Seminari - Corso "Deep Learning and Robot Perception"

Dettagli

"VISUAL TRANSFORMERS FOR INDUSTRIALIMAGE INTELLIGENCE: FROM RETRIEVAL TO KNOWLEDGE-DRIVENWORKFLOWS" - Seminario "ANALYSIS AND INTEGRATION OF ALGORITHMS AND ML MODELS TO SUPPORT AUTOMATIC CREATION OF 'CERTIFICAZIONI NOTARILI'" - Seminario

Venerdì 5 dicembre 2025 dalle ore 11.30, nell'Aula 11 del Dipartimento di Ingegneria, si terranno due seminari, che sono parte integrante del corso di Deep Learning and Robot Perception.

  • 11.30 - 12.30: VISUAL TRANSFORMERS FOR INDUSTRIAL IMAGE INTELLIGENCE: FROM RETRIEVAL TO KNOWLEDGE-DRIVENWORKFLOWS - azienda: Red Lync
    Speaker: Enrico Bellocchio
    Abstract
    Visual transformers are reshaping industrial computer vision by unifying representation learning, region-aware perception, and metadata fusion within a single scalable pipeline. This talk surveys how technical drawings, component photos, and PDF schematics can be indexed through automatic ROI extraction, transformer-based embeddings, and ERP/MES data alignment to support rapid similarity search, semantic filtering, and decision support across archives of tens of thousands of assets. We discuss practical design choices—patch-level detection, multi-modal embeddings, asynchronous progress tracking, and PDF reporting—that enable engineers to retrieve prior work, compare variants, and maintain living repositories of manufacturing knowledge. The resulting workflows illustrate how transformer-driven visual search bridges descriptive data and imagery, accelerating design reuse and collaborative diagnostics in industrialsettings.
    Locandina

  • 12.30 - 13.30: ANALYSIS AND INTEGRATION OF ALGORITHMS AND ML MODELS TO SUPPORT AUTOMATIC CREATION OF 'CERTIFICAZIONI NOTARILI' - azienda: Weedea
    Speaker: Gabriele Mawi, Nicola Cucina
    Abstract
    The fields of software engineering and legal certification share a heavy reliance on individual expertise, with the latter often depending on operator intuition and decision-making processes that are difficult to codify using standard heuristics.Investigating the feasibility of representing these complex, experience-based workflows, we put emphasis on the differences between a classical deterministic approach versus the usage of Machine Learning and Artificial Intelligence algorithms, analyzing the trade-off between manual knowledge codification, which demands extensive domain study, and the potential for offloading complex logic to predictivemodels.
    Locandina

Notizie ed eventi - Elenco