SearcharxivSearch

arXiv subjects

Andrea Gentili

Publications and source records attributed to Andrea Gentili.

3 recordsLinked to original sources

Fine-tuning Large Enterprise Language Models via Ontological Reasoning

Large Language Models (LLMs) exploit fine-tuning as a technique to adapt to diverse goals, thanks to task-specific training data. Task specificity should go hand in hand with domain orientation, that is, the specialization of an LLM to accurately address the tasks of a given realm of interest. However, models are usually fine-tuned over publicly available data or, at most, over ground data from databases, ignoring business-level definitions and domain experience. On the other hand, Enterprise Knowledge Graphs (EKGs) are able to capture and augment such domain knowledge via ontological reasoning. With the goal of combining LLM flexibility with the domain orientation of EKGs, we propose a novel neurosymbolic architecture that leverages the power of ontological reasoning to build task- and domain-specific corpora for LLM fine-tuning.

cs.CL

Some remarks on blueprints and ${\mathbb F}_1$-schemes

Over the past two decades several different approaches to defining a geometry over ${\mathbb F}_1$ have been proposed. In this paper, relying on Toën and Vaquié's formalism, we investigate a new category ${\mathsf{Sch}}_{\widetilde{\mathsf B}}$ of schemes admitting a Zariski cover by affine schemes relative to the category of blueprints introduced by Lorscheid. A blueprint, that may be thought of as a pair consisting of a monoid $M$ and a relation on the semiring $M \otimes_{{\mathbb F}_1} \mathbb N$, is a monoid object in a certain symmetric monoidal category $\mathsf B$, which is shown to be complete, cocomplete, and closed. We prove that every $\widetilde{\mathsf B}$-scheme $Σ$ can be associated, through adjunctions, with both a classical scheme $Σ_{\mathbb Z}$ and a scheme $\underlineΣ$ over ${\mathbb F}_1$ in the sense of Deitmar, together with a natural transformation $Λ\colon Σ_{\mathbb Z}\to \underlineΣ\otimes_{{\mathbb F}_1} {\mathbb Z}$. Furthermore, as an application, we show that the category of "${\mathbb F}_1$-schemes" defined by A. Connes and C. Consani can be naturally merged with that of $\widetilde{\mathsf B}$-schemes to obtain a larger category, whose objects we call "${\mathbb F}_1$-schemes with relations".

math.AG

COVID-19 and Company Knowledge Graphs: Assessing Golden Powers and Economic Impact of Selective Lockdown via AI Reasoning

In the COVID-19 outbreak, governments have applied progressive restrictions to production activities, permitting only those that are considered strategic or that provide essential services. This is particularly apparent in countries that have been stricken hard by the virus, with Italy being a major example. Yet we know that companies are not just isolated entities: They organize themselves into intricate shareholding structures --- forming company networks --- distributing decision power and dividends in sophisticated schemes for various purposes. One tool from the Artificial Intelligence (AI) toolbox that is particularly effective to perform reasoning tasks on domains characterized by many entities highly interconnected with one another is Knowledge Graphs (KG). In this work, we present a visionary opinion and report on ongoing work about the application of Automated Reasoning and Knowledge Graph technology to address the impact of the COVID-19 outbreak on the network of Italian companies and support the application of legal instruments for the protection of strategic companies from takeovers.

cs.AI