-
Particle production and hadronization temperature of the massive Schwinger model
Authors:
Laura Batini,
Lara Kuhn,
Jürgen Berges,
Stefan Floerchinger
Abstract:
We study the pair production, string breaking, and hadronization of a receding electron-positron pair using the bosonized version of the massive Schwinger model in quantum electrodynamics in 1+1 space-time dimensions. Specifically, we study the dynamics of the electric field in Bjorken coordinates by splitting it into a coherent field and its Gaussian fluctuations. We find that the electric field…
▽ More
We study the pair production, string breaking, and hadronization of a receding electron-positron pair using the bosonized version of the massive Schwinger model in quantum electrodynamics in 1+1 space-time dimensions. Specifically, we study the dynamics of the electric field in Bjorken coordinates by splitting it into a coherent field and its Gaussian fluctuations. We find that the electric field shows damped oscillations, reflecting pair production. Interestingly, the computation of the asymptotic total particle density per rapidity interval for large masses can be fitted using a Boltzmann factor, where the temperature can be related to the hadronization temperature in QCD. Lastly, we discuss the possibility of an analog quantum simulation of the massive Schwinger model using ultracold atoms, explicitly matching the potential of the Schwinger model to the effective potential for the relative phase of two linearly coupled Bose-Einstein condensates.
△ Less
Submitted 7 June, 2024;
originally announced June 2024.
-
Informed Graph Learning By Domain Knowledge Injection and Smooth Graph Signal Representation
Authors:
Keivan Faghih Niresi,
Lucas Kuhn,
Gaëtan Frusque,
Olga Fink
Abstract:
Graph signal processing represents an important advancement in the field of data analysis, extending conventional signal processing methodologies to complex networks and thereby facilitating the exploration of informative patterns and structures across various domains. However, acquiring the underlying graphs for specific applications remains a challenging task. While graph inference based on smoo…
▽ More
Graph signal processing represents an important advancement in the field of data analysis, extending conventional signal processing methodologies to complex networks and thereby facilitating the exploration of informative patterns and structures across various domains. However, acquiring the underlying graphs for specific applications remains a challenging task. While graph inference based on smooth graph signal representation has become one of the state-of-the-art methods, these approaches usually overlook the unique properties of networks, which are generally derived from domain-specific knowledge. Overlooking this information could make the approaches less interpretable and less effective overall. In this study, we propose a new graph inference method that leverages available domain knowledge. The proposed methodology is evaluated on the task of denoising and imputing missing sensor data, utilizing graph signal reconstruction techniques. The results demonstrate that incorporating domain knowledge into the graph inference process can improve graph signal reconstruction in district heating networks. Our code is available at \href{https://github.com/Keiv4n/IGL}{github.com/Keiv4n/IGL}.
△ Less
Submitted 6 June, 2024;
originally announced June 2024.
-
Neutron phase filtering for separating phase- and attenuation signal in aluminium and anodic aluminium oxide
Authors:
Estrid Buhl Naver,
Okan Yetik,
Noémie Ott,
Matteo Busi,
Pavel Trtik,
Luise Theil Kuhn,
Markus Strobl
Abstract:
Neutron imaging has gained significant importance as a material characterisation technique and is particularly useful to visualise hydrogenous materials in objects opaque to other radiations. Particular fields of application include investigations of hydrogen in metals as well as metal corrosion, thanks to the fact that neutrons can penetrate metals better than e.g. X-rays and are at the same time…
▽ More
Neutron imaging has gained significant importance as a material characterisation technique and is particularly useful to visualise hydrogenous materials in objects opaque to other radiations. Particular fields of application include investigations of hydrogen in metals as well as metal corrosion, thanks to the fact that neutrons can penetrate metals better than e.g. X-rays and are at the same time highly sensitive to hydrogen. However at interfaces for example those that are prone to corrosion, refraction effects sometimes obscure the attenuation image, which is used to for hydrogen quantification. Refraction, as a differential phase effect, diverts the neutron beam away from the interface in the image which leads to intensity gain and intensity loss regions, which are superimposed to the attenuation image, thus obscuring the interface region and hindering quantitative analyses of e.g. hydrogen content in the vicinity of the interface or in an oxide layer. For corresponding effects in X-ray imaging, a phase filter approach was developed and is generally based on transport-of-intensity considerations. Here, we compare such an approach, that has been adapted to neutrons, with another simulation-based assessment using the ray-tracing software McStas. The latter appears superior and promising for future extensions which enable fitting forward models via simulations in order to separate phase and attenuation effects and thus pave the way for overcoming quantitative limitations at refracting interfaces.
