157082 research outputs found
Sort by
NLP verification:towards a general methodology for certifying robustness
No full textMachine learning has exhibited substantial success in the field of natural language processing (NLP). For example, large language models have empirically proven to be capable of producing text of high complexity and cohesion. However, at the same time, they are prone to inaccuracies and hallucinations. As these systems are increasingly integrated into real-world applications, ensuring their safety and reliability becomes a primary concern. There are safety critical contexts where such models must be robust to variability or attack and give guarantees over their output. Computer vision had pioneered the use of formal verification of neural networks for such scenarios and developed common verification standards and pipelines, leveraging precise formal reasoning about geometric properties of data manifolds. In contrast, NLP verification methods have only recently appeared in the literature. While presenting sophisticated algorithms in their own right, these papers have not yet crystallised into a common methodology. They are often light on the pragmatical issues of NLP verification, and the area remains fragmented. In this paper, we attempt to distil and evaluate general components of an NLP verification pipeline that emerges from the progress in the field to date. Our contributions are twofold. First, we propose a general methodology to analyse the effect of the embedding gap - a problem that refers to the discrepancy between verification of geometric subspaces, and the semantic meaning of sentences which the geometric subspaces are supposed to represent. We propose a number of practical NLP methods that can help to quantify the effects of the embedding gap. Second, we give a general method for training and verification of neural networks that leverages a more precise geometric estimation of semantic similarity of sentences in the embedding space and helps to overcome the effects of the embedding gap in practice.</p
Run Sum Hotelling's <i>T</i><sup>2</sup> Chart for Autocorrelated Processes
No full textAutocorrelation has recently gained significant attention as many industrial outcomes are autocorrelated. This work presents a multivariate run sum T2 (MRS) chart for monitoring autocorrelated processes by utilizing a first-order vector autoregressive (VAR(1)) model with an s-skip sampling strategy. To evaluate the performance of the proposed chart, the average run length (ARL) measure is used. Comparative analyses reveal that the proposed MRS chart significantly outperforms the basic T2 charts for autocorrelated processes. The s-skip sampling strategy is incorporated into the MRS chart for autocorrelated processes to reduce the impact of autocorrelation. An illustrative example is included to demonstrate the practical implementation of the proposed chart, highlighting its advantages in real-world applications and emphasizing its potential to improve process monitoring across various industrial settings.</p
Enhanced distribution-free Cucconi schemes based on runs: An application to the surveillance of daily average customer service time
No full textThe measurement of daily average customer service times in a service centre holds paramount importance as it serves as a key metric for assessing service performance and quality. It is essential to ensure that the average service time (AST) aligns with predetermined quality standards. Hence, regular monitoring of service times is vital to uphold and enhance service quality. This study proposes two enhanced distribution-free Cucconi schemes integrated with runs rules (RR), i.e., the 2-of-3 (2/3) RR Cucconi and the synthetic Cucconi schemes. The practical utility of these proposed schemes is illustrated through an application in the surveillance of the daily average customer service time at an Australian service centre. The inherent nonparametric feature of these schemes renders them versatile in a wide range of applications, regardless of the underlying process distribution. In addition, these newly devised schemes possess the capability to surveil both process location and scale parameters simultaneously. Through comprehensive Monte-Carlo simulations, we demonstrate the superior performance of the 2/3 RR Cucconi scheme against four existing schemes, namely the Shewhart-Lepage, 2/3 RR Lepage, synthetic Lepage, and Shewhart-Cucconi schemes. Therefore, the 2/3 RR Cucconi scheme is recommended over all competing schemes considered in this paper due to its excellent performance and ease of use
Seismic stratigraphy of the Guinea Plateau before, during and after the opening of the Equatorial Atlantic Gateway
