Questa è la nostra ultima selezione di pubblicazioni e brevetti in inglese su Large Language Models (LLM), tra numerose riviste scientifiche online, classificate e incentrate su large language model, LLM, generative pre-trained transformer, pre-training, transformer architecture, gradient descent, GPT, tokenization, generative model, self-attention mechanism, masked language model e MLM.
Large language model-based systems and methods for automating complex query processing in infrastructure graph databases
Patent published on the 2026-10-15 in US under Ref US20260320050 by JPMORGAN CHASE BANK N A [US] (Gourgoulias Konstantinos [gb], Yeghiazarian Rozie Tereza [us], Guest William Joel [us], Ibraimoski Senad [gb], Peikon Bruce [us], Ghalyan Najah [us], Moran Sean [gb])
Abstract: [0000] A method may include: receiving, by a computer program and from a user electronic device, a user input comprising a user request to a graph database; interpreting, by the computer program and using a large language model (LLM), the user request in context of the graph database; identifying, by the computer program and using the LLM, a relevant graph query for the interpreted user request; executing, by the computer program, the relevant graph query on the graph database; transforming, by [...]
Our summary: The method automates complex query processing in graph databases using a large language model. It interprets user requests, identifies relevant graph queries, and executes them. The system transforms output into tabular data and generates summaries for user retrieval.
large language model, query processing, graph database, automation
Patent
Systems and methods for prompt-based threat model generation
Patent published on the 2026-10-15 in US under Ref US20260320199 by THREATMODELER SOFTWARE INC [US] (Agarwwal Anuraag [us], Nagalia Nikunj [ae])
Abstract: [0000] A prompt-based threat modeling technique uses stored prompt templates to generate threat models using an LLM. Architecture data describing the target computing environment is used to construct a context-inserted prompt, which is transmitted to a large language model. The large language model provides a response to the context-inserted prompt, which is parsed to extract a plurality of components associated with the target computing environment, one or more relationships between the plurali[...]
Our summary: A prompt-based threat modeling technique utilizes stored templates to generate models using a large language model. It constructs a context-inserted prompt from architecture data of the target environment. The model is generated from parsed responses and stored as structured threat records.
threat modeling, prompt templates, large language model, context-inserted prompt
Patent
Text-to-animation generation
Patent published on the 2026-10-15 in US under Ref US20260320890 by SNAP INC [US] (Zhou Bing [us], Guo Chuan [us], Wang Jian [us])
Abstract: [0000] Examples relate to systems and methods for generating an animation from a text prompt. The systems and methods receive a text prompt describing a desired animation and expand the text prompt using a large language model (LLM) to generate a detailed motion description that includes specific movements of different skeletal joints. The systems and methods generate, using a diffusion-based transformer model conditioned on the detailed motion description, a latent vector representation of a mo[...]
Our summary: The system generates animations from text prompts by expanding them into detailed motion descriptions. It utilizes a diffusion-based transformer model to create a latent vector representation of the motion sequence. Finally, it decodes this representation into animation data and applies post-processing to enhance the final output.
text-to-animation, large language model, motion description, diffusion-based transformer
Patent
Personalized virtual layout for collaborative metaverse sessions
Patent published on the 2026-10-15 in US under Ref US20260320944 by INT BUSINESS MACHINES CORPORATION [US] (Polgrean Heather Nicole [us], Majdabadi Hamid [ca], Rakshit Sarbajit Kumar [in], Garcia Delgado Carolina [mx])
Abstract: [0000] In some implementations, a computer may obtain sensing information indicating real-time information associated with a physical space of a user. The computer may obtain a participant profile of the user, where the participant profile indicates at least one of a role, a responsibility, or a mobility type of the user for a metaverse collaboration session. The computer may generate, using a generative model, a personalized virtual layout within the metaverse collaboration session based on the[...]
