Opportunity: ASSESSMENT OF OPEN-SOURCE LARGE LANGUAGE MODELS WITHIN THE SATCOM SECTOR (ARTES FPE 1A.128)

  • Closing date 27-09-2024

Background

Artificial Intelligence-powered Natural Language Processing (NLP) models are revolutionising the way we operate and function within our society. Large Language Models (LLMs) based on these technologies enable us to process and comprehend vast amounts of information quickly and accurately. The space sector is no exception to the current era of digital overload and the exponential growth in the amount of complex technical and legal documentation that the sector needs to deal with, makes the use of LLMs, highly attractive. Examples of the tasks and capabilities that could be supported by LLMs include:

  • Summarising Documents: LLMs can be trained to summarise technically and legally complex documentation, allowing for quicker extraction of critical information from lengthy documents.
  • Advanced Search Functions: LLMs have the capability to analyse vast volumes of historical data to identify patterns and similarities, assisting in decision-making processes and facilitating the identification of missing features in future developments or proposals.
  • Smart Conversational Agent (i.e. ChatBot): LLMs can be integrated into chatbot interfaces to provide conversational assistance, potentially helping in the creation and refinement of new Satcom-related products and services, supporting brainstorming of new mission concepts, answering highly technical and legal queries, or providing expert guidance in any domain.

Furthermore, LLMs could be hosted on-board communication satellites for task planning and decision-making for applications where LLMs could perform tasks that require high-level reasoning and/or intuition.

Undeniably, several commercial initiatives have made significant strides in advancing the capabilities of language models and European industry is already making use of these algorithms daily. It is, however, important to consider the potential commercial and security risks associated with using these proprietary LLMs when applying them to sensitive domains such as the space sector: there are concerns regarding the unclear confidentiality and data protection policies surrounding these models.

Open-source LLMs have also experienced rapid evolution in recent times, making them potential alternatives to proprietary systems. These models offer the key advantage of data protection (as they can be safely stored locally in computers within each company), as well as offering transparency, flexibility, and community- driven development. The proposed activity therefore aims to generate a Satcom- specific open-source LLM for use within the participating state Satcom sector.

 

Key Focus Area

The main goal of the proposed activity would be the creation of a SatCom Virtual Expert/ Chatbot assistant. The activity will include selection of the best open-source LLM to use as a starting point; enhancement of the base performance of the model by fine-tuning (i.e. training) with Satcom domain-specific data; and comparison of its enhanced performance against state-of-the-art commercial models. Finally, the activity will generate recommendations and advice on how organisations and companies within the European Satcom community can further train/fine-tune the model with their own proprietary data for their own (private) use.

The activity is part of the Advanced Research in Telecommunications System (ARTES) Future Preparation programme line. This programme element is the beginning of the ARTES ‘feeding chain’ that offers the possibility to acquire knowledge on future Satcom market perspectives, investigate future system concepts and prepare initial ‘dossiers’ on strategic initiatives; that cannot be developed usually at every Member State’s level. It is based on the concept of a European common effort to produce quality results to set the future of SatCom.

 

What We Offer

ESA will fully fund this opportunity and will provide oversight and Satcom specific technical support.

 

What We Are Looking For?

We are looking for experts in Machine Learning, with demonstrated hands-on experience working with, developing and fine-tuning Large Language Models, and with knowledge about SatCom.

 

Who Can Apply?

This call is open to industry within the following ESA Member States: Austria (AT), Belgium (BE), Canada (CA), Switzerland (CH), Czechia (CZ), Germany (DE), Denmark (DK), Estonia (EE), Greece (GR), Spain (ES), Finland (FI), France (FR), United Kingdom (GB), Hungary (HU), Ireland (IE), Italy (IT), Lithuania (LT), Luxembourg (LU), the Netherlands (NL), Norway (NO), Poland (PL), Portugal (PT), Romania (RO), Sweden (SE), Slovenia (SI).

 

How To Apply

Proposals can be submitted via esa-star (please find link below) until 27/09/2024 13:00 CET
https://esastar-publication-ext.sso.esa.int/ESATenderActions/details/90625