No full textThe Guinea Plateau contains an ∼200 Myr stratigraphic record, encompassing the mid-Cretaceous opening of the Equatorial Atlantic Gateway (EAG). Here we present new 2D seismic data to constrain the structural and stratigraphic evolution of the plateau. Seismic stratigraphic analysis reveals five megasequences of ∼25–65 Myr duration: M1, a Jurassic synrift sequence with prominent seaward-dipping reflections; M2, a late Jurassic–Early Cretaceous post-rift carbonate platform; M3, a late Early Cretaceous transform clastic-dominated sequence; M4, an Albian–Maastrichtian ocean–continent transform to post-transform sequence; and M5, a Maastrichtian–recent passive margin sequence with low sedimentation rates. These megasequences also contain prominent transgressive–regressive cycles of 5–10 Myr duration, interpreted to be the result of dynamic topography.The boundary between M3 and M4 is a major erosional unconformity documenting the final continental break-up during the opening of the EAG. Above this, a pronounced Albian to Cenomanian/Turonian marine transgression resulted in marine inundation of the plateau. Structural deformation continued into the early Cenomanian along the Guinea Marginal Ridge, a potential structural barrier that restricted marine connection across the EAG. Bulk geochemical data from the shallow Guinea Plateau indicate that enhanced carbon burial in this setting was primarily driven by the deposition of reworked, oxidized organic matter during an oceanic anoxic event, independent of gateway opening
The Least Action Admissibility Principle
No full textThis paper provides a new admissibility criterion for choosing physically relevant weak solutions of the equations of Lagrangian and continuum mechanics when non-uniqueness of solutions to the initial value problem occurs. The criterion is motivated by the classical least action principle but is now applied to initial value problems which exhibit non-unique solutions. Examples are provided for Lagrangian mechanics and the Euler equations of barotropic fluid mechanics. In particular, we show that the least action admissibility principle prefers the classical two shock solution to the Riemann initial value problem to certain solutions generated by convex integration. On the other hand, Dafermos’s entropy criterion prefers convex integration solutions to the two shock solutions. Furthermore, when the pressure is given by p(ρ)=ρ2, we show that the two shock solution is always preferred whenever the convex integration solutions are defined for the same initial data
Purpose versus profit: How institutions shape entrepreneurial success across countries
No full textWhat does ‘success’ mean for entrepreneurs in different institutional environments? Drawing on institutional theory, we explore how the success perceptions of entrepreneurs are shaped by their interpretations of the institutional environment across countries. Based on 87 semi-structured, in-depth interviews with digital start-ups in China, Germany, and the United Kingdom (UK), we find that success perceptions differ substantially across institutional contexts. Our findings suggest that success perceptions balance the individual preferences of entrepreneurs and the need to adapt to the institutional environment. We contribute to understanding on differences in entrepreneurship across countries by examining how institutions can influence entrepreneurial response strategies and outcomes. In addition, we provide a novel perspective on the role of entrepreneurial agency in the context of strong institutional influences
Characteristics of Wind Field Observed by Synthetic Aperture Radar and Microwave Radiometer in Tropical Cyclone
No full textWind field structure of tropical cyclone (TC) can be resolved by remotely sensed sensors operated at microwave frequency, i.e. synthetic aperture radar (SAR) and microwave radiometer. The main purpose of this study is to investigate the characteristics of TC wind observed by Sentinel-1 (S-1) and soil moisture active passive (SMAP) during over 100 TCs from 2016 to 2023. The swath coverage of CyclObs winds inverted from S-1 images is about 500 km with a spatial resolution of about 500 m, while the SMAP wind products have a swath coverage of 1000 km with spatial resolution of 0.25° grids. Three TC parameters, geographic location of eye, maximum wind speed and radius of maximum wind speed, 34, 50, and 64 knot (kt) wind radii are estimated from CyclObs and SMAP winds. It is found that the distance between TC eyes derived from these two remote-sensed products and those from International Best Track Archive for Climate Stewardship (IBTrACS) reanalysis increases with central pressure increasing. Compared with IBTrACS data, the Root Mean Squared Error (RMSE) and Correlation Coefficient (COR) of maximum wind speed by CyclObs are 9.94 m s–1 and 0.63, whereas those by SMAP are 12.34 m s–1 and 0.51. In contrast, about 23 m RMSE of radius of maximum wind speed and about 0.7 COR is achieved from both CyclObs and SMAP winds. Here, the angle is defined as follows: 0°−135° clockwise relative with TC movement represents the right side, 135°−225° represents the back side, and 0°−135° counterclockwise represents the left side. The angle between the maximum wind velocity direction relative to TC translation direction (forward direction) is not correlated well with NHC maximum wind speed or NHC radius of maximum wind speed. Interestingly, as the radius of maximum wind speed increases, the angle in CyclObs wind increases at the back and left sides of TC centre, whereas the angle decreases at the right side. This behaviour is only observed at the left side in SMAP wind