△ Less
Submitted 23 May, 2024;
originally announced May 2024.
-
Towards Real-Time Urban Physics Simulations with Digital Twins
Authors:
Jacopo Bonari,
Lisa Kuehn,
Max von Danwitz,
Alexander Popp
Abstract:
Urban populations continue to grow, highlighting the critical need to safeguard civilians against potential disruptions, such as dangerous gas contaminant dispersion. The digital twin (DT) framework offers promise in analyzing and predicting such events. This study presents a computational framework for modelling airborne contaminant dispersion in built environments. Leveraging automatic generatio…
▽ More
Urban populations continue to grow, highlighting the critical need to safeguard civilians against potential disruptions, such as dangerous gas contaminant dispersion. The digital twin (DT) framework offers promise in analyzing and predicting such events. This study presents a computational framework for modelling airborne contaminant dispersion in built environments. Leveraging automatic generation of computational domains and solution processes, the proposed framework solves the underlying physical model equations with the finite element method (FEM) for numerical solutions. Model order reduction (MOR) methods are investigated to enhance computational efficiency without compromising accuracy. The study outlines the automatic model generation process, the details of the employed model, and the future perspectives for the realization of a DT. Throughout this research, the aim is to develop a reliable predictive model combining physics and data in a hybrid DT to provide informed real-time support within evacuation scenarios.
△ Less
Submitted 16 May, 2024;
originally announced May 2024.
-
Broad-line region geometry from multiple emission lines in a single-epoch spectrum
Authors:
L. Kuhn,
J. Shangguan,
R. Davies,
A. W. S. Man,
Y. Cao,
J. Dexter,
F. Eisenhauer,
N. M. Förster Schreiber,
H. Feuchtgruber,
R. Genzel,
S. Gillessen,
S. Hönig,
D. Lutz,
H. Netzer,
T. Ott,
S. Rabien,
D. J. D. Santos,
T. Shimizu,
E. Sturm,
L. J. Tacconi
Abstract:
The broad-line region (BLR) of active galactic nuclei (AGNs) traces gas close to the central supermassive black hole (BH). Recent reverberation map** (RM) and interferometric spectro-astrometry data have enabled detailed investigations of the BLR structure and dynamics, as well as estimates of the BH mass. These exciting developments motivate comparative investigations of BLR structures using di…
▽ More
The broad-line region (BLR) of active galactic nuclei (AGNs) traces gas close to the central supermassive black hole (BH). Recent reverberation map** (RM) and interferometric spectro-astrometry data have enabled detailed investigations of the BLR structure and dynamics, as well as estimates of the BH mass. These exciting developments motivate comparative investigations of BLR structures using different broad emission lines. In this work, we have developed a method to simultaneously model multiple broad lines of the BLR from a single-epoch spectrum. We apply this method to the five strongest broad emission lines (H$α$, H$β$, H$γ$, Pa$β$, and He $I\;λ$5876) in the UV-to-NIR spectrum of NGC 3783, a nearby Type I AGN which has been well studied by RM and interferometric observations. Fixing the BH mass to the published value, we fit these line profiles simultaneously to constrain the BLR structure. We find that the differences between line profiles can be explained almost entirely as being due to different radial distributions of the line emission. We find that using multiple lines in this way also enables one to measure some important physical parameters, such as the inclination angle and virial factor of the BLR. The ratios of the derived BLR time lags are consistent with the expectation of theoretical model calculations and RM measurements.
△ Less
Submitted 22 January, 2024;
originally announced January 2024.
-
On the Non-Associativity of Analog Computations
Authors:
Lisa Kuhn,
Bernhard Klein,
Holger Fröning
Abstract:
The energy efficiency of analog forms of computing makes it one of the most promising candidates to deploy resource-hungry machine learning tasks on resource-constrained system such as mobile or embedded devices. However, it is well known that for analog computations the safety net of discretization is missing, thus all analog computations are exposed to a variety of imperfections of corresponding…
▽ More
The energy efficiency of analog forms of computing makes it one of the most promising candidates to deploy resource-hungry machine learning tasks on resource-constrained system such as mobile or embedded devices. However, it is well known that for analog computations the safety net of discretization is missing, thus all analog computations are exposed to a variety of imperfections of corresponding implementations. Examples include non-linearities, saturation effect and various forms of noise. In this work, we observe that the ordering of input operands of an analog operation also has an impact on the output result, which essentially makes analog computations non-associative, even though the underlying operation might be mathematically associative. We conduct a simple test by creating a model of a real analog processor which captures such ordering effects. With this model we assess the importance of ordering by comparing the test accuracy of a neural network for keyword spotting, which is trained based either on an ordered model, on a non-ordered variant, and on real hardware. The results prove the existence of ordering effects as well as their high impact, as neglecting ordering results in substantial accuracy drops.
