A5011973574- Topic Modeling
- Natural Language Processing Techniques
- Seismic Waves and Analysis
- Seismology and Earthquake Studies
- Web Data Mining and Analysis
- Semantic Web and Ontologies
- Multimodal Machine Learning Applications
- Seismic Imaging and Inversion Techniques
- Geophysics and Sensor Technology
- Earthquake Detection and Analysis
- earthquake and tectonic studies
- Text Readability and Simplification
- Speech and dialogue systems
- Data Quality and Management
- Biomedical Text Mining and Ontologies
- Advanced Text Analysis Techniques
- Consumer Market Behavior and Pricing
- Text and Document Classification Technologies
- Landslides and related hazards
- Constraint Satisfaction and Optimization
- Logic, Reasoning, and Knowledge
- Advanced Fiber Optic Sensors
- Auction Theory and Applications
- Domain Adaptation and Few-Shot Learning
- Speech Recognition and Synthesis
Institut des Sciences de la Terre
2021-2025
Université Savoie Mont Blanc
2021-2025
Université Grenoble Alpes
2021-2025
Université Gustave Eiffel
2021-2025
Centre National de la Recherche Scientifique
2021-2025
Institut de Recherche pour le Développement
2021-2024
Université Libre de Bruxelles
2024
Walloon Excellence in Lifesciences and Biotechnology
2024
GNS Science
2024
Victoria University of Wellington
2018-2022
Manually querying search engines in order to accumulate a large bodyof factual information is tedious, error-prone process of piecemealsearch. Search retrieve and rank potentially relevantdocuments for human perusal, but do not extract facts, assessconfidence, or fuse from multiple documents. This paperintroduces KnowItAll, system that aims automate the tedious ofextracting collections facts web an autonomous,domain-independent, scalable manner.The paper describes preliminary experiments...
Traditional information extraction systems have focused on satisfying precise, narrow, pre-specified requests from small, homogeneous corpora. In contrast, the TextRunner system demonstrates a new kind of extraction, called Open Information Extraction (OIE), in which makes single, data-driven pass over entire corpus and extracts large set relational tuples, without requiring any human input. (Banko et al., 2007) is fully-implemented, highly scalable example OIE. TextRunner's extractions are...
Natural Language Interfaces to Databases (NLIs) can benefit from the advances in statistical parsing over last fifteen years or so. However, parsers require training on a massive, labeled corpus, and manually creating such corpus for each database is prohibitively expensive. To address this quandary, paper reports PRECISE NLI, which uses parser as "plug in". The shows how strong semantic model coupled with "light re-training" enables overcome errors, correctly map parsed questions...
Recognizing names and linking them to structured data is a fundamental task in text analysis. Existing approaches typically perform these two steps using pipeline architecture: they use Named-Entity Recognition (NER) system find the boundaries of mentions text, an Entity Linking (EL) connect entries or semi-structured repositories like Wikipedia. However, tasks are tightly coupled, each type can benefit significantly from kind information provided by other. We present joint model for NER EL,...
As product prices become increasingly available on the World Wide Web, consumers attempt to understand how corporations vary these over time. However, change based proprietary algorithms and hidden variables (e.g., number of unsold seats a flight). Is it possible develop data mining techniques that will enable predict price changes under conditions?This paper reports pilot study in domain airline ticket where we recorded 12,000 observations 41 day period. When trained this data, Hamlet ---...
Supervised sequence-labeling systems in natural language processing often suffer from data sparsity because they use word types as features their prediction tasks. Consequently, have difficulty estimating parameters for which appear the test set, but seldom (or never) training set. We demonstrate that distributional representations of types, trained on unannotated text, can be used to improve performance rare words. incorporate aspects these into feature space our systems. In an experiment a...
Abstract Ambient noise interferometry is becoming increasingly popular for studying seismic velocity changes. Such changes contain information on the structural and mechanical properties of Earth systems. Application to monitoring, however, complicated by large number processes capable inducing crustal We demonstrate this at White Island volcano over a 10‐year period containing multiple well‐documented eruptions. Using individual stations, we detect perturbations that ascribe volcanic...
Using 3D terrestrial laser scan (TLS) technology, we have recorded postseismic deformation on and adjacent to the surface rupture formed during 6th April 2009 L'Aquila normal faulting earthquake (Mw 6.3). modeling techniques repeated surveys 8–124 days after earthquake, produced a 4D dataset of across 3 × 65 m area at high horizontal spatial resolution. We detected millimetre‐scale movements partitioned between discrete slip development hangingwall syncline over 10's meters. interpret...
Finding the right representations for words is critical building accurate NLP systems when domain-specific labeled data task scarce. This article investigates novel techniques extracting features from n-gram models, Hidden Markov Models, and other statistical language including a Partial Lattice Random Field model. Experiments on part-of-speech tagging information extraction, among tasks, indicate that taken in combination with more traditional features, outperform alone, graphical model...
Abstract The Whakaari/White Island volcano, located ~ 50 km off the east coast of North in New Zealand, has experienced sequences quiescence, unrest, magmatic and phreatic eruptions over last decades. For 15 years, seismic data have been continuously archived providing potential insight into this frequently active volcano. Here we take advantage unusually long time series to retrospectively process using ambient noise tremor-based methodologies. We investigate (RSAM) frequency (Power...
Abstract Volcanic inflation and deflation often precede eruptions can lead to seismic velocity changes ( ) in the subsurface. Recently, interferometry on coda of ambient noise‐cross‐correlation functions yielded encouraging results detecting these at active volcanoes. Here, we analyze data recorded Klyuchevskoy Group Kamchatka, Russia, between summer 2015 2016 study signals related volcanic activity. However, ubiquitous tremors introduce distortions noise wavefield that cause artifacts...
Continuous monitoring of volcanic gas emissions is crucial for understanding activity and potential eruptions. However, gases underwater are infrequently studied or quantified. This study explores the Distributed Acoustic Sensing (DAS) technology to monitor degassing. DAS converts fiber-optic cables into high-resolution vibration recording arrays, providing measurements at unprecedented spatio-temporal resolution. We conducted an experiment Laacher See volcano in Germany, immersing a cable...
In urban environments, shallow geothermal heating and cooling systems can play a crucial role in the transition towards renewable energy sources. One such site (USquare) is transformed military barracks Brussels, where over one hundred boreholes were drilled (~120 m) equipped with heat exchangers as part of low-enthalpy network for multi-use development project. Fourteen these fiber optic cables that provide continuous temporal monitoring downhole conditions during operation. This includes...
Open-conduit basaltic volcanoes are susceptible to sudden transitions from mild activity violent explosive eruptions with little no warning. Such was the case at Stromboli in summer of 2019, when two paroxysmal explosions occurred within approximately months (July 3 and August 28). We apply coda wave interferometry identify possible behavior build-up these events, computing seismic velocity changes using five broadband stations on volcano between 2013–2022. This timeframe encompasses a range...
As household appliances grow in complexity and sophistication, they become harder to use, particularly because of their tiny display screens limited keyboards. This paper describes a strategy for building natural language interfaces that circumvents these problems. Our approach leverages decades research on planning databases by reducing the appliance problem database problem; reduction provably maintains desirable properties interface. The goes describe implementation evaluation EXACT...