GEOAI Distinguished Lecture

Title
To be defined

Speaker
To be defined

Affiliation
To be defined

Email
To be defined

Synopsis
To be defined

Back to top

ISSMGE Bright Spark Lectures

The Bright Spark Lecture award was introduced by ISSMGE past president, Professor Charles W.W. Ng to promote young members of ISSMGE to play a major role in international and regional conferences by delivering keynote and invited lectures at these conferences.

3FOMLIG will host two Bright Spark Lectures to award a central role to the next generation of geotechnical engineers working with ML/AI in research or academia. Lectures can focus on aspects related to advancements in ML/AI in geotechnical research, practice, and education.

Conditions for eligibility include:

• The candidate is an outstanding and promising young member of ISSMGE
• The candidate will be 35 years or under on the last day of 3FOMLIG (October 17, 2025)
  • Three recommendation letters are prepared to support nomination (at least one letter from the candidate’s Member Society; at least one letter from the candidate’s University, Company or similar)

Interested persons are welcome to obtain further details about Bright Spark Lectures on the ISSMGE Bright Spark Lecture Award page and download the application form. Completed application forms are to be sent to the Chair of the 3 FOMLIG Organizing Committee, Prof. Marco UZIELLI (marco.uzielli@unifi.it) by 23:59 CET on April 30th, 2025 at the latest.

Successful candidates will be invited to give a 30-minute presentation at 3FOMLIG and write a plenary paper which will be included in a Special Issue of Geodata and AI. A certificate will be awarded to the recipient by the ISSMGE President (or their designate) at 3FOMLIG as a token of congratulations, recognition, and appreciation. Registration to 3FOMLIG will be waived.

Back to top

Parallel Technical Sessions

PS01 – Open geodatabases and their use in site characterization in data-scarce regions

Conveners
Monica LÖFMAN (Ramboll, Finland – monica.lofman@ramboll.fi ), Paul VARDANEGA (University of Bristol, UK – p.j.vardanega@bristol.ac.uk)

Abstract
This session aims to showcase and describe various open access geodatabases and their applications in the field of site characterization – especially in data-scarce regions. Contributions to this session may deal with any type of geodatabases, such as soil, rock, seismic, geophysics and foundation engineering related databases. Besides introducing new open geodatabases, we also welcome contributions showcasing open geodatabase collections and frameworks for geodatabase sharing. In addition, we encourage submissions of application examples, how the existing open geodatabases (such as those included in the TC304 compendium of databases, “304dB”) can support the site characterization process. The presented examples may utilize ML/AI methods, but also other applications are welcome, including but not limited to: state-of-the-practice application examples, empirical correlations, outlier detection and data filtering, and frameworks to support engineering judgement and decision-making.

PS02 – Enhancing the value of data using ML/AI: temporal evolution of landslide hazard due to climate change in Hong Kong

Conveners
Andy LEUNG (The Hong Kong Polytechnic University – andy.yf.leung@polyu.edu.hk), Edward CHU (GEO – edwardkhchu@cedd.gov.hk)

Abstract
Hong Kong has a hilly natural terrain with little flat land. Landslide risk assessment and management for both man-made slopes and the natural terrain are therefore high on the agenda of the engineering community of Hong Kong.  The effects of climate change on landslide risks are also crucial aspects that warrant ongoing scientific investigation and engineering adaptation.  Over the past few decades, the Civil Engineering and Development Department (CEDD) of the Hong Kong SAR government have compiled large volumes of data on rainfall intensity, landslide events and other geotechnical information across the city, to support the decision-making processes on slope upgrading and other geotechnical projects.

The datasets include the following:

• Raingauge data maintained by CEDD
• Borehole logs and laboratory testing data
• Geological map
• Information of registered man-made slopes
• Digital Terrain Model based on LiDAR Survey
• Natural terrain landslide inventory
• Dataset of reported landslide incidents

Participants of this session will be provided with subsets of the abovementioned data several months ahead of the 3FOMLIG Workshop. These can serve as valuable resources for the development, fine-tuning or validation of machine learning models, on potential topics including (but not limited to) 3D subsurface characterization, rainfall pattern predictions under the influence of climate change, landslide-rainfall correlation and landslide hazard zonation.  Focused (online) meetings will be held among the participants, data owners and session organizers from April to August 2025, with key findings to be discussed and presented at the main 3FOMLIG event.