Our summary: A computer generates a personalized virtual layout for metaverse collaboration sessions using real-time sensing information and participant profiles. The layout is customized based on user roles, responsibilities, and mobility types. This arrangement includes virtual objects tailored to specific constraints from the user s environment and profile.
virtual layout, metaverse, collaborative sessions, generative model
Patent
Hypernetworks for content recontextualization
Patent published on the 2026-10-15 in US under Ref US20260320860 by EBAY INC [US] (Balicki Jakub Adam [pl], Caudevilla Torras Oriol [us], Drozdz Tomasz [pl], Karanowski Konrad [pl], Lorek Pawel [pl], Lyu Hong [us], Skorupa Aleksander Piotr [pl], Tomczak Jakub Mikolaj [nl], Trzcinski Tomaz [pl], Zieba Maciej [pl])
Abstract: [0000] Some aspects relate to technologies for content recontextualization using hypernetworks to generate Low-Rank Adaptation (LoRA) parameters for adapting a generative model. In some aspects, content recontextualization involves accessing a content item and recontextualization text, generating LoRA parameters by processing the content item with a hypernetwork model, and creating a prompt from the recontextualization text. The generative model, adapted with the LoRA parameters provided by the [...]
Our summary: Hypernetworks are utilized to generate LoRA parameters for adapting generative models. The process involves accessing content and generating prompts from recontextualization text. Training the hypernetwork model requires a dataset of samples for effective parameter updates while keeping the generative model parameters frozen.
Hypernetworks, content recontextualization, Low-Rank Adaptation, generative model
Patent
Machine-learning framework for generating records from user-generated content
Patent published on the 2026-10-15 in US under Ref US20260320112 by ANCESTRY COM OPERATIONS INC [US] (Reese Jack [us], Archibald Taylor Neil [us], Chen Siteng [us], Subraveti Suraj [us], Fabiano Maria Antonia [us], Murahari Kalyan Chakravarthi [us], Paradkar Vishwajeet Vijay [us])
Abstract: [0000] The present disclosure relates to systems, non-transitory computer-readable media, and methods for processing user-generated genealogical content. In particular, in one or more embodiments, the disclosed systems receive digital images reflecting user-generated genealogical-content items and utilize a content-classification model to generate record classifications and route the digital images to one or more specialized processing pipelines to extract genealogical data from the digital imag[...]
Our summary: The system processes user-generated genealogical content by classifying digital images and extracting genealogical data. It employs an ensemble of models for classification and data extraction. Additionally, layout analysis is performed to optimize the reading order of the digital images.
machine learning, genealogical data, content classification, digital image processing
Patent
A large language model-based agent system for interactive large-scale project scheduling
Published on 2026-09-14 by @OXFORD
Abstract: AbstractAdvanced optimization techniques generate highly complex schedules for large-scale engineering projects. While large language models have recently emerged as a transformative alternative for project scheduling, two primary hurdles must be overcome to realize a high-precision framework: performance degradation in large-scale projects and the functional separation between planning and analysis. To address these challenges, we propose an on-premises autonomous large language model agent sys[...]
Our summary: This paper presents an autonomous agent system utilizing large language models for interactive project scheduling. It addresses performance issues and functional separation through a LangGraph-based framework. The proposed system achieves approximately 90% accuracy in large-scale scheduling by integrating retrieval-augmented generation with analytical tools.
large language models, project scheduling, optimization techniques, knowledge graph
Publication
Provenance grounds trust in autonomous science
Published on 2026-08-20 by Robert MacKnight, Ivan M. Novitskiy, Rushik Radadiya, Gabe Gomes @NATURE
Abstract: Nature Computational Science, Published online: 20 August 2026; doi:10.1038/s43588-026-01035-4As large language models begin to plan and run experiments on their own, the reflex is to demand that they be interpretable before they are trusted. We argue that what grounds trust in autonomous science is not a view inside the model; it is provenance: a complete, re-openable record of what was reasoned, done and measured that can be audited and set right when it is wrong, making autonomy a corrective [...]
Our summary: Provenance is essential for establishing trust in autonomous scientific processes. A complete record of reasoning, actions, and measurements allows for auditing and correction. While interpretability is important, the provenance record is crucial for ensuring rigor and accountability.
provenance, trust, autonomous science, interpretability
Publication

















