Coarse-grained Level Research Articles

Dual-Grained Lightweight Strategy.

Removing redundant parameters and computations before the model training has attracted a great interest as it can effectively reduce the storage space of the model, speed up the training and inference of the model, and save energy consumption during the running of the model. In addition, the simplification of deep neural network models can enable high-performance network models to be deployed to resource-constrained edge devices, thus promoting the development of the intelligent world. However, current pruning at initialization methods exhibit poor performance at extreme sparsity. In order to improve the performance of the model under extreme sparsity, this paper proposes a dual-grained lightweight strategy-TEDEPR. This is the first time that TEDEPR has used tensor theory in the pruning at initialization method to optimize the structure of a sparse sub-network model and improve its performance. Specifically, first, at the coarse-grained level, we represent the weight matrix or weight tensor of the model as a low-rank tensor decomposition form and use multi-step chain operations to enhance the feature extraction capability of the base module to construct a low-rank compact network model. Second, unimportant weights are pruned at a fine-grained level based on the trainability of the weights in the low-rank model before the training of the model, resulting in the final compressed model. To evaluate the superiority of TEDEPR, we conducted extensive experiments on MNIST, UCF11, CIFAR-10, CIFAR-100, Tiny-ImageNet and ImageNet datasets with LeNet, LSTM, VGGNet, ResNet and Transformer architectures, and compared with state-of-the-art methods. The experimental results show that TEDEPR has higher accuracy, faster training and inference, and less storage space than other pruning at initialization methods under extreme sparsity.

Coarse-Graining the Recognition of a Glycolipid by the C-Type Lectin Mincle Receptor.

Macrophage inducible Ca2+-dependent lectin (Mincle) receptor recognizes Mycobacterium tuberculosis glycolipids to trigger an immune response. This host membrane receptor is thus a key player in the modulation of the immune response to infection by M. tuberculosis and has emerged as a promising target for the development of new vaccines against tuberculosis. The recent development of the Martini 3 force field for coarse-grained (CG) molecular modeling allows the study of interactions of soluble proteins with small ligands which was not typically modeled well with the previous Martini 2 model. Here, we present a refined approach detailing a protocol for modeling interactions between a glycolipid and its receptor at a CG level using the Martini 3 force field. Using this approach, we studied Mincle and identified critical parameters governing ligand recognition, such as loop flexibility and the regulation of hydrophobic groove formation by calcium ions. In addition, we assessed ligand affinity using free energy perturbation calculations. Our results offer mechanistic insight into the interactions between Mincle and glycolipids, providing a basis for the rational design of molecules targeting this type of membrane receptors.

Pseudo-label refinement via hierarchical contrastive learning for source-free unsupervised domain adaptation

Source-free unsupervised domain adaptation aims to adapt a source model to an unlabeled target domain without accessing the source data due to privacy considerations. Existing works mainly solve the problem by self-training methods and representation learning. However, these works typically learn the representation on a single semantic level and barely exploit the rich hierarchical semantic information to obtain clear decision boundaries, which makes it hard for these methods to achieve satisfactory generalization performance. In this paper, we propose a novel hierarchical contrastive domain adaptation algorithm that exploits self-supervised contrastive learning on both fine-grained instances and coarse-grained cluster semantics. On the one hand, we propose an adaptive prototype pseudo-labeling strategy to obtain much more reliable labels. On the other hand, we propose hierarchical contrastive representation learning on both fine-grained instance-wise level and coarse-grained cluster level to reduce the negative effect of label noise and stabilize the whole training procedure. Extensive experiments are conducted on primary unsupervised domain adaptation benchmark datasets, and the results demonstrate the effectiveness of the proposed method.

GraphBNC: Machine Learning-Aided Prediction of Interactions Between Metal Nanoclusters and Blood Proteins.