△ Less
Submitted 25 September, 2023;
originally announced September 2023.
-
Measuring 3D tree imbalance of plant models using graph-theoretical approaches
Authors:
Sophie J. Kersting,
A. Luise Kühn,
Mareike Fischer
Abstract:
Imbalance in the 3D structure of plants can be an important indicator of insufficient light or nutrient supply, as well as excessive wind, (formerly present) physical barriers, neighbor or storm damage. It can also be a simple means to detect certain illnesses, since some diseases like the apple proliferation disease, an infection with the barley yellow dwarf virus or plant canker can cause abnorm…
▽ More
Imbalance in the 3D structure of plants can be an important indicator of insufficient light or nutrient supply, as well as excessive wind, (formerly present) physical barriers, neighbor or storm damage. It can also be a simple means to detect certain illnesses, since some diseases like the apple proliferation disease, an infection with the barley yellow dwarf virus or plant canker can cause abnormal growth, like \enquote{witches' brooms} or burls, resulting in a deviating 3D plant architecture. However, quantifying imbalance of plant growth is not an easy task, and it requires a mathematically sound 3D model of plants to which imbalance indices can be applied. Current models of plants are often based on stacked cylinders or voxel matrices and do not allow for measuring the degree of 3D imbalance in the branching structure of the whole plant.
On the other hand, various imbalance indices are readily available for so-called graph-theoretical trees and are frequently used in areas like phylogenetics and computer science. While only some basic ideas of these indices can be transferred to the 3D setting, graph-theoretical trees are a logical foundation for 3D plant models that allow for elegant and natural imbalance measures.
In this manuscript, our aim is thus threefold: We first present a new graph-theoretical 3D model of plants and discuss desirable properties of imbalance measures in the 3D setting. We then introduce and analyze eight different 3D imbalance indices and their properties. Thirdly, we illustrate all our findings using a data set of 63 bush beans. Moreover, we implemented all our indices in the publicly available \textsf{R}-software package \textsf{treeDbalance} accompanying this manuscript.
△ Less
Submitted 8 December, 2023; v1 submitted 26 July, 2023;
originally announced July 2023.
-
Inverse Design of All-dielectric Metasurfaces with Bound States in the Continuum
Authors:
Sergei Gladyshev,
Theodosios D. Karamanos,
Lina Kuhn,
Dominik Beutel,
Thomas Weiss,
Carsten Rockstuhl,
Andrey Bogdanov
Abstract:
Metasurfaces with bound states in the continuum (BICs) have proven to be a powerful platform for drastically enhancing light-matter interactions, improving biosensing, and precisely manipulating near- and far-fields. However, engineering metasurfaces to provide an on-demand spectral and angular position for a BIC remains a prime challenge. A conventional solution involves a fine adjustment of geom…
▽ More
Metasurfaces with bound states in the continuum (BICs) have proven to be a powerful platform for drastically enhancing light-matter interactions, improving biosensing, and precisely manipulating near- and far-fields. However, engineering metasurfaces to provide an on-demand spectral and angular position for a BIC remains a prime challenge. A conventional solution involves a fine adjustment of geometrical parameters, requiring multiple time-consuming calculations. In this work, to circumvent such tedious processes, we develop a physics-inspired, inverse design method on all-dielectric metasurfaces for an on-demand spectral and angular position of a BIC. Our suggested method predicts the core-shell particles that constitute the unit cell of the metasurface, while considering practical limitations on geometry and available materials. Our method is based on a smart combination of a semi-analytical solution, for predicting the required dipolar Mie coefficients of the meta-atom, and a machine learning algorithm, for finding a practical design of the meta-atom that provides these Mie coefficients. Although our approach is exemplified in designing a metasurface sustaining a BIC, it can, also, be applied to many more objective functions. With that, we pave the way toward a general framework for the inverse design of metasurfaces in specific and nanophotonic structures in general.
△ Less
Submitted 17 May, 2023; v1 submitted 17 May, 2023;
originally announced May 2023.
