Search

Multi-document summarization (MDS) is a difficult task in Natural Language Processing, aiming to summarize information from several documents. However, the source documents are often insufficient to obtain a qualitative summary. We propose a retriever-guided model combined with non-parametric memory for summary generation. This model retrieves relevant candidates from a database and then generates the summary considering the candidates with a copy mechanism and the source documents. The retriever is implemented with Approximate Nearest Neighbor Search (ANN) to search large databases. Our method is evaluated on the MultiXScience dataset which includes scientific articles. Finally, we discuss our results and possible directions for future work., Comment: International Conference Recent Advances in Natural Language Processing (RANLP), Sep 2023, Varna, Bulgaria

Machine Learning (ML) has emerged as a core technology to provide learning models to perform complex tasks. Boosted by Machine Learning as a Service (MLaaS), the number of applications relying on ML capabilities is ever increasing. However, ML models are the source of different privacy violations through passive or active attacks from different entities. In this paper, we present MixNN a proxy-based privacy-preserving system for federated learning to protect the privacy of participants against a curious or malicious aggregation server trying to infer sensitive attributes. MixNN receives the model updates from participants and mixes layers between participants before sending the mixed updates to the aggregation server. This mixing strategy drastically reduces privacy without any trade-off with utility. Indeed, mixing the updates of the model has no impact on the result of the aggregation of the updates computed by the server. We experimentally evaluate MixNN and design a new attribute inference attack, Sim, exploiting the privacy vulnerability of SGD algorithm to quantify privacy leakage in different settings (i.e., the aggregation server can conduct a passive or an active attack). We show that MixNN significantly limits the attribute inference compared to a baseline using noisy gradient (well known to damage the utility) while keeping the same level of utility as classic federated learning.

Stylolites are ubiquitous geo-patterns observed in rocks in the upper crust, from geological reservoirs in sedimentary rocks to deformation zones, in folds, faults, and shear zones. These rough surfaces play a major role in the dissolution of rocks around stressed contacts, the transport of dissolved material and the precipitation in surrounding pores. Consequently, they play an active role in the evolution of rock microstructures and rheological properties in the Earth's crust. They are observed individually or in networks, in proximity to fractures and joints, and in numerous geological settings. This review article deals with their geometrical and compositional characteristics and the factors leading to their genesis. The main questions this review focuses on are the following: How do they form? How can they be used to measure strain and formation stress? How do they control fluid flow in the upper crust? Geometrically, stylolites have fractal roughness, with fractal geometrical properties exhibiting typically three scaling regimes: a self-affine scaling with Hurst exponent 1.1+/-0.1 at small scale (up to tens or hundreds of microns), another one with Hurst exponent around 0.5 to 0.6 at intermediate scale (up to millimeters or centimeters), and in the case of sedimentary stylolites, a flat scaling at large scale. More complicated anisotropic scaling (scaling laws depending of the direction of the profile considered) is found in the case of tectonic stylolites. We report models based on first principles from physical chemistry and statistical physics, including a mechanical component for the free-energy associated with stress concentrations, and a precise tracking of the influence of grain-scale heterogeneities and disorder on the resulting (micro)structures. Experimental efforts to reproduce stylolites in the laboratory are also reviewed. We show that although micrometer-size stylolite teeth are obtained in laboratory experiments, teeth deforming numerous grains have not yet been obtained experimentally, which is understandable given the very long formation time of such geometries. Finally, the applications of stylolites as strain and stress markers, to determine paleostress magnitude are reviewed. We show that the scalings in stylolite heights and the crossover scale between these scalings can be used to determine the stress magnitude (its scalar value) perpendicular to the stylolite surface during the stylolite formation, and that the stress anisotropy in the stylolite plane can be determined for the case of tectonic stylolites. We also show that the crossover between medium (millimetric) scales and large (pluricentimetric) scales, in the case of sedimentary stylolites, provides a good marker for the total amount of dissolution, which is still valid even when the largest teeth start to dissolve -- which leads to the loss of information, since the total deformation is not anymore recorded in a single marker structure. We discuss the impact of the stylolites on the evolution of the transport properties of the hosting rock, and show that they promote a permeability increase parallel to the stylolites, whereas their effect on the permeability transverse to the stylolite can be negligible, or may reduce the permeability, depending on the development of the stylolite. Highlights: Stylolite formation depends on rock composition and structure, stress and fluids. Stylolite geometry, fractal and self-affine properties, network structure, are investigated. The experiments and physics-based numerical models for their formation are reviewed. Stylolites can be used as markers of strain, paleostress orientation and magnitude. Stylolites impact transport properties, as function of maturity and flow direction.