Hybrid nanostructures between biomolecules and inorganic nanomaterials constitute a largely unexplored field of research, with the potential for novel applications in bioimaging, biosensing, and nanomedicine. Developing such applications relies critically on understanding the dynamical properties of the nano-bio interface. This work introduces and validates a strategy to predict atom-scale interactions between water-soluble gold nanoclusters (AuNCs) and a set of blood proteins (albumin, apolipoprotein, immunoglobulin, and fibrinogen). Graph theory and neural networks are utilized to predict the strengths of interactions in AuNC-protein complexes on a coarse-grained level, which are then optimized in Monte Carlo-based structure search and refined to atomic-scale structures. The training data is based on extensive molecular dynamics (MD) simulations of AuNC-protein complexes, and the validating MD simulations show the robustness of the predictions. This strategy can be generalized to any complexes of inorganic nanostructures and biomolecules provided that one generates enough data about the interactions, and the bioactive parts of the nanostructure can be coarse-grainedrationally.

CFCN: An HLA-peptide Prediction Model based on Taylor Extension Theory and Multi-view Learning

Background: With the increasing development of biotechnology, many cancer solutions have been proposed nowadays. In recent years, Neo-peptides-based methods have made significant contributions, with an essential prerequisite of bindings between peptides and HLA molecules. However, the binding is hard to predict, and the accuracy is expected to improve further. Methods: Therefore, we propose the Crossed Feature Correction Network (CFCN) with deep learning method, which can automatically extract and adaptively learn the discriminative features in HLA-peptide binding, in order to make more accurate predictions on HLA-peptide binding tasks. With the fancy structure of encoding and feature extracting process for peptides, as well as the feature fusion process between fine-grained and coarse-grained level, it shows many advantages on given tasks. Results: The experiment illustrates that CFCN achieves better performances overall, compared with other fancy models in many aspects. Conclusion: In addition, we also consider to use multi-view learning methods for the feature fusion process, in order to find out further relations among binding features. Eventually, we encapsulate our model as a useful tool for further research on binding tasks.

A Method for Extracting Fine-Grained Knowledge of the Wheat Production Chain

The knowledge within wheat production chain data has multiple levels and complex semantic relationships, making it difficult to extract knowledge from them. Therefore, this paper proposes a fine-grained knowledge extraction method for the wheat production chain based on ontology. For the first time, the conceptual layers of ploughing, planting, managing, and harvesting were defined around the main agricultural activities of the wheat production chain. Based on this, the entities, relationships, and attributes in the conceptual layers were defined at a fine-grained level, and a spatial–temporal association pattern layer with four conceptual layers, twenty-eight entities, and forty-two relationships was constructed. Then, based on the characteristics of the self-constructed dataset, the Word2vec-BiLSTM-CRF model was designed for extracting the knowledge within it, i.e., the entity–relationship–attribute model and the Word2vec-BiLSTM-CRF model in this paper were compared with the four SOTA models. The results show that the accuracy and F1 value improved by 8.44% and 8.89%, respectively, compared with the BiLSTM-CRF model. Furthermore, the entities of the pest and disease dataset were divided into two different granularities for the comparison experiment; the results show that for entities with “disease names” and “pest names”, the recognition accuracy at the fine-grained level is improved by 32.71% and 31.58%, respectively, compared to the coarse-grained level, and the recognition performance of various fine-grained entities has been improved.

EAAC-Net: An Efficient Adaptive Attention and Convolution Fusion Network for Skin Lesion Segmentation.

Accurate segmentation of skin lesions in dermoscopic images is of key importance for quantitative analysis of melanoma. Although existing medical image segmentation methods significantly improve skin lesion segmentation, they still have limitations in extracting local features with global information, do not handle challenging lesions well, and usually have a large number of parameters and high computational complexity. To address these issues, this paper proposes an efficient adaptive attention and convolutional fusion network for skin lesion segmentation (EAAC-Net). We designed two parallel encoders, where the efficient adaptive attention feature extraction module (EAAM) adaptively establishes global spatial dependence and global channel dependence by constructing the adjacency matrix of the directed graph and can adaptively filter out the least relevant tokens at the coarse-grained region level, thus reducing the computational complexity of the self-attention mechanism. The efficient multiscale attention-based convolution module (EMA⋅C) utilizes multiscale attention for cross-space learning of local features extracted from the convolutional layer to enhance the representation of richly detailed local features. In addition, we designed a reverse attention feature fusion module (RAFM) to enhance the effective boundary information gradually. To validate the performance of our proposed network, we compared it with other methods on ISIC 2016, ISIC 2018, and PH2 public datasets, and the experimental results show that EAAC-Net has superior segmentation performance under commonly used evaluation metrics.

