BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.2.3.1//EN TZID:Europe/Paris X-WR-TIMEZONE:Europe/Paris BEGIN:VEVENT UID:1190@cran.univ-lorraine.fr DTSTART;TZID=Europe/Paris:20260206T140000 DTEND;TZID=Europe/Paris:20260206T150000 DTSTAMP:20260305T132828Z URL:https://www.cran.univ-lorraine.fr/events/seminaire-de-roberto-cilli-ge oressources/ SUMMARY:Séminaire de Roberto Cilli (GeoRessources) DESCRIPTION:Speaker: Roberto Cilli\, GeoRessources (Nancy)\nTitle: Transfor mer-Based Geo-technical Classification of Borehole Logs with Benchmarking of Uncertainty Quantification Methods\nLocation: SiMul meeting room\, Facu lté des Sciences et Technologies\, Henri Poincaré Building\, 4th floor\n Abstract:\nGeological borehole descriptions are a fundamental resource for subsurface modeling\, yet they are often stored as unstructured text\, li miting their usability for automated analysis. While a growing number of s tudies have applied natural language processing to geological texts\, most approaches disregard the semantic continuity between adjacent lithologica l descriptions and overlook the structured nature of stratigraphic success ions within borehole logs. In this study\, we introduce a context-aware se quence labeling framework that applies natural language processing and pos itional encoding to classify lithological units from borehole log descript ions modeled as structured sequences by combining a pre-trained Sentence-B ERT model for semantic encoding with a single-layer Transformer encoder th at captures contextual and positional relationships (Reimers et al\, 2019\ , Vaswani et al. 2017). We evaluate our approach on a dataset of manually labeled boreholes from the Pianello hillslope\, located in Southern Italy\ , focusing on five lithological classes relevant to slope stability analys is. The proposed method achieves an accuracy gain of approximately 15% com pared to a baseline random forest classifier fed with Sentence-BERT embedd ings\, revealing that the context of a lithological description can signif icantly improve the classification performance of NLP algorithms. Furtherm ore\, we show that the proposed architecture\, consisting of less than 350 k learnable parameters\, is lightweight and scalable\, enabling rapid proc essing and confirming its practical applicability in real-world scenarios especially in low-resource computing environments (Asus ROG Flow X13 AMD R yzen9 16GB RAM equipped with a NVIDIA RTX3050 4GB VRAM).\n\nMoreover\, a c omparison between benchmark uncertainty quantification (UQ) algorithms\, i ncluding the Bayesian by Backprop (Blundell et al. 2015)\, Deep Ensemble ( Lakshminarayanan et al. 2016)\, MC Dropout (Gal and Ghahramani\, 2016) and a custom Bayesian framework inspired by Kendall and Gal\, 2016. Results i ndicate that the framework inspired Kendall and Gal\, 2016 is the only cap able to disentangle the epistemic and aleatoric components of uncertainty\ , among those tested. However\, uncertainty estimates still need further v alidation since we observed a weak negative correlation between epistemic uncertainty and classification accuracy in real unseen samples while undes ired behaviours are observed when the network is fed with synthetic and me aningless inputs.\n\nKendall and Gal\, 2016. What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?\nLakshminarayanan et al. 2 017\, Simple and Scalable Predictive Uncertainty Estimation using Deep Ens embles.\nGal and Z. Ghahramani\, 2016. Dropout as a Bayesian approximation : Representing model uncertainty in deep learning.\nKirkwood et al\, 2022. Bayesian Deep Learning for Spatial Interpolation in the Presence of Auxil iary Information.\nN. Reimers\, I. Gurevych\, Sentence-BERT: Sentence embe ddings using siamese BERT-Networks (2019).\nVaswani et al\, 2017. Attentio n is all you need. CATEGORIES:Séminaires projet SiMul LOCATION:CRAN - FST - 4ème\, Campus Sciences\, Boulevard des Aiguillettes\ , Vandoeuvre-lès-Nancy\, 54506\, France X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Campus Sciences\, Boulevard des Aiguillettes\, Vandoeuvre-lès-Nancy\, 54506\, France;X-APPLE-RADIUS= 100;X-TITLE=CRAN - FST - 4ème:geo:0,0 END:VEVENT BEGIN:VTIMEZONE TZID:Europe/Paris X-LIC-LOCATION:Europe/Paris BEGIN:STANDARD DTSTART:20251026T020000 TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET END:STANDARD END:VTIMEZONE END:VCALENDAR