arXiv subjects

Francesco Carella

Publications and source records attributed to Francesco Carella.

3 recordsLinked to original sources

arXiv2026

Towards Dependable Retrieval-Augmented Generation Using Factual Confidence Prediction

Incorporating specific knowledge into large language models via retrieval-augmented generation (RAG) is a widespread technique that fuels many of today's industry AI applications. A fundamental problem is to assess if the context retrieved by some similarity search provides indeed supporting facts, or instead misguides the generator with irrelevant information. It is critical to associate meaningful confidence measures about the factuality of the retrieval process with the generated answers. We present a new, two-staged approach to predict fact faithfulness of the output of retrieval-augmented generations. First, we employ conformal prediction to select only those retrieved chunks who have a high chance to come from the correct source. This approach in itself can improve answer quality by up to 6% in some of the studied datasets, however, the associated statistical guarantees do not hold generally, since the assumption of sample exchangeability depends on the retriever setup. We present diagnostic metrics to assess whether a setup is suitable. Second, we quantify confidence in the consistency of a generated final answer with a given retrieved context, using an attention-based factuality classifier. This approach can detect inconsistent answers with a chance of up to 77%. Our work helps to establish a novel type of certified RAG systems for a broad range of natural language industry applications.

cs.IR↗

arXiv2025

Clustering Wind data at 1 AU to contextualize magnetic reconnection in the solar wind

Context. Magnetic reconnection events are frequently observed in the solar wind. Understanding the patterns and structures within the solar wind is crucial to put observed magnetic reconnection events into context, since their occurrence rate and properties are likely influenced by solar wind conditions. Aims. We employed unsupervised learning techniques such as self-organizing maps (SOM) and K-Means to cluster and interpret solar wind data at 1 AU for an improved understanding of the conditions that lead to magnetic reconnection in the solar wind. Methods. We collected magnetic field data and proton density, proton temperature, and solar wind speed measurements taken by the Wind spacecraft. After preprocessing the data, we trained a SOM to visualize the high-dimensional data in a lower-dimensional space and applied K-Means clustering to identify distinct clusters within the solar wind data. Results. Our analysis revealed that the reconnection events are distributed across five different clusters: a) slow solar wind, b) compressed slow wind, c) highly Alfvénic wind, d) compressed fast wind, and e) ejecta. Compressed slow and fast wind and ejecta are clusters associated with solar wind transients such as stream interaction regions and interplanetary coronal mass ejections. The majority of the reconnection events are associated with the slow solar wind, followed by the highly Alfvénic wind, compressed slow wind, and compressed fast wind, and a small fraction of the reconnection events are associated with ejecta. Conclusions. Unsupervised learning approaches with SOM and K-Means lead to physically interpretable solar wind clusters based on their transients and allow for the contextualization of magnetic reconnection exhausts' occurrence in the solar wind.

astro-ph.SR↗

arXiv2025

Towards Automated Safety Requirements Derivation Using Agent-based RAG

We study the automated derivation of safety requirements in a self-driving vehicle use case, leveraging LLMs in combination with agent-based retrieval-augmented generation. Conventional approaches that utilise pre-trained LLMs to assist in safety analyses typically lack domain-specific knowledge. Existing RAG approaches address this issue, yet their performance deteriorates when handling complex queries and it becomes increasingly harder to retrieve the most relevant information. This is particularly relevant for safety-relevant applications. In this paper, we propose the use of agent-based RAG to derive safety requirements and show that the retrieved information is more relevant to the queries. We implement an agent-based approach on a document pool of automotive standards and the Apollo case study, as a representative example of an automated driving perception system. Our solution is tested on a data set of safety requirement questions and answers, extracted from the Apollo data. Evaluating a set of selected RAG metrics, we present and discuss advantages of a agent-based approach compared to default RAG methods.

cs.AI↗