Distribution and origin of carbon dioxide in the East Irish Sea Basin: implications for carbon storage:implications for carbon storage
No full textThe natural occurrence of carbon dioxide in subsurface reservoirs is proof of concept that it can be securely stored over geological timescales. Gas accumulations naturally enriched in CO2 were identified in the East Irish Sea Basin, and their origin was evaluated using a large well and geophysical database. Legacy petroleum fluid samples indicate that CO2 is regionally negligible, except within the proximal North Morecambe and Rhyl gas fields in the northern basin. Despite relatively elevated ionic concentrations within the northern basin, interpretations of CO2 dissolution from formation water samples are not conclusive due to widespread contamination. Geochemical measurements of Carboniferous coal and shale samples indicate that units are typically mature and are lacking any further generative potential. While the accumulated CO2 may have been generated from Carboniferous limestones or formerly organic-rich units, this is likely to have been limited based on their burial history and widespread extent compared to the local present-day distribution of CO2. Instead, thick and densely spaced Paleogene igneous dykes were mapped near the Rhyl Field. Despite being the most likely origin, igneous intrusions are interpreted across the northern basin and near several accumulations that lack CO2, suggesting that other geological elements have influenced its contemporary distribution, such as the cap rock or migration
Low-Crosstalk and Independent Amplitude/ Polarization Control in Near- and Far-Fields using a Dielectric Metasurface
No full textMetasurfaces offer precise control over multidimensional light fields at subwavelength resolution, positioning them as powerful platforms for manipulating both near- and far-field optical distributions. Recent progress has concentrated on achieving simultaneous amplitude and polarization modulation in both fields using single-layer metasurfaces to increase information capacity. However, existing multiplexing techniques remain limited in enabling arbitrary, independent customization of amplitude and polarization characteristics across near- and far-fields within a single metasurface design. Here, a vectorial metasurface capable of fully decoupled near- and far-field multiplexing is presented, allowing independent control over amplitude and polarization in both spatial and spectral domains. By employing geometric-phase metasurfaces with four-nanopillar supercells, the generation of two distinct vectorial light fields with different amplitude and polarization distributions in the near- and far-field is experimentally demonstrated. This complete decoupling is achieved using a modified two-loop-iteration GS algorithm that simultaneously optimizes amplitudes and polarization profiles across both optical regimes. This approach establishes a new paradigm in multidimensional vector-field multiplexing, with applications spanning polarization-encoded encryption, complex vector beam generation, and high-density data storage
A novel kinetically-driven approach to forming columnar {110}-textured lithium metal anodes with extended cycle life
No full textConstructing {110}‐textured lithium (Li) metal anodes is a promising strategy to extend battery life. While preparation of such anodes has been the subject of a few studies, their focus has been exclusively on thermodynamically‐driven (equilibrium) strategies. Through a systematic screening of bath conditions, the study reported here identifies a novel kinetically‐driven protocol that enhances the volume fraction of {110} texture by more than fivefold compared to equilibrium approaches. The protocol involves Li deposition at high current densities or low temperatures in a commonly used LiNO3‐containing ether‐based electrolyte. Columnar {110}‐oriented grains are formed through a growth rate selection process arising from the stronger electronic coupling and faster electron transfer rate between Li(110) and Li+ cations compared to other lattice planes. LiNO3 plays a crucial role by inhibiting the deposition on the Li(110) plane less than the other planes. Simple bath condition adjustments yield optimized {110}‐textured Li anodes with improved plating/stripping homogeneity that suppresses dendrite formation and electrolyte consumption, resulting in extended cycle life in lean‐electrolyte full cells. This kinetically‐driven approach offers mechanistic insight into Li texture formation and a promising route to high‐performance Li metal anodes
Access Repository Dashboard
Do you manage Heriot Watt Pure? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!