The Compact Linear Collider (CLIC) is a multi-TeV high-luminosity linear e+e- collider under development. For an optimal exploitation of its physics potential, CLIC is foreseen to be built and operated in a staged approach with three centre-of-mass energy stages ranging from a few hundred GeV up to 3 TeV. The first stage will focus on precision Standard Model physics, in particular Higgs and top-quark measurements. Subsequent stages will focus on measurements of rare Higgs processes, as well as searches for new physics processes and precision measurements of new states, e.g. states previously discovered at LHC or at CLIC itself. In the 2012 CLIC Conceptual Design Report, a fully optimised 3 TeV collider was presented, while the proposed lower energy stages were not studied to the same level of detail. This report presents an updated baseline staging scenario for CLIC. The scenario is the result of a comprehensive study addressing the performance, cost and power of the CLIC accelerator complex as a function of centre-of-mass energy and it targets optimal physics output based on the current physics landscape. The optimised staging scenario foresees three main centre-of-mass energy stages at 380 GeV, 1.5 TeV and 3 TeV for a full CLIC programme spanning 22 years. For the first stage, an alternative to the CLIC drive beam scheme is presented in which the main linac power is produced using X-band klystrons., Comment: 57 pages, 27 figures, 12 tables, published as CERN Yellow Report. Updated version: Minor layout changes for print version

We develop and test a methodology to infer paleostress from the morphology of stylolites within borehole cores. This non-destructive method is based on the analysis of the stylolite trace along the outer cylindrical surface of the cores. It relies on an automatic digitization of high-resolution photographs and on the spatial Fourier spectrum analysis of the stylolite traces. We test and show, on both synthetic and natural examples, that the information from this outer cylindrical surface is equivalent to the one obtained from the destructive planar sections traditionally used. The assessment of paleostress from the stylolite morphology analysis is made using a recent theoretical model, which links the morphological properties to the physical processes acting during stylolite evolution. This model shows that two scaling regimes are to be expected for the stylolite height power spectrum, separated by a cross-over length that depends on the magnitude of the paleostress during formation. We develop a non linear fit method to automatically extract the cross-over lengths from the digitized stylolite profiles. Results on cores from boreholes drilled in the surroundings of the Andra Underground Research Laboratory located at Bure, France, show that different groups of sedimentary stylolites can be distinguished, and correspond to different estimated vertical paleostress values. For the Oxfordian formation, one group of stylolites indicate a paleostress of around 10 MPa, while another group yields 15 MPa. For the Dogger formation, two stylolites indicate a paleostress of around 10 MPa, while others appear to have stopped growing at paleostresses between 30 and 22 MPa, starting at an erosion phase that initiated in the late Cretaceous and continues today. This method has a high potential for further applications on reservoirs or other geological contexts where stylolites are present., Comment: International Journal of Rock Mechanics and Mining Sciences (2013) online first

In January 1999, CERN began to develop CASTOR ("CERN Advanced STORage manager"). This Hierarchical Storage Manager targetted at HEP applications has been in full production at CERN since May 2001. It now contains more than two Petabyte of data in roughly 9 million files. In 2002, 350 Terabytes of data were stored for COMPASS at 45 MB/s and a Data Challenge was run for ALICE in preparation for the LHC startup in 2007 and sustained a data transfer to tape of 300 MB/s for one week (180 TB). The major functionality improvements were the support for files larger than 2 GB (in collaboration with IN2P3) and the development of Grid interfaces to CASTOR: GridFTP and SRM ("Storage Resource Manager"). An ongoing effort is taking place to copy the existing data from obsolete media like 9940 A to better cost effective offerings. CASTOR has also been deployed at several HEP sites with little effort. In 2003, we plan to continue working on Grid interfaces and to improve performance not only for Central Data Recording but also for Data Analysis applications where thousands of processes possibly access the same hot data. This could imply the selection of another filesystem or the use of replication (hardware or software)., Comment: Talk from the 2003 Computing in High Energy and Nuclear Physics (CHEP03), La Jolla, Ca, USA, March 2003, 2 pages, PDF. PSN TUDT007

The ALICE experiment at CERN will propose unprecedented requirements for event building and data recording. New technologies will be adopted as well as ad-hoc frameworks, from the acquisition of experimental data up to the transfer onto permanent media and its later access. These issues justify a careful, in-depth planning and preparation. The ALICE Data Challenge is a very important step of this development process where simulated detector data is moved from dummy data sources up to the recording media using processing elements and data-paths as realistic as possible. We will review herein the current status of past, present and future ALICE Data Challenges, with particular reference to the sessions held in 2002 when - for the first time - streams worth one week of ALICE data were recorded onto tape media at sustained rates exceeding 300 MB/s., Comment: Talk from the 2003 Computing in High Energy and Nuclear Physics (CHEP03), La Jolla, Ca, USA, March 2003, 9 pages, PDF. PSN MOGT007