-
Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation
Authors:
Lorenz Kuhn,
Yarin Gal,
Sebastian Farquhar
Abstract:
We introduce a method to measure uncertainty in large language models. For tasks like question answering, it is essential to know when we can trust the natural language outputs of foundation models. We show that measuring uncertainty in natural language is challenging because of "semantic equivalence" -- different sentences can mean the same thing. To overcome these challenges we introduce semanti…
▽ More
We introduce a method to measure uncertainty in large language models. For tasks like question answering, it is essential to know when we can trust the natural language outputs of foundation models. We show that measuring uncertainty in natural language is challenging because of "semantic equivalence" -- different sentences can mean the same thing. To overcome these challenges we introduce semantic entropy -- an entropy which incorporates linguistic invariances created by shared meanings. Our method is unsupervised, uses only a single model, and requires no modifications to off-the-shelf language models. In comprehensive ablation studies we show that the semantic entropy is more predictive of model accuracy on question answering data sets than comparable baselines.
△ Less
Submitted 15 April, 2023; v1 submitted 19 February, 2023;
originally announced February 2023.
-
CLAM: Selective Clarification for Ambiguous Questions with Generative Language Models
Authors:
Lorenz Kuhn,
Yarin Gal,
Sebastian Farquhar
Abstract:
Users often ask dialogue systems ambiguous questions that require clarification. We show that current language models rarely ask users to clarify ambiguous questions and instead provide incorrect answers. To address this, we introduce CLAM: a framework for getting language models to selectively ask for clarification about ambiguous user questions. In particular, we show that we can prompt language…
▽ More
Users often ask dialogue systems ambiguous questions that require clarification. We show that current language models rarely ask users to clarify ambiguous questions and instead provide incorrect answers. To address this, we introduce CLAM: a framework for getting language models to selectively ask for clarification about ambiguous user questions. In particular, we show that we can prompt language models to detect whether a given question is ambiguous, generate an appropriate clarifying question to ask the user, and give a final answer after receiving clarification. We also show that we can simulate users by providing language models with privileged information. This lets us automatically evaluate multi-turn clarification dialogues. Finally, CLAM significantly improves language models' accuracy on mixed ambiguous and unambiguous questions relative to SotA.
△ Less
Submitted 20 February, 2023; v1 submitted 15 December, 2022;
originally announced December 2022.
-
Polychromatic neutron phase contrast imaging of weakly absorbing samples enabled by phase retrieval
Authors:
Maja Østergaard,
Estrid Buhl Naver,
Anders Kaestner,
Peter K. Willendrup,
Annemarie Brüel,
Henning Osholm Sørensen,
Jesper Skovhus Thomsen,
Søren Schmidt,
Henning Friis Poulsen,
Luise Theil Kuhn,
Henrik Birkedal
Abstract:
We demonstrate the use of a phase retrieval technique for propagation-based phase contrast neutron imaging with a polychromatic beam. This enables imaging samples with low absorption contrast and/or improving the signal-to-noise ratio to facilitate e.g. time resolved measurements. A metal sample, designed to be close to a pure phase object, and a bone sample with canals partially filled with D2O w…
▽ More
We demonstrate the use of a phase retrieval technique for propagation-based phase contrast neutron imaging with a polychromatic beam. This enables imaging samples with low absorption contrast and/or improving the signal-to-noise ratio to facilitate e.g. time resolved measurements. A metal sample, designed to be close to a pure phase object, and a bone sample with canals partially filled with D2O were used for demonstrating the technique. These samples were imaged with a polychromatic neutron beam followed by phase retrieval. For both samples the signal-to-noise ratio were significantly improved and in case of the bone sample, the phase retrieval allowed for separation of bone and D2O, which is important for example for in situ flow experiments. The use of deuteration-contrast avoids the use of chemical contrast enhancement and makes neutron imaging an interesting complementary method to X-ray imaging of bone.
△ Less
Submitted 4 October, 2022;
originally announced October 2022.