PS03 – Machine learning techniques for modeling slope–vegetation–atmosphere interactions under changing climatic conditions

Conveners
Michele CALVELLO (University of Salerno, Italy – mcalvello@unisa.it), Luca COMEGNA (Università della Campania “Luigi Vanvitelli – luca.comegna@unicampania.it), Luca PICIULLO (Norwegian Geotechnical Institute, Norway – luca.piciullo@ngi.no), Guido RIANNA (CMCC, Italy – guido.rianna@cmcc.it)

Abstract
This session aims to assess the capabilities of innovative applications of machine learning (ML) for understanding complex soil–vegetation–atmosphere interactions within slopes, particularly in the context of climate change. Contributions to this session will address the use of ML techniques to analyze large-scale and high-resolution datasets and monitoring data enabling more accurate modeling of weather-induced hazards for natural and artificial slopes (e.g., landslides, hyper-concentrated flows and flash floods, soil erosion, dam and levee failures). Emphasis will be placed on the integration of in-situ and remote sensing monitoring of soils with meteorological and climatic data to enhance the hydro-mechanical soil characterization and to assess slope behavior and stability, both at local and regional scales. By bridging the gap between data-driven modeling and traditional geotechnical approaches, this session aims to foster interdisciplinary collaboration. Relevant topics for submissions include, but are not limited to:

• modeling and predicting the interactions between soil and the atmosphere in slopes under varying environmental and land cover conditions;
  • integration of ML and physically-based numerical analysis for complex thermo-hydro-mechanical soil behaviors;
• evaluation over time of soil-related meteo-climatic hazards, including real-time assessments and short-term (early warning) and long-term (land use planning) predictions;
• soil suction and water retention dynamics;
• utilizing ML for evaluating the evolving hazards posed by changing climatic conditions.

PS04 – Applications of ML/DL Methods to slope stability monitoring, modelling, and early warning in natural and mining environments

Conveners
Filippo CATANI (University of Padova, Italy)

Abstract
This session seeks to address the knowledge transfer between novel ML/DL technology currently being developed and operational applications of slope stability analysis at different scales and stages of risk mitigation, in both natural and mining environments. Scales to be considered include large regional assessments for planning and early warning, medium-scale multi-slope applications, as well as single-slope analysis.

Examples to be presented are expected to cover one or more aspects related to the application of ML and DL methods to slope stability monitoring, slope displacement and deformation analysis, the integration of field data with numerical models, support for decision support systems (DSS) in risk reduction, and the assimilation of remote sensing data in numerical and statistical models for slope hazard assessment.

PS05 – Data-knowledge collaborative-driven approaches for geoscience and geoengineering

Conveners
Wengang ZHANG (Chongqing University, China – zhangwg@cqu.edu.cn), Weixin SUN (Chongqing University, China – 20201601052@cqu.edu.cn), Songlin LIU (Chongqing University, China – songlinl@cqu.edu.cn)

Abstract
The integration of domain-specific knowledge with data-driven machine learning techniques has emerged as a crucial approach to enhancing the performance, interpretability, and robustness of AI models. This session aims to explore data-knowledge collaborative-driven approaches adopted for solution of geoscience and geoengineering problems, with a particular focus on Physics-Informed Neural Networks (PINNs), SHapley Additive exPlanations (SHAP), and hybrid models etc. that combine machine learning algorithms with expert knowledge or physical principles. PINNs generally adopt a novel way of incorporating physical laws into machine learning models, enabling more accurate solutions to complex differential equations and simulations of physical phenomena. SHAP, on the other hand, provides a framework for explaining machine learning model predictions by quantifying the contribution of each input feature, enhancing model transparency and interpretability. By exploring both theoretical advancements and real-world applications, this session will highlight how these methodologies, when integrated with domain knowledge, can significantly improve model accuracy, decision-making, and predictive capabilities. Contributions to this session would focus on the development of new algorithms, application case studies, and hybrid approaches that bridge the gap between data-driven insights and expert knowledge.

