PNRR M4C2 I1.3 Partenariato Esteso FAIR - Bando a Cascata SPOKE 8 - “Pervasive AI” - riservato a Imprese

I progetti dovranno indicare uno degli obiettivi individuati :

Objective #1: Classification, identification and prediction of rare events in time series

Description of the objective:

Many advances have been done in applications like anomaly detection, classification and prediction, studied in WP8.1. However, quite often these events are rare and evidence of these events in the data set of time series is relatively small compared to traces of “normal” behaviour. We plan to explore in this project, cases in which the events to classify, detect and predict are extremely rare and not represented in data. We will follow two approaches, one based on data augmentation and one on autoencoders and anomalous reconstruction error detection.

Contribution: 171.263,90 €

Objective #2: Make AI pervasive in space

Description of the objective:

Space applications are characterized by the deployment of devices in harsh and remote environments in which the greater the behavior intelligence the larger the probability of mission success. This objective aims at contaminating conventional methods for space platform operations with AI to grant a larger level of self management to spaceborne systems.

Fundamental problems in AI design must be addressed originating from the limited computational resources available and the small bandwidth between spaceborn devices and ground facilities. At the same time, as it is recognized that more than half of a long-term space mission costs are in keeping the platform operative, solving such fundamental problems may substantially ease access to space economy.

Contribution: 171.263,90 €

Objective #3: Measuring the unmeasurable

Description of the objective: Follow the provoking Galileo’s prescription and exploit nowadays capability of hyper-sensorizing accessible systems to make inaccessible systems also measurable. The physics of interacting systems make phenomena interlinked at every level. Intelligence can be used to capture interrelations and infer unmeasurable quantities from redundant measured ones.

Bringing this “sensor virtualization” to the extreme poses significant fundamental challenges but may also be the key to substantial advancements in many practical applications from management of industrial plants, to monitoring of large and/or distributed infrastructure, to rescue activity in harsh environments.

Contribution: 171.263,90 €

Objective #4: 3D reconstruction of objects and spaces for interactive navigation and anomaly prediction

Description of the objective:. 3D generative models and 3D reconstructing approaches, such as the ones NERF-based, are the state-of-the art of reconstructive Computer vision and offer challenging possibilities for exploiting such a knowledge in many application domains. Examples are interactive Robot navigation, and large object reconstruction. This can be used in industry for maintenance problems and in assistive home robotics for interactive exploration- as living digital twins. 3D computer vision, also enriched by visual and language navigation (VLN) results could be absolutely disruptive for the new generation of collaborative and interactive AI-based robotics. The objective is not only to define the use case but to exploit and refine state-of-the art results for addressing both efficiency and accuracy in the results.and to collaborate with FAIR also hosting grats and Phd works.

Contribution: 200.510,04 € RESERVED TO SOUTH

Objective #5: Develop domain specific Lage multimodal models for document understanding, for financial and Public institutions

Description of the objective:

Multimodal visual and language models are becoming profitable for interacting with users in many specific domains: among them understanding multimodal documents based on texts, images, graphs and numeric tables are still cumbersome, if developed with open source facilities, possibly in multilingual and Italiaan. The goal of objective is to collaborate with the TP VLMC on specific use case of document interaction, Visual question & answering, summarization considering both the visual understanding and the textual interpretation as well as Retrieval facilities. The solution should follow the criteria of European definition for trustworthy AI. Finally solutions should incorporate Domain knowledge in implicit or explicit form.

Contribution: 171.263,90 €

Objective #6: Integration of data-driven and knowledge based models for logistics

Description of the objective:

Logistic applications are complex and appear in many industrial sectors. These applications have been traditionally solved by building knowledge based systems by interacting with domain experts. Recent advances in machine learning enable us to use data to integrate the model with data-driven components. In general, the language in which these two models are expressed are quite different. Knowledge based systems are in general expressed in terms of symbolic, human-understandable terms, while data-driven models are expressed as neural networks, namely huge mathematical models. Integrating these two languages requires the development of new methodologies and techniques that will be applied to logistic applications.

Contribution: 171.263,90 €

Objective #7: Generative AI in the Audiovisual Industry

Description of the objective:

The project should contribute to developing and applying new generative AI–driven production models for the audiovisual sector. Generative AI should be employed to create music, audio and video, integrating it with existing sound and image production as well as post-production technologies. The main objective is to experiment and standardize changes and innovations brought by generative AI to the production processes, training traditional and new professional figures. The new production model will be assessed by producing audiovisual content.

The project should deliver the following expected outcomes:

• Empowering the engaged territory, through lasting skill transfer and professional training.

• Developing specific tools to share methodologies, implementation schemes, and production standards.

• Providing concrete applications of the new model with one or more audiovisual products as case studies.

Contribution: 160.000,00 € RESERVED TO SOUTH