlog of technical requirements and aligning them with the business requirements and the user journeys, (c) obtaining some early perspectives on the available manufacturing data (from the XMANAI demonstrators and open data sources), and (d) consolidating the Minimum Viable Product (MVP) that summarizes the expected features on which XMANAI shall focus (by the end of the project) for maximizing the expected added value to manufacturers while ensuring innovation from a scientific and technical perspective.
Available for download here.
D2.1: Asset Management Bundles Methods and System Designs
Deliverable 2.1 deals with the architectural design for the asset management layer of the overall XMANAI Platform. This layer has the central task of importing/extracting data from external data sources (i.e. legacy and operational manufacturing systems) and ensuring data explainability in order to make them available for running AI pipelines.
In order to specify the asset management-related services, a detailed state-of-the-art analysis was performed.
On the one hand, this includes all necessary methods for the execution of all asset management and sharing processes. And on the other hand, current technologies for fulfilling the XMANAI requirements were examined.
Based on these results, a detailed architecture for the management of industrial assets and AI models was designed and accompanied by the selection of technologies and the elaboration of mock-ups to demonstrate the expected user interactions.
Available for download here.
D3.1: AI Bundles Methods and System Designs
This deliverable reports on the results produced through WP3 activities towards the delivery of the
XMANAI Artificial Intelligence (AI) bundles designs and methods. It presents insights gained through the landscape analysis on research dimensions and technical advancements in the broader scope of AI pipelines and reflects upon the positioning of the XMANAI solution targeting explainable AI pipelines in the manufacturing domain. In this scope, the role of data models is explored and relevant standards are studied. The deliverable reports on the development of the XMANAI data model and presents its early concepts, relationships, and foreseen lifecycle management mechanisms. Finally, the detailed WP3 architecture design is discussed and each of its components is specified, including designed functionalities and offered methods, considered technologies for its implementation and indicative mockups.
Available for download here.
D5.1: System Architecture, Bundles Placement Plan and APIs Design
This deliverable provides the design of the XMANAI reference architecture by: (a) designating the
architecture blueprints across the tiers, services bundles, components and application perspectives, (b) designating the core workflows along the user journeys in Explainable AI for business users, data scientists and data engineers, for the interaction among the different XMANAI components, (b) defining the core functionalities, the mapping to the XMANAI technical requirements and MVP features, and the main interactions of the XMANAI components that shall be delivered in the XMANAI Centralized Cloud and OnPremise (Private Cloud) installations, (c) obtaining some early perspectives on the features of the XMANAI manufacturing apps and their alignment with the business requirements of the XMANAI demonstrators.
Available for download here.
D8.2: Project Website and Communication Channels Instantiation
D8.2 is the first result of XMANAI task T8.3 “Communication Activities and Publicity” which is responsible to construct the project’s website from the very early implementation stages. It also reports the set-up of all needed Web 2.0 and social channels that will be used during the project for communication. This document is accompanying the launch of the XMANAI website and Social Media channels which actually compose the deliverable.
Available for download here.
D9.2: Ethics and Data Management Plan
This document outlines the main elements of the data management policy with regard to all research datasets that will be generated during the pilots’ execution and the efficient management of publications will be agreed and followed by the Consortium, in accordance with the H2020 guidelines regarding Open Access to Scientific Publications and Research Data.
Available for download here.