PS06 – Meta-modelling for geotechnical engineering: bridging data gaps with AI and simulations

Conveners
Marco D’IGNAZIO (Ramboll, Finland – marco.dignazio@ramboll.fi)

Abstract
The lack of publicly available geotechnical data, particularly in offshore environments, limits the application of machine learning (ML) in site characterisation, foundation design, and risk assessment. Meta-modelling offers a powerful solution by using numerical simulations (e.g., FEM, DEM, CFD) to develop surrogate models that approximate complex geotechnical behaviour with high efficiency. Numerical simulations, particularly finite element modelling (FEM), are well-established and reliable tools for modelling foundation performance, with parameter determination typically relying on advanced laboratory and in-situ testing, as well as limited publicly available and industry-owned databases. Data-driven models based on numerical simulations enable rapid predictions, optimisation, and uncertainty quantification, reducing reliance on unavailable or confidential field data. This session will explore best practices in building and validating meta-models, their integration with ML workflows, and real-world applications in offshore and onshore geotechnics. By combining physics-based simulations with AI-driven modelling, meta-modelling paves the way for more scalable, efficient, and data-informed geotechnical solutions.

PS07 – Intelligent technologies for risk assessment in tunnel and underground engineering

Conveners
Dongming ZHANG (Tongji University, China – 09zhang@tongji.edu.cn), Mingliang ZHOU (Tongji University, China – zhoum@tongji.edu.cn), Shuai ZHAO (The Hong Kong Polytechnic University – zhaosoi@126.com)

Abstract
This session will explore integrating advanced AI/ML technologies with traditional tunnel engineering, focusing on machine learning’s role in risk assessment and management. Topics include real-time anomaly detection, intelligent sensing, and optimization for safety. The session will highlight the synergy between data-driven and physics-based approaches to improve predictive accuracy and risk control. Interdisciplinary collaboration is encouraged, with an emphasis on novel materials and hybrid models.  Relevant topics for submission include, but are not limited to:

• Advancements in machine learning models for predicting the stress-strain behavior of geotechnical materials under complex conditions, such as high stress, variable moisture content, and seismic activity;
• Breakthroughs in intelligent sensing technologies and AI-driven algorithms for real-time risk assessment, early warning systems, and optimization in tunnel engineering;
• The integration of AI/ML with classical mechanics methods to improve the analysis and evaluation of tunnel and underground construction safety, enabling intelligent warnings and data-driven decision-making;
• Interdisciplinary research on novel materials and hybrid models aimed at enhancing structural resilience and long-term safety in tunnel and underground engineering.

PS08 – Big data and foundation models in tunneling engineering

Conveners
Chao ZHANG (College of Civil Engineering, Hunan University, China – chao_zhang@hnu.edu.cn), Guowen XU (Department of Civil Engineering, Southwest Jiaotong University, China – xuguowen@swjtu.edu.cn)

Abstract
Tunneling engineering focuses on optimizing underground space use, such as in shield tunneling, while tackling challenges like geological prediction, risk warning, safety control, and green construction. With the advancement of artificial intelligence, harnessing big data and foundation models offers unprecedented opportunities to optimize design, construction, and maintenance processes. This session aims to bridge the latest advancements in academic research with practical applications in tunneling engineering, offering a platform for innovative ideas and groundbreaking solutions in this rapidly evolving field. We encourage submissions of recent research on, but not limited to, the following topics: Integration and management of large-scale geological and operational datasets; Data-driven insights for real-time decision-making in tunneling operations; Application of AI foundation models (e.g., LLMs, multimodal models) for decision support in tunneling; Innovations in machine learning for ground condition assessment and risk prediction; Data-driven approaches to reduce energy consumption and carbon emissions; AI-powered solutions for optimizing material usage and waste reduction; Multimodal frameworks combining foundation models, big data analytics, and green construction practices.

PS09 – Data-integrated risk assessment in geological and geotechnical engineering

Conveners
Lü QING (Zhejiang University, China – lvqing@zju.edu.cn), Honglei SUN (Zhejiang University of Technology, China – sunhonglei@zjut.edu.cn), Yuanqin TAO (Zhejiang University of Technology, China – taoyuanqin@zjut.edu.cn)

Abstract
This session focuses on the applications and prospective advancements of data-driven methods in geological and geotechnical practices. It aims to explore the transformative impact of increasingly accessible and large-scale data in reshaping conventional methodologies for geological hazard assessment, geotechnical construction, design, and maintenance. Topics of interest include but are not limited to: development of site characterization databases and statistical analysis of geotechnical properties; risk assessment and early warning technologies of geological hazards such as landslides, rockfalls, debris flows, and subsurface cavities; uncertainty and reliability analysis of rock excavation; data-knowledge-driven prediction and active control of excavation-induced deformations.