-
Comparing NED and SIMBAD classifications across the contents of nearby galaxies
Authors:
L. Kuhn,
M. Shubat,
P. Barmby
Abstract:
Cataloguing and classifying celestial objects is one of the fundamental activities of observational astrophysics. In this work, we compare the contents of two comprehensive databases, the NASA Extragalactic Database (NED) and Set of Identifications, Measurements and Bibliography for Astronomical Data (SIMBAD) in the vicinity of nearby galaxies. These two databases employ different classification s…
▽ More
Cataloguing and classifying celestial objects is one of the fundamental activities of observational astrophysics. In this work, we compare the contents of two comprehensive databases, the NASA Extragalactic Database (NED) and Set of Identifications, Measurements and Bibliography for Astronomical Data (SIMBAD) in the vicinity of nearby galaxies. These two databases employ different classification schemes -- one flat and one hierarchical -- and our goal was to determine the compatibility of classifications for objects in common. Searching both databases for objects within the respective isophotal radius of each of the ~1300 individual galaxies in the Local Volume Galaxy sample, we found that on average, NED contains about ten times as many entries as SIMBAD and about two thirds of SIMBAD objects are matched by position to a NED object, at 5 arcsecond tolerance. These quantities do not depend strongly on the properties of the parent galaxies. We developed an algorithm to compare individual object classifications between the two databases and found that 88% of the classifications agree; we conclude that NED and SIMBAD contain consistent information for sources in common in the vicinity of nearby galaxies. Because many galaxies have numerous sources contained only in one of NED or SIMBAD, researchers seeking the most complete picture of an individual galaxy's contents are best served by using both databases.
△ Less
Submitted 27 June, 2022;
originally announced June 2022.
-
Nanocriticality in the magnetic phase transition of CoO nanoparticles
Authors:
Machteld E. Kamminga,
Jonas Okkels Birk,
Jari í Hjøllum,
Henrik Jacobsen,
Jakob Lass,
Thorbjørn L. Koch,
Niels B. Christensen,
Christof Niedermayer,
Lukas Keller,
Luise Theil Kuhn,
Elisabeth T. Ulrikkeholm,
Erik Brok,
Cathrine Frandsen,
Kim Lefmann
Abstract:
The universal theory of critical phase transitions describes the critical behavior at second-order phase transitions in infinitely large systems. With the increased contemporary interest in nanoscale materials, we investigated CoO nanoparticles by means of neutron scattering and found how the theory of critical phenomena breaks down in the nanoscale regime. Using CoO as a model system, we have ide…
▽ More
The universal theory of critical phase transitions describes the critical behavior at second-order phase transitions in infinitely large systems. With the increased contemporary interest in nanoscale materials, we investigated CoO nanoparticles by means of neutron scattering and found how the theory of critical phenomena breaks down in the nanoscale regime. Using CoO as a model system, we have identified a size-dependent nanocritical temperature region close to the antiferromagnetic phase transition where the magnetic correlation length of the nanoparticles converges to a constant value, which is significantly smaller than that of the saturated state found at low temperatures. This is in clear contrast to the divergence around $T_{\rm N}$ observed for bulk systems. Our findings of nanocriticality in the magnetic phase transition is of great importance for the understanding of phase transitions at the nanoscale.
△ Less
Submitted 30 May, 2022; v1 submitted 27 May, 2022;
originally announced May 2022.
-
Manipulating organic semiconductor morphology with visible light
Authors:
Michael Korning Sorensen,
Anders Skovbo Gertsen,
Rocco Peter Fornari,
Binbin Zhou,
Peter Uhd Jepsen,
Edoardo Stanzani,
Shinhee Yun,
Marcial Fernandez Castro,
Matthias Schwartzkopf,
Alexandros Koutsioubas,
Piotr de Silva,
Moises Espindola Rodriguez,
Luise Theil Kuhn,
Jens Wenzel Andreasen
Abstract:
We present a method to manipulate the final morphology of roll-to-roll slot-die coated poly(3-hexylthiophene) (P3HT) by optically exciting the p-type polymer in solution while coating. Our results provide a comprehensive picture of the entire knowledge chain, from demonstrating how to apply our method to a fundamental understanding of the changes in morphology and physical properties induced by ex…
▽ More
We present a method to manipulate the final morphology of roll-to-roll slot-die coated poly(3-hexylthiophene) (P3HT) by optically exciting the p-type polymer in solution while coating. Our results provide a comprehensive picture of the entire knowledge chain, from demonstrating how to apply our method to a fundamental understanding of the changes in morphology and physical properties induced by exciting P3HT while coating. By combining results from density functional theory and molecular dynamics simulations with a variety of X-ray experiments, absorption spectroscopy, and THz spectroscopy, we demonstrate the relationship between morphology and physical properties of the thin film. Specifically, in P3HT films excited with light during deposition, we observe changes in crystallinity and texture with more face-on orientation and increased out-of-plane charge mobility.
△ Less
Submitted 28 February, 2022;
originally announced March 2022.
