arXiv - PHYS - Statistical Mechanics最新文献_第5页

Input Space Mode Connectivity in Deep Neural Networks 深度神经网络的输入空间模式连接性

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-09 DOI: arxiv-2409.05800

Jakub Vrabel, Ori Shem-Ur, Yaron Oz, David Krueger

{"title":"Input Space Mode Connectivity in Deep Neural Networks","authors":"Jakub Vrabel, Ori Shem-Ur, Yaron Oz, David Krueger","doi":"arxiv-2409.05800","DOIUrl":"https://doi.org/arxiv-2409.05800","url":null,"abstract":"We extend the concept of loss landscape mode connectivity to the input space\u0000of deep neural networks. Mode connectivity was originally studied within\u0000parameter space, where it describes the existence of low-loss paths between\u0000different solutions (loss minimizers) obtained through gradient descent. We\u0000present theoretical and empirical evidence of its presence in the input space\u0000of deep networks, thereby highlighting the broader nature of the phenomenon. We\u0000observe that different input images with similar predictions are generally\u0000connected, and for trained models, the path tends to be simple, with only a\u0000small deviation from being a linear path. Our methodology utilizes real,\u0000interpolated, and synthetic inputs created using the input optimization\u0000technique for feature visualization. We conjecture that input space mode\u0000connectivity in high-dimensional spaces is a geometric effect that takes place\u0000even in untrained models and can be explained through percolation theory. We\u0000exploit mode connectivity to obtain new insights about adversarial examples and\u0000demonstrate its potential for adversarial detection. Additionally, we discuss\u0000applications for the interpretability of deep networks.","PeriodicalId":501520,"journal":{"name":"arXiv - PHYS - Statistical Mechanics","volume":"181 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142196176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Systematic Coarse-Graining of Sequence-Dependent Structure and Elasticity of Double-Stranded DNA 系统粗粒化双链 DNA 的序列相关结构和弹性

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-09 DOI: arxiv-2409.05510

Enrico Skoruppa, Helmut Schiessel

{"title":"Systematic Coarse-Graining of Sequence-Dependent Structure and Elasticity of Double-Stranded DNA","authors":"Enrico Skoruppa, Helmut Schiessel","doi":"arxiv-2409.05510","DOIUrl":"https://doi.org/arxiv-2409.05510","url":null,"abstract":"Coarse-grained models have played an important role in the study of the\u0000behavior of DNA at length scales beyond a few hundred base pairs.\u0000Traditionally, these models have relied on structurally featureless and\u0000sequence-independent approaches, such as the twistable worm-like chain.\u0000However, research over the past decade has illuminated the substantial impact\u0000of DNA sequence even at the kilo-base pair scale. Several robust\u0000sequence-dependent models have emerged, capturing intricacies at the base\u0000pair-step level. Here we introduce an analytical framework for coarse-graining\u0000such models to the 2 to 40-base pair scale while preserving essential\u0000structural and dynamical features. These faithful coarse-grained\u0000parametrizations enable efficient sampling of large molecules. Rather than\u0000providing a fully parametrized model, we present the methodology and software\u0000necessary for mapping any base pair-step model to the desired level of\u0000coarse-graining. Finally, we provide application examples of our method,\u0000including estimates of the persistence length and effective torsional stiffness\u0000of DNA in a setup mimicking a freely orbiting tweezer, as well as simulations\u0000of intrinsically helical DNA.","PeriodicalId":501520,"journal":{"name":"arXiv - PHYS - Statistical Mechanics","volume":"26 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142196179","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

On the motion of passive and active particles with harmonic and viscous forces 被动和主动粒子在谐波和粘性力作用下的运动

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-08 DOI: arxiv-2409.05164

Jae-Won Jung, Sung Kyu Seo, Kyungsik Kim

引用次数: 0

Fisher zeroes and dynamical quantum phase transitions for two- and three-dimensional models 二维和三维模型的费雪零点和动态量子相变

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-07 DOI: arxiv-2409.09070

Tomasz Masłowski, Hadi Cheraghi, Jesko Sirker, Nicholas Sedlmayr

引用次数: 0

Re-entrant percolation in active Brownian hard disks 活动布朗硬盘中的再入流渗流

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-06 DOI: arxiv-2409.04141

David Evans, José Martín-Roca, Nathan J. Harmer, Chantal Valeriani, Mark A. Miller

{"title":"Re-entrant percolation in active Brownian hard disks","authors":"David Evans, José Martín-Roca, Nathan J. Harmer, Chantal Valeriani, Mark A. Miller","doi":"arxiv-2409.04141","DOIUrl":"https://doi.org/arxiv-2409.04141","url":null,"abstract":"Non-equilibrium clustering and percolation are investigated in an archetypal\u0000model of two-dimensional active matter using dynamic simulations of\u0000self-propelled Brownian repulsive particles. We concentrate on the single-phase\u0000region up to moderate levels of activity, before motility-induced phase\u0000separation (MIPS) sets in. Weak activity promotes cluster formation and lowers\u0000the percolation threshold. However, driving the system further out of\u0000equilibrium partly reverses this effect, resulting in a minimum in the critical\u0000density for the formation of system-spanning clusters and introducing\u0000re-entrant percolation as a function of activity in the pre-MIPS regime. This\u0000non-monotonic behaviour arises from competition between activity-induced\u0000effective attraction (which eventually leads to MIPS) and activity-driven\u0000cluster breakup. Using an adapted iterative Boltzmann inversion method, we\u0000derive effective potentials to map weakly active cases onto a passive\u0000(equilibrium) model with conservative attraction, which can be characterised by\u0000Monte Carlo simulations. While the active and passive systems have practically\u0000identical radial distribution functions, we find decisive differences in\u0000higher-order structural correlations, to which the percolation threshold is\u0000highly sensitive. For sufficiently strong activity, no passive pairwise\u0000potential can reproduce the radial distribution function of the active system.","PeriodicalId":501520,"journal":{"name":"arXiv - PHYS - Statistical Mechanics","volume":"40 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142196221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Time-dependent dynamics in the confined lattice Lorentz gas 约束晶格洛伦兹气体中随时间变化的动力学