PS10 – AI-based risk and resilience analysis for underground engineering

Conveners
Yue PAN (Shanghai Jiao Tong University, China – panyue001@sjtu.edu.cn), Zhong-Kai HUANG (Tongji University, China –  5huangzhongkai@tongji.edu.cn), Xin WEI (University of Michigan, United States – xincwei@umich.edu)

Abstract
The rapid development of underground engineering, particularly in urban infrastructure projects such as metro systems, deep foundation pits, and tunnelling, presents considerable challenges in the areas of risk assessment and resilience optimization. Traditional risk management and resilience analysis in these complex environments is often constrained by the inherent uncertainty of geotechnical data, unpredictable environmental factors, and the dynamic nature of subsurface construction. Artificial Intelligence (AI) holds substantial promise in overcoming these limitations, facilitating advanced predictive analytics, real-time monitoring, and adaptive decision-making. This session focuses on the application of AI-based techniques in risk and resilience analysis for both predictive and prescriptive analytics in underground engineering projects, such as machine learning, deep learning, and data-driven decision-making and optimization. It aims to improve safety, operational efficiency, and environmental sustainability in underground construction projects. Contributions to this special topic are invited to explore innovative AI applications, methodological advancements, and case studies that strengthen the resilience and risk management capacities of underground engineering systems, thereby advancing the development of safer and more robust infrastructure in the face of complex and unpredictable challenges.

PS11 – Machine learning for resilient geoinfrastructure

Conveners
Enrico SORANZO (BOKU University, Austria – enrico.soranzo@boku.ac.at), Zhongqiang LIU (Norwegian Geotechnical Institute, Norway – zhongqiang.liu@ngi.no)

Abstract
This session will explore the integration of advanced machine learning techniques in the field of geotechnics, aiming to enhance the resilience and reliability of geoinfrastructure. This session will present methods for improving geotechnical data quality and reliability through standardized datasets. Additionally, the session will discuss techniques for quantifying and managing uncertainties, addressing the heterogeneity of geomaterials and modeling nonlinear behaviors in geosystems. Potential topics include also the application of generative AI techniques to geotechnical problems, the integration of physical laws into neural network models, case studies on the application of machine learning to geo-infrastructures and geo-hazards and techniques to make machine learning solutions more transparent and understandable to engineers and stakeholders.

PS12 – Values of machine learning in geotechnical reliability and risk

Conveners
Zi-Jun CAO (Southwest Jiaotong University, China – zijuncao@swjtu.edu.cn), Tengyuan ZHAO (Xi’an Jiaotong University, China – tyzhao@xjtu.edu.cn)

Abstract
Machine Learning (ML) has emerged as a powerful tool in geotechnical engineering, providing significant advancements in the fields of reliability analysis and risk management. This special session will explore the value of ML in enhancing the accuracy, efficiency, and adaptability of geotechnical design and analysis. Traditional geotechnical models often struggle with the complexity and uncertainty inherent in soil behavior, environmental factors, and structural performance. ML techniques offer a transformative solution, with their ability to process large datasets, uncover non-linear patterns, and make predictions under uncertainty. This special session highlights key applications of ML, including but not limited to, soil property prediction, slope stability analysis, foundation design, and seismic risk assessment. By incorporating ML, geotechnical engineers can improve the reliability of safety predictions, optimize design parameters, and enhance risk assessments through rational uncertainty quantification. Discussions on challenges such as data quality, model transparency, and the integration of ML with conventional engineering knowledge are also welcomed. Ultimately, this session provides a forum to demonstrate how ML is reshaping the way geotechnical professionals approach risk and reliability, offering more accurate, data-driven solutions to complex engineering problems.