-
Tree balance indices: a comprehensive survey
Authors:
Mareike Fischer,
Lina Herbst,
Sophie Kersting,
Luise Kühn,
Kristina Wicke
Abstract:
Tree balance plays an important role in phylogenetics and other research areas, which is why several indices to measure tree balance have been introduced over the years. Nevertheless, a formal definition of what a balance index actually is and what makes it a useful measure of balance (or, in other cases, imbalance), has so far not been introduced in the literature. While the established indices a…
▽ More
Tree balance plays an important role in phylogenetics and other research areas, which is why several indices to measure tree balance have been introduced over the years. Nevertheless, a formal definition of what a balance index actually is and what makes it a useful measure of balance (or, in other cases, imbalance), has so far not been introduced in the literature. While the established indices all summarize the (im)balance of a tree in a single number, they vary in their definitions and underlying principles. It is the aim of the present manuscript to introduce formal definitions of balance and imbalance indices that classify desirable properties of such indices and to analyze and categorize established indices accordingly. In this regard, we review 19 established (im)balance indices from the literature, summarize their general, statistical and combinatorial properties (where known), prove numerous additional results and indicate directions for future research by making explicit open questions and gaps in the literature. We also prove that a few tree shape statistics that have been used to measure tree balance in the literature do not fulfill our definition of an (im)balance index, which might indicate that their properties are not as useful for practical purposes. Moreover, we show that five additional tree shape statistics from other contexts actually are tree (im)balance indices according to our definition. The manuscript is accompanied by the website \url{treebalance.wordpress.com} containing fact sheets of the discussed indices. Moreover, we introduce the software package \verb|treebalance| implemented in $\mathsf{R}$ that can be used to calculate all indices discussed.
△ Less
Submitted 9 November, 2023; v1 submitted 25 September, 2021;
originally announced September 2021.
-
A multimodal operando neutron study of the phase evolution in a graphite electrode
Authors:
Monica-Elisabeta Lăcătuşu,
Luise Theil Kuhn,
Rune E. Johnsen,
Patrick K. M. Tung,
Søren Schmidt,
Takenao Shinohara,
Ryoji Kiyanagi,
Anton S. Tremsin,
Nancy Elewa,
Robin Woracek,
Markus Strobl
Abstract:
Obtaining a complete picture of local processes still poses a significant challenge in battery research. Here we demonstrate an in-situ combination of multimodal neutron imaging with neutron diffraction for spatially resolved operando observations of the lithiation-delithiation of a graphite electrode in a Li-ion battery cell. Throughout the lithiation-delithiation process we image the Li distribu…
▽ More
Obtaining a complete picture of local processes still poses a significant challenge in battery research. Here we demonstrate an in-situ combination of multimodal neutron imaging with neutron diffraction for spatially resolved operando observations of the lithiation-delithiation of a graphite electrode in a Li-ion battery cell. Throughout the lithiation-delithiation process we image the Li distribution based on the local beam attenuation. Simultaneously, we observe the development of the lithiated graphite phases as a function of cycling time and electrode thickness and integral throughout its volume by diffraction contrast imaging and diffraction, respectively. While the conventional imaging data allows to observe the Li uptake in graphite already during the formation of the solid electrolyte interphase, diffraction indicates the onset and development of the Li insertion/extraction globally, which supports the local structural transformation observations by diffraction contrast imaging.
△ Less
Submitted 8 April, 2021;
originally announced April 2021.
-
Robustness to Pruning Predicts Generalization in Deep Neural Networks
Authors:
Lorenz Kuhn,
Clare Lyle,
Aidan N. Gomez,
Jonas Rothfuss,
Yarin Gal
Abstract:
Existing generalization measures that aim to capture a model's simplicity based on parameter counts or norms fail to explain generalization in overparameterized deep neural networks. In this paper, we introduce a new, theoretically motivated measure of a network's simplicity which we call prunability: the smallest \emph{fraction} of the network's parameters that can be kept while pruning without a…
▽ More
Existing generalization measures that aim to capture a model's simplicity based on parameter counts or norms fail to explain generalization in overparameterized deep neural networks. In this paper, we introduce a new, theoretically motivated measure of a network's simplicity which we call prunability: the smallest \emph{fraction} of the network's parameters that can be kept while pruning without adversely affecting its training loss. We show that this measure is highly predictive of a model's generalization performance across a large set of convolutional networks trained on CIFAR-10, does not grow with network size unlike existing pruning-based measures, and exhibits high correlation with test set loss even in a particularly challenging double descent setting. Lastly, we show that the success of prunability cannot be explained by its relation to known complexity measures based on models' margin, flatness of minima and optimization speed, finding that our new measure is similar to -- but more predictive than -- existing flatness-based measures, and that its predictions exhibit low mutual information with those of other baselines.