Faster but Not Sweeter: A Model of Escherichia coli Re-level Lipopolysaccharide for Martini 3 and a Martini 2 Version with Accelerated Kinetics.

Lipopolysaccharide (LPS) is a complex glycolipid molecule that is the main lipidic component of the outer leaflet of the outer membrane of Gram-negative bacteria. It has very limited lateral motion compared to phospholipids, which are more ubiquitous in biological membranes, including in the inner leaflet of the outer membrane of Gram-negative bacteria. The slow-moving nature of LPS can present a hurdle for molecular dynamics simulations, given that the (pragmatically) accessible timescales to simulations are currently limited to microseconds, during which LPS displays some conformational dynamics but hardly any lateral diffusion. Thus, it is not feasible to observe phenomena such as insertion of molecules, including antibiotics/antimicrobials, directly into the outer membrane from the extracellular side nor to observe LPS dissociating from proteins via molecular dynamics using currently available models at the atomistic and more coarse-grained levels of granularity. Here, we present a model of deep rough LPS compatible with the Martini 2 coarse-grained force field with scaled down nonbonded interactions to enable faster diffusion. We show that the faster-diffusing LPS model is able to reproduce the salient biophysical properties of the standard models, but due to its faster lateral motion, molecules are able to penetrate deeper into membranes containing the faster model. We show that the fast ReLPS model is able to reproduce experimentally determined patterns of interaction with outer membrane proteins while also allowing for LPS to associate and dissociate with proteins within microsecond timescales. We also complete the Martini 3 LPS toolkit for Escherichia coli by presenting a (standard) model of deep rough LPS for this force field.

Novel behavior-enhanced long- and short-term interest model for sequential recommendation

In the realm of modern recommender systems, user-item interaction data often exhibit sequential patterns in relation to various behaviors, such as clicks and purchases on e-commerce platforms. The objective of heterogeneous sequential recommendation (HSR) is to predict each user's next item of interest under a specific behavior based on these interactions. However, existing approaches to HSR struggle to fully capture item transition relationships, as they consider these relationships from a single dimension and at a coarse-grained level. To bridge this gap, we propose the novel Behavior-enhanced Long- and Short-term Interest (BLSI) model, which explores fine-grained item transition relationships in both local and global dimensionalities. At its core, BLSI incorporates a behavior-enhanced self-attention network (BSAN) to capture short-term user preferences. BSAN distinguishes the effects of different behaviors and considers cross-type behavior influences during the linear projection and attention score calculation stages. Additionally, BLSI employs a heterogeneous graph neural network (HGNN) to model long-term user interests by discriminatively aggregating the information of neighboring nodes according to their behavior transition relationships. Furthermore, a gating mechanism is implemented to adaptively fuse short- and long-term preferences for personalized recommendations. Extensive experimental results on three datasets demonstrate that BLSI significantly outperforms state-of-the-art recommendation methods, highlighting the advantages of leveraging sequentiality and behavioral heterogeneity.