Back to top

Hackathons

1^st EduHackathon – Assessing the usefulness of Large Language Models in geotechnical education

Conveners

Main convener: Michele CALVELLO (University of Salerno, Italy – mcalvello@unisa.it)

Co-conveners (confirmed so far)

• Marco UZIELLI (University of Florence, Italy – uzielli@unifi.it)
• Tae-Hyuk KWON (Korea Advanced Institute of Science and Technology, South Korea – kwon@kaist.ac.kr)
• Ana HEITOR (University of Leeds, UK – Heitor@leeds.ac.uk)
• Emilio BILOTTA (University of Napoli Federico II, Italy – bilotta@unina.it)
• Alessia CUCCURULLO (Université libre de Bruxelles, Belgium – cuccurullo@ulb.be)
• Stephen WU (Institute of Statistical Mathematics, Japan – stewu@ism.ac.jp)
• Andy YF LEUNG (Hong Kong Polytechnic University, China – yf.leung@polyu.edu.hk)
• Paul VARDANEGA (University of Bristol, UK – j.vardanega@bristol.ac.uk)
• Charles MAC ROBERT (Stellenbosch University, South Africa – macrobert@sun.ac.za)

Abstract
This AI geo-educational hackathon seeks to explore the possible use of large language models (LLM) and other generative AI tools in:

1. creating support educational material for soil mechanics and geotechnical courses;
2. exploring questions posed to LLM by geotechnical engineering instructors on specific geotechnical topics, and the related LLM answers, as bases for collective discussions among experts on those topics.

The plenary hackathon session during the conference should be seen as part of a longer process involving a core group of conveners who will, in the months before the conference:

• define the topics to tackle (e.g., about which geotechnical educators feel a need for creating new material or exploring existing material with AI tools);
• explore LLM and generative AI tools which can be used for the purpose;
• create some reporting, and possibly some new educational material, on the issues tackled;
• collectively discuss these preliminary results;
• define the format of the conference hackathon, which will include the presentation to the workshop participants of the work conducted by the group of conveners.

2nd GeoTechathon: Multiagent LLMs

Conveners

Main conveners: Stephen WU (The Institute of Statistical Mathematics, Japan – stewu@ism.ac.jp); Andy YF LEUNG (Hong Kong Polytechnic University, China – andy.yf.leung@polyu.edu.hk)

Co-conveners: Marco UZIELLI (University of Florence, Italy – marco.uzielli@unifi.it); Michele CALVELLO (University of Salerno, Italy – mcalvello@unisa.it); Yuanqin TAO (Zhejiang University of Technology, China – taoyuanqin@zjut.edu.cn); Kok Kwang PHOON (Singapore University of Technology and Design, Singapore – kkphoon@sutd.edu.sg)

Abstract
This hackathon will focus on developing multiagent AI systems that integrate large language models (LLMs) with other geotechnical computational tools to solve geotechnical problems in a predefined application domain. The participants are required to form small groups and decide on specific problems to solve by themselves. The organizer will provide a batch of sample agents for participants to use or build on in order to design and implement a workflow that solves the targeted problem robustly and efficiently. A judge panel will evaluate the solutions and decide the winners of this hackathon event.

Target applications: Geohazards related tasks

[Hazard example – landslide, liquefaction, levee breach, sinkhole and underground cavity, etc.]

[Task example – forecast, detection, fast investigation, real-time assessment, post-event evaluation, report generation, etc.]

Target participants: geotechnical engineers from industry, young researchers, graduate students

Prerequisites for a team: at least one person specialized in python programming, at least one person has experience in using LLMs through API in python. (Can be the same person that satisfies both requirements)

Hackathon timeline (tentative)
March – April 2025: Team forming and event application
May 2025: Tutorial sessions for building multiagent system based on LLMs
June – August 2025: Hackathon working period
September 2025: Initial solution submission and feedback from event supporters
October 1^st: 2025: All teams submit final solution

At 3FOMLIG: Present solutions and judges provide evaluations

Please contact one of the main conveners if you are interested in joining, and we will follow up with you on the application process.