△ Less
Submitted 10 March, 2021;
originally announced March 2021.
-
Efficient Smoothing of Dilated Convolutions for Image Segmentation
Authors:
Thomas Ziegler,
Manuel Fritsche,
Lorenz Kuhn,
Konstantin Donhauser
Abstract:
Dilated Convolutions have been shown to be highly useful for the task of image segmentation. By introducing gaps into convolutional filters, they enable the use of larger receptive fields without increasing the original kernel size. Even though this allows for the inexpensive capturing of features at different scales, the structure of the dilated convolutional filter leads to a loss of information…
▽ More
Dilated Convolutions have been shown to be highly useful for the task of image segmentation. By introducing gaps into convolutional filters, they enable the use of larger receptive fields without increasing the original kernel size. Even though this allows for the inexpensive capturing of features at different scales, the structure of the dilated convolutional filter leads to a loss of information. We hypothesise that inexpensive modifications to Dilated Convolutional Neural Networks, such as additional averaging layers, could overcome this limitation. In this project we test this hypothesis by evaluating the effect of these modifications for a state-of-the art image segmentation system and compare them to existing approaches with the same objective. Our experiments show that our proposed methods improve the performance of dilated convolutions for image segmentation. Crucially, our modifications achieve these results at a much lower computational cost than previous smoothing approaches.
△ Less
Submitted 19 March, 2019;
originally announced March 2019.
-
Patient Risk Assessment and Warning Symptom Detection Using Deep Attention-Based Neural Networks
Authors:
Ivan Girardi,
Pengfei Ji,
An-phi Nguyen,
Nora Hollenstein,
Adam Ivankay,
Lorenz Kuhn,
Chiara Marchiori,
Ce Zhang
Abstract:
We present an operational component of a real-world patient triage system. Given a specific patient presentation, the system is able to assess the level of medical urgency and issue the most appropriate recommendation in terms of best point of care and time to treat. We use an attention-based convolutional neural network architecture trained on 600,000 doctor notes in German. We compare two approa…
▽ More
We present an operational component of a real-world patient triage system. Given a specific patient presentation, the system is able to assess the level of medical urgency and issue the most appropriate recommendation in terms of best point of care and time to treat. We use an attention-based convolutional neural network architecture trained on 600,000 doctor notes in German. We compare two approaches, one that uses the full text of the medical notes and one that uses only a selected list of medical entities extracted from the text. These approaches achieve 79% and 66% precision, respectively, but on a confidence threshold of 0.6, precision increases to 85% and 75%, respectively. In addition, a method to detect warning symptoms is implemented to render the classification task transparent from a medical perspective. The method is based on the learning of attention scores and a method of automatic validation using the same data.
△ Less
Submitted 27 September, 2018;
originally announced September 2018.
-
Three Dimensional Polarimetric Neutron Tomography of Magnetic Fields
Authors:
Morten Sales,
Markus Strobl,
Takenao Shinohara,
Anton Tremsin,
Luise Theil Kuhn,
William R. B. Lionheart,
Naeem M. Desai,
Anders Bjorholm Dahl,
Søren Schmidt
Abstract:
Through the use of Time-of-Flight Three Dimensional Polarimetric Neutron Tomography (ToF 3DPNT) we have for the first time successfully demonstrated a technique capable of measuring and reconstructing three dimensional magnetic field strengths and directions unobtrusively and non-destructively with the potential to probe the interior of bulk samples which is not amenable otherwise.
Using a pione…
▽ More
Through the use of Time-of-Flight Three Dimensional Polarimetric Neutron Tomography (ToF 3DPNT) we have for the first time successfully demonstrated a technique capable of measuring and reconstructing three dimensional magnetic field strengths and directions unobtrusively and non-destructively with the potential to probe the interior of bulk samples which is not amenable otherwise.
Using a pioneering polarimetric set-up for ToF neutron instrumentation in combination with a newly developed tailored reconstruction algorithm, the magnetic field generated by a current carrying solenoid has been measured and reconstructed, thereby providing the proof-of-principle of a technique able to reveal hitherto unobtainable information on the magnetic fields in the bulk of materials and devices, due to a high degree of penetration into many materials, including metals, and the sensitivity of neutron polarisation to magnetic fields. The technique puts the potential of the ToF time structure of pulsed neutron sources to full use in order to optimise the recorded information quality and reduce measurement time.
△ Less
Submitted 2 February, 2018; v1 submitted 17 April, 2017;
originally announced April 2017.