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-06 DOI: arxiv-2409.04293

A. Squarcini, A. Tinti, P. Illien, O. Bénichou, T. Franosch

{"title":"Time-dependent dynamics in the confined lattice Lorentz gas","authors":"A. Squarcini, A. Tinti, P. Illien, O. Bénichou, T. Franosch","doi":"arxiv-2409.04293","DOIUrl":"https://doi.org/arxiv-2409.04293","url":null,"abstract":"We study a lattice model describing the non-equilibrium dynamics emerging\u0000from the pulling of a tracer particle through a disordered medium occupied by\u0000randomly placed obstacles. The model is considered in a restricted geometry\u0000pertinent for the investigation of confinement-induced effects. We analytically\u0000derive exact results for the characteristic function of the moments valid to\u0000first order in the obstacle density. By calculating the velocity\u0000autocorrelation function and its long-time tail we find that already in\u0000equilibrium the system exhibits a dimensional crossover. This picture is\u0000further confirmed by the approach of the drift velocity to its terminal value\u0000attained in the non-equilibrium stationary state. At large times the diffusion\u0000coefficient is affected by both the driving and confinement in a way that we\u0000quantify analytically. The force-induced diffusion coefficient depends\u0000sensitively on the presence of confinement. The latter is able to modify\u0000qualitatively the non-analytic behavior in the force observed for the unbounded\u0000model. We then examine the fluctuations of the tracer particle along the\u0000driving force. We show that in the intermediate regime superdiffusive anomalous\u0000behavior persists even in the presence of confinement. Stochastic simulations\u0000are employed in order to test the validity of the analytic results, exact to\u0000first order in the obstacle density and valid for arbitrary force and\u0000confinement.","PeriodicalId":501520,"journal":{"name":"arXiv - PHYS - Statistical Mechanics","volume":"6 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142196182","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Stochastic Scovil--Schulz-DuBois machine 随机斯科维尔--舒尔茨--杜博瓦机器

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-06 DOI: arxiv-2409.04124

Fei Liu

引用次数: 0

Stripe and Bubble Ratchets on Asymmetric Substrates 不对称基底上的条纹和气泡棘轮

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-06 DOI: arxiv-2409.04646

C. Reichhardt, C. J. O. Reichhardt

引用次数: 0

Quantum percolation on Lieb Lattices 利布网格上的量子渗流

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-06 DOI: arxiv-2409.04610

W. S. Oliveira, J. Pimentel de Lima, Raimundo R. dos Santos

引用次数: 0

Collective chemotactic search strategies 集体趋化搜索策略

arXiv - PHYS - Statistical Mechanics Pub Date : 2024-09-06 DOI: arxiv-2409.04262

Hugues Meyer, Adam Wysocki, Heiko Rieger

{"title":"Collective chemotactic search strategies","authors":"Hugues Meyer, Adam Wysocki, Heiko Rieger","doi":"arxiv-2409.04262","DOIUrl":"https://doi.org/arxiv-2409.04262","url":null,"abstract":"Chemotactic biological or synthetic active matter shapes its environment by\u0000secretions of chemical signals from its self-propelled constituents, like\u0000cells, organisms or active colloids. From this indirect interaction collective\u0000effects emerge that can be used by the agents to migrate collectively, to form\u0000patterns or to search for targets more efficiently. Here, we use paradigmatic\u0000models to study the efficiency of collective search strategies of a large group\u0000of motile agents that release during their movement repulsive auto-chemotactic\u0000signals forcing them to move away from high concentrations of the chemical\u0000clue. We show that the repulsive chemotactic interactions improve the search\u0000efficiency, measured by the mean first passage time to find a randomly located\u0000target, by orders of magnitude depending on the strength of the chemotactic\u0000coupling. The mechanism for this improvement relies on two factors: the\u0000increase of the persistence length due to the agent's self-interaction with its\u0000own chemotactic field and by a more homogeneous distribution of the agents due\u0000to their mutual indirect repulsion mediated by the chemotactic field. At\u0000stronger particle-field coupling the chemotactic searchers self-organize into\u0000ballistically moving bands reminiscent of search-chains formed in search and\u0000rescue operations, whose efficiency depends on the number of searchers\u0000involved. Our comprehensive study of collective search strategies of large\u0000groups of interacting agents is not only relevant for chemotactic active matter\u0000but also for a wide range of fields like ethology, information engineering,\u0000robotics, and social engineering.","PeriodicalId":501520,"journal":{"name":"arXiv - PHYS - Statistical Mechanics","volume":"26 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142196186","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0