Back to top

3FOMLIG Nextgen Initiatives: a Journey from Theory to Practice

The 3FOMLIG NextGen Group, involving undergraduate and postgraduate students, explores the integration of AI in geotechnical engineering through three main activities:

• “Museum of AI” is a curated science gallery highlighting the chronological development of AI models, themed with the history of Florence. This thematic gallery showcases the progression of AI models, presenting its applications in geotechnical engineering.
• “Student Competition” encourages AI learners to compete and make collaborations to address a specific geotechnical problem. Their solutions will be judged for innovation and accuracy of results. The competition will be advertised on some platforms, such as Kaggle, to attract global attention and participation.
• “3FOMLIG NextGen Paper” is the productive game of the 3FOMLIG NextGen Group – the game of collaboration! It will be mainly played at the workshop, in a teamwork, guided by PhD students. Each team tries to solve a broken part of the bigger puzzle, which is a real-world geotechnical problem. The problem will be tackled through hands-on programming and AI models evaluation by each team during the workshop. The problem’s components address diverse geotechnical challenges, from field and lab testing to geotechnical design. This session will culminate in a multi-disciplinary paper capturing the collective expertise of all contributors. All AI learners –whether from academia or industry– are welcome to play the game!

If you are an undergraduate or postgraduate student and wish to receive further information and/or join one or more of the 3FOMLIG NextGen initiatives, please contact the 3FOMLIG NextGen coordinators Alireza DUZANDEH (University of Florence, Italy – alireza.duzandeh@unifi.it) and Mohammadsadegh FARHADI (Tampere University, Finland – mohammadsadegh.farhadi@tuni.fi)

Back to top

7TH Machine Learning in Geotechnics Dialogue (7MLIGD)

Trustworthy data-centric geotechnics

Panel
Kok Kwang PHOON (Singapore University of Technology and Design, Singapore – kkphoon@sutd.edu.sg), Patrizia VITALE (Norwegian Geotechnical Institute, Norway – patrizia.vitale@ngi.no)

Abstract
Data-centric geotechnics is an emerging interdisciplinary field that involves the integration of information, data, techniques, tools, perspectives, concepts, theories, and/or experiences from geotechnical engineering, machine learning (ML), and artificial intelligence (AI). Currently, ML and AI are at most deployed as proofs of concept in geotechnics. Embracing data-centric geotechnics is the only way for geotechnical engineers to remain in the driver seat to deploy ML/AI responsibly and safely, to the benefit of the community and stakeholders and in compliance with technical regulations and ethical paradigms. A greater appetite for data-driven innovations and, correspondingly, the capability to suitably manage the risks which come with the adoption of these momentous innovations, would allow the geotechnical community to take advantage of major breakthroughs occurring elsewhere. The agenda to introduce trustworthiness explicitly in geotechnical research, practice, and education is called “trustworthy data-centric geotechnics”. The 7th Machine Learning in Geotechnics Dialogue will provide a dynamic, interactive opportunity to discuss technical and non-technical approaches and opportunities to attract researchers, practitioners, and the next generation of geotechnical engineers to address the critical trustworthiness issue in a comprehensive, systemic, and strategic manner.

Back to top

Free Short Course for Students “A Primer to Machine Learning and Artificial Intelligence in Geotechnical Engineering”

October 15 2025 09:00-13:00

This free short course will be held prior to the opening of 3FOMLIG and will provide undergraduate and postgraduate students – including those who are new to the topic – with fundamental insights into the roles machine learning and artificial intelligence in geotechnical engineering and into the advantages which these approaches may bring to geotechnical research and practice. Lectures will be held by leading researchers and will cover themes such as databases for data-centric geotechnics, Bayesian machine learning in site characterization, and the application of Large Language Models in geotechnics.

Course program
09:00-09:45 – Databases for data-centric geotechnics – Kok-Kwang PHOON (Singapore University of Technology and Design, Singapore – kkphoon@sutd.edu.sg)
09:45-10:30 – Bayesian machine learning in geotechnical site characterization – Jianye CHING (National Taiwan University, Taipei – jyching@ntu.edu.tw)
10:30-11:00 – Coffee break
11:00-11:45 – Accelerating problem-solving in geotechnical engineering with large language models (LLMs) – Stephen WU (Institute of Statistical Mathematics, Japan – stewu@ism.ac.jp)
11:45-12:30 – Application of LLM to landslide investigations – Andy YF LEUNG (Hong Kong Polytechnic University, China – yf.leung@polyu.edu.hk)

If you are an undergraduate or postgraduate student and wish to join the short course, please send an email to the 3FOMLIG Secretary Dr. Monica LÖFMAN at monica.lofman@ramboll.fi. Enrolment in the short course will be on a first come-first serve basis until the maximum number of participants is reached.

Back to top