-
Implicit Negative Feedback in Clinical Information Retrieval
Authors:
Lorenz Kuhn,
Carsten Eickhoff
Abstract:
In this paper, we reflect on ways to improve the quality of bio-medical information retrieval by drawing implicit negative feedback from negated information in noisy natural language search queries. We begin by studying the extent to which negations occur in clinical texts and quantify their detrimental effect on retrieval performance. Subsequently, we present a number of query reformulation and r…
▽ More
In this paper, we reflect on ways to improve the quality of bio-medical information retrieval by drawing implicit negative feedback from negated information in noisy natural language search queries. We begin by studying the extent to which negations occur in clinical texts and quantify their detrimental effect on retrieval performance. Subsequently, we present a number of query reformulation and ranking approaches that remedy these shortcomings by resolving natural language negations. Our experimental results are based on data collected in the course of the TREC Clinical Decision Support Track and show consistent improvements compared to state-of-the-art methods. Using our novel algorithms, we are able to reduce the negative impact of negations on early precision by up to 65%.
△ Less
Submitted 12 July, 2016;
originally announced July 2016.
-
Dynamic rotor mode in antiferromagnetic nanoparticles
Authors:
K. Lefmann,
H. Jacobsen,
J. Garde,
P. Hedegard,
A. Wischnewski,
S. N. Ancona,
H. S. Jacobsen,
C. R. H. Bahl,
L. Theil Kuhn
Abstract:
We present experimental, numerical, and theoretical evidence for a new mode of antiferromagnetic dynamics in nanoparticles. Elastic neutron scattering experiments on 8 nm particles of hematite display a loss of diffraction intensity with temperature, the intensity vanishing around 150 K. However, the signal from inelastic neutron scattering remains above that temperature, indicating a magnetic sys…
▽ More
We present experimental, numerical, and theoretical evidence for a new mode of antiferromagnetic dynamics in nanoparticles. Elastic neutron scattering experiments on 8 nm particles of hematite display a loss of diffraction intensity with temperature, the intensity vanishing around 150 K. However, the signal from inelastic neutron scattering remains above that temperature, indicating a magnetic system in constant motion. In addition, the precession frequency of the inelastic magnetic signal shows an increase above 100 K. Numerical Langevin simulations of spin dynamics reproduce all measured neutron data and reveal that thermally activated spin canting gives rise to a new type of coherent magnetic precession mode. This "rotor" mode can be seen as a high-temperature version of superparamagnetism and is driven by exchange interactions between the two magnetic sublattices. The frequency of the rotor mode behaves in fair agreement with a simple analytical model, based on a high temperature approximation of the generally accepted Hamiltonian of the system. The extracted model parameters, as the magnetic interaction and the axial anisotropy, are in excellent agreement with results from Mossbauer spectroscopy.
△ Less
Submitted 1 January, 2015;
originally announced January 2015.
-
Solitonic lattice and Yukawa forces in the rare earth orthoferrite TbFeO3
Authors:
Sergey Artyukhin,
Maxim Mostovoy,
Niels Paduraru Jensen,
Duc Le,
Karel Prokes,
Vinícus G. Paula,
Heloisa N. Bordallo,
Andrey Maljuk,
Sven Landsgesell,
Hanjo Ryll,
Bastian Klemke,
Sebastian Paeckel,
Klaus Kiefer,
Kim Lefmann,
Luise Theil Kuhn,
Dimitri N. Argyriou
Abstract:
The control of domains in ferroic devices lies at the heart of their potential for technological applications. Multiferroic materials offer another level of complexity as domains can be either or both of a ferroelectric and magnetic nature. Here we report the discovery of a novel magnetic state in the orthoferrite TbFeO3 using neutron diffraction under an applied magnetic field. This state has a v…
▽ More
The control of domains in ferroic devices lies at the heart of their potential for technological applications. Multiferroic materials offer another level of complexity as domains can be either or both of a ferroelectric and magnetic nature. Here we report the discovery of a novel magnetic state in the orthoferrite TbFeO3 using neutron diffraction under an applied magnetic field. This state has a very long incommensurate period ranging from 340 Angstrom at 3K to 2700 Angstrom at the lowest temperatures and exhibits an anomalously large number of higher-order harmonics, allowing us to identify it with the periodic array of sharp domain walls of Tb spins separated by many lattice constants. The Tb domain walls interact by exchanging spin waves propagating through the Fe magnetic sublattice. The resulting Yukawa-like force, familiar from particle physics, has a finite range that determines the period of the incommensurate state.
△ Less
Submitted 22 March, 2011;
originally announced March 2011.