Inference Process Research Articles

Automatic compliance checking of BIM models against quality standards based on ontology technology

Many countries have issued standards to stipulate the quality of BIM models. However, due to the complexity of BIM models and the textual expression of the standards, checking the quality of BIM models against the standards manually is challenging and time-consuming. In order to solve the problem, this paper presents an ontology-based approach for the automatic compliance checking of BIM models against the standards. Taking the Chinese standard as an example, it is represented in OWL to obtain an ontology, and SWRL is used for representing the checking rules. Then, an