Multi-grained contrastive representation learning for label-efficient lesion segmentation and onset time classification of acute ischemic stroke

Ischemic lesion segmentation and the time since stroke (TSS) onset classification from paired multi-modal MRI imaging of unwitnessed acute ischemic stroke (AIS) patients is crucial, which supports tissue plasminogen activator (tPA) thrombolysis decision-making. Deep learning methods demonstrate superiority in TSS classification. However, they often overfit task-irrelevant features due to insufficient paired labeled data, resulting in poor generalization. We observed that unpaired data are readily available and inherently carry task-relevant cues, but are less often considered and explored. Based on this, in this paper, we propose to fully excavate the potential of unpaired unlabeled data and use them to facilitate the downstream AIS analysis task. We first analyze the utility of features at the varied grain and propose a multi-grained contrastive learning (MGCL) framework to learn task-related prior representations from both coarse-grained and fine-grained levels. The former can learn global prior representations to enhance the location ability for the ischemic lesions and perceive the healthy surroundings, while the latter can learn local prior representations to enhance the perception ability for semantic relation between the ischemic lesion and other health regions. To better transfer and utilize the learned task-related representation, we designed a novel multi-task framework to simultaneously achieve ischemic lesion segmentation and TSS classification with limited labeled data. In addition, a multi-modal region-related feature fusion module is proposed to enable the feature correlation and synergy between multi-modal deep image features for more accurate TSS decision-making. Extensive experiments on the large-scale multi-center MRI dataset demonstrate the superiority of the proposed framework. Therefore, it is promising that it helps better stroke evaluation and treatment decision-making.

Holographic coarse-grained states and the necessity of perfect entanglement

In the framework of the holographic principle, focusing on a central concept, conditional mutual information, we construct a class of coarse-grained states, which are intuitively connected to a family of thread configurations. These coarse-grained states characterize the entanglement structure of holographic systems at a coarse-grained level. Importantly, these coarse-grained states can be used to further reveal nontrivial requirements for the holographic entanglement structure. Specifically, we employ these coarse-grained states to probe the entanglement entropies of disconnected regions and the entanglement wedge cross section dual to the inherent correlation in a bipartite mixed state. The investigations demonstrate the necessity of perfect tensor state entanglement. Moreover, in a certain sense, our work establishes the equivalence between the holographic entanglement of purification and the holographic balanced partial entropy. We also construct a thread configuration with the multiscale entanglement renormalization ansatz (MERA) structure, reexamining the connection between the MERA structure and kinematic space. Published by the American Physical Society 2024

Hypergraph-enhanced multi-interest learning for multi-behavior sequential recommendation

Learning dynamic user preference has become an increasingly important component for many online platforms (e.g., video-sharing sites, e-commerce systems) to make sequential recommendations. In these platforms, user–item interactive behaviors are often multi-typed (e.g., click, add-to-favorite, purchase) with complex cross-type behavior inter-dependencies. Consequently, the multi-behavior sequence recommendation (MBSR) is gaining growing attention to meet practical application needs. However, most of the existing MBSR methods have not adequately explored the latent multi-dimensional real interests and multi-order multi-behavior dependencies, hampering the accurate inference of user preferences and further limiting recommendation performance. To this end, we devise a Hypergraph-Enhanced Multi-interest Learning Framework (HEML) equipped with a time-sensitive sequence module and a temporal-free hypergraph module, i.e., to learn both multi-interest and multi-behavior dependencies. For multi-interest extraction, a dual-scale transformer is designed to encode sequence patterns from coarse-grained level to fine-grained level, respectively. A classic capsule network is then exploited to extract the hidden two-level multi-interests explicitly. An interest-matching mechanism is presented to further adaptively match the most relevant general interests of the users at the current time. For multi-behavior dependencies, a user-tailored multi-behavior hypergraph is established to capture global multi-order (e.g., triadic or even high-order) dependencies across behaviors. A lightweight hypergraph convolutional network is then designed to perform a two-stage refined ‘node-hyperedge-node’ feature transformation on the hypergraph structure. We also introduce a cross-view co-guided learning mechanism to encourage the aggregation of sequence and hypergraph information across views. Numerous empirical investigations conducted across three authentic datasets demonstrate the consistent superiority of HEML over a diverse array of recommendation methodologies. We have released the implementation code at https://github.com/Breeze-del/HEMLCODE.

Martini-Based Coarse-Grained Soil Organic Matter Model Derived from Atomistic Simulations.

The significance of soil organic matter (SOM) in environmental contexts, particularly its role in pollutant adsorption, has prompted an increased utilization of molecular simulations to understand microscopic interactions. This study introduces a coarse-grained SOM model, parametrized within the framework of the versatile Martini 3 force field. Utilizing models generated by the Vienna Soil Organic Matter Modeler 2, which constructs humic substance systems from a fragment database, we employed Swarm-CG to parametrize the fragments and subsequently assembled them into macromolecules. Direct Boltzmann inversion (DBI) facilitated the determination of bonded parameters between fragments. The parametrization yielded favorable agreement in the radius of gyration and solvent-accessible surface area. Transfer free energies exhibited a strong correlation with hexadecane-water and chloroform-water values, albeit deviations were noted for octanol-water values. Comparing densities of modeled Leonardite humic acid systems at coarse-grained and atomistic levels revealed promising agreement, particularly at higher water concentrations. The DBI approach effectively reproduced average values of bonded interactions between fragments. Radial distribution functions between carboxylate groups and calcium ions showed partial agreement, however, reproducing certain peaks was challenging due to fixed bead sizes. Detailed analysis of atomistic systems revealed different configurations between the groups, explaining discrepancies. The present contribution provides a comprehensive insight into the properties, strengths, and weaknesses of the coarse-grained SOM model, serving as a foundation for future investigations encompassing pollutant interactions and varied SOM compositions.

Simultaneous Determination of Four Catechins in Black Tea via NIR Spectroscopy and Feature Wavelength Selection: A Novel Approach.

As a non-destructive, fast, and cost-effective technique, near-infrared (NIR) spectroscopy has been widely used to determine the content of bioactive components in tea. However, due to the similar chemical structures of various catechins in black tea, the NIR spectra of black tea severely overlap in certain bands, causing nonlinear relationships and reducing analytical accuracy. In addition, the number of NIR spectral wavelengths is much larger than that of the modeled samples, and the small-sample learning problem is rather typical. These issues make the use of NIRS to simultaneously determine black tea catechins challenging. To address the above problems, this study innovatively proposed a wavelength selection algorithm based on feature interval combination sensitivity segmentation (FIC-SS). This algorithm extracts wavelengths at both coarse-grained and fine-grained levels, achieving higher accuracy and stability in feature wavelength extraction. On this basis, the study built four simultaneous prediction models for catechins based on extreme learning machines (ELMs), utilizing their powerful nonlinear learning ability and simple model structure to achieve simultaneous and accurate prediction of catechins. The experimental results showed that for the full spectrum, the ELM model has better prediction performance than the partial least squares model for epicatechin (EC), epicatechin gallate (ECG), epigallocatechin (EGC), and epigallocatechin gallate (EGCG). For the feature wavelengths, our proposed FIC-SS-ELM model enjoys higher prediction performance than ELM models based on other wavelength selection algorithms; it can simultaneously and accurately predict the content of EC (Rp2 = 0.91, RMSEP = 0.019), ECG (Rp2 = 0.96, RMSEP = 0.11), EGC (Rp2 = 0.97, RMSEP = 0.15), and EGCG (Rp2 = 0.97, RMSEP = 0.35) in black tea. The results of this study provide a new method for the quantitative determination of the bioactive components of black tea.

Effect of relative timescale on a system of particles sliding on a fluctuating energy landscape: Exact derivation of product measure condition.

We consider a system of hardcore particles advected by a fluctuating potential energy landscape, whose dynamics is in turn affected by the particles. Earlier studies have shown that as a result of two-way coupling between the landscape and the particles, the system shows an interesting phase diagram as the coupling parameters are varied. The phase diagram consists of various different kinds of ordered phases and a disordered phase. We introduce a relative timescale ω between the particle and landscape dynamics, and study its effect on the steady state properties. We find there exists a critical value ω=ω_{c} when all configurations of the system are equally likely in the steady state. We prove this result exactly in a discrete lattice system and obtain an exact expression for ω_{c} in terms of the coupling parameters of the system. We show that ω_{c} is finite in the disordered phase, diverges at the boundary between the ordered and disordered phase, and is undefined in the ordered phase. We also derive ω_{c} from a coarse-grained level description of the system using linear hydrodynamics. We start with the assumption that there is a specific value ω^{*} of the relative timescale when correlations in the system vanish, and mean-field theory gives exact expressions for the current Jacobian matrix A and compressibility matrix K. Our exact calculations show that Onsager-type current symmetry relation AK=KA^{T} can be satisfied if and only if ω^{*}=ω_{c}. Our coarse-grained model calculations can be easily generalized to other coupled systems.

MTG_CD: Multi-scale learnable transformation graph for fault classification and diagnosis in microservices

The rapid advancement of microservice architecture in the cloud has led to the necessity of effectively detecting, classifying, and diagnosing run failures in microservice applications. Due to the high dynamics of cloud environments and the complex dependencies between microservices, it is challenging to achieve robust real-time system fault identification. This paper proposes an interpretable fault diagnosis framework tailored for microservice architecture, namely Multi-scale Learnable Transformation Graph for Fault Classification and Diagnosis(MTG_CD). Firstly, we employ multi-scale neural transformation and graph structure adjacency matrix learning to enhance data diversity while extracting temporal-structural features from system monitoring metrics Secondly, a graph convolutional network (GCN) is utilized to fuse the extracted temporal-structural features in a multi-feature modeling approach, which helps to improve the accuracy of anomaly detection. To identify the root cause of system faults, we finally conduct a coarse-grained level diagnosis and exploration after obtaining the results of classifying the fault data. We evaluate the performance of MTG_CD on the microservice benchmark SockShop, demonstrating its superiority over several baseline methods in detecting CPU usage overhead, memory leak, and network delay faults. The average macro F1 score improves by 14.05%.

Diagnosis-Guided Deep Subspace Clustering Association Study for Pathogenetic Markers Identification of Alzheimer's Disease Based on Comparative Atlases.

The roles of brain region activities and genotypic functions in the pathogenesis of Alzheimer's disease (AD) remain unclear. Meanwhile, current imaging genetics methods are difficult to identify potential pathogenetic markers by correlation analysis between brain network and genetic variation. To discover disease-related brain connectome from the specific brain structure and the fine-grained level, based on the Automated Anatomical Labeling (AAL) and human Brainnetome atlases, the functional brain network is first constructed for each subject. Specifically, the upper triangle elements of the functional connectivity matrix are extracted as connectivity features. The clustering coefficient and the average weighted node degree are developed to assess the significance of every brain area. Since the constructed brain network and genetic data are characterized by non-linearity, high-dimensionality, and few subjects, the deep subspace clustering algorithm is proposed to reconstruct the original data. Our multilayer neural network helps capture the non-linear manifolds, and subspace clustering learns pairwise affinities between samples. Moreover, most approaches in neuroimaging genetics are unsupervised learning, neglecting the diagnostic information related to diseases. We presented a label constraint with diagnostic status to instruct the imaging genetics correlation analysis. To this end, a diagnosis-guided deep subspace clustering association (DDSCA) method is developed to discover brain connectome and risk genetic factors by integrating genotypes with functional network phenotypes. Extensive experiments prove that DDSCA achieves superior performance to most association methods and effectively selects disease-relevant genetic markers and brain connectome at the coarse-grained and fine-grained levels.

Advanced computational approaches to understand protein aggregation.

Protein aggregation is a widespread phenomenon implicated in debilitating diseases like Alzheimer's, Parkinson's, and cataracts, presenting complex hurdles for the field of molecular biology. In this review, we explore the evolving realm of computational methods and bioinformatics tools that have revolutionized our comprehension of protein aggregation. Beginning with a discussion of the multifaceted challenges associated with understanding this process and emphasizing the critical need for precise predictive tools, we highlight how computational techniques have become indispensable for understanding protein aggregation. We focus on molecular simulations, notably molecular dynamics (MD) simulations, spanning from atomistic to coarse-grained levels, which have emerged as pivotal tools in unraveling the complex dynamics governing protein aggregation in diseases such as cataracts, Alzheimer's, and Parkinson's. MD simulations provide microscopic insights into protein interactions and the subtleties of aggregation pathways, with advanced techniques like replica exchange molecular dynamics, Metadynamics (MetaD), and umbrella sampling enhancing our understanding by probing intricate energy landscapes and transition states. We delve into specific applications of MD simulations, elucidating the chaperone mechanism underlying cataract formation using Markov state modeling and the intricate pathways and interactions driving the toxic aggregate formation in Alzheimer's and Parkinson's disease. Transitioning we highlight how computational techniques, including bioinformatics, sequence analysis, structural data, machine learning algorithms, and artificial intelligence have become indispensable for predicting protein aggregation propensity and locating aggregation-prone regions within protein sequences. Throughout our exploration, we underscore the symbiotic relationship between computational approaches and empirical data, which has paved the way for potential therapeutic strategies against protein aggregation-related diseases. In conclusion, this review offers a comprehensive overview of advanced computational methodologies and bioinformatics tools that have catalyzed breakthroughs in unraveling the molecular basis of protein aggregation, with significant implications for clinical interventions, standing at the intersection of computational biology and experimental research.

Milestoning estimators of dissipation in systems observed at a coarse resolution

Many nonequilibrium, active processes are observed at a coarse-grained level, where different microscopic configurations are projected onto the same observable state. Such "lumped" observables display memory, and in many cases, the irreversible character of the underlying microscopic dynamics becomes blurred, e.g., when the projection hides dissipative cycles. As a result, the observations appear less irreversible, and it is very challenging to infer the degree of broken time-reversal symmetry. Here we show, contrary to intuition, that by ignoring parts of the already coarse-grained state space we may-via a process called milestoning-improve entropy-production estimates. We present diverse examples where milestoning systematically renders observations "closer to underlying microscopic dynamics" and thereby improves thermodynamic inference from lumped data assuming a given range of memory, and we hypothesize that this effect is quite general. Moreover, whereas the correct general physical definition of time reversal in the presence of memory remains unknown, we here show by means of physically relevant examples that at least for semi-Markov processes of first and second order, waiting-time contributions arising from adopting a naive Markovian definition of time reversal generally must be discarded.

SM-GRSNet: sparse mapping-based graph representation segmentation network for honeycomb lung lesion

Objective. Honeycomb lung is a rare but severe disease characterized by honeycomb-like imaging features and distinct radiological characteristics. Therefore, this study aims to develop a deep-learning model capable of segmenting honeycomb lung lesions from Computed Tomography (CT) scans to address the efficacy issue of honeycomb lung segmentation. Methods. This study proposes a sparse mapping-based graph representation segmentation network (SM-GRSNet). SM-GRSNet integrates an attention affinity mechanism to effectively filter redundant features at a coarse-grained region level. The attention encoder generated by this mechanism specifically focuses on the lesion area. Additionally, we introduce a graph representation module based on sparse links in SM-GRSNet. Subsequently, graph representation operations are performed on the sparse graph, yielding detailed lesion segmentation results. Finally, we construct a pyramid-structured cascaded decoder in SM-GRSNet, which combines features from the sparse link-based graph representation modules and attention encoders to generate the final segmentation mask. Results. Experimental results demonstrate that the proposed SM-GRSNet achieves state-of-the-art performance on a dataset comprising 7170 honeycomb lung CT images. Our model attains the highest IOU (87.62%), Dice(93.41%). Furthermore, our model also achieves the lowest HD95 (6.95) and ASD (2.47). Significance. The SM-GRSNet method proposed in this paper can be used for automatic segmentation of honeycomb lung CT images, which enhances the segmentation performance of Honeycomb lung lesions under small sample datasets. It will help doctors with early screening, accurate diagnosis, and customized treatment. This method maintains a high correlation and consistency between the automatic segmentation results and the expert manual segmentation results. Accurate automatic segmentation of the honeycomb lung lesion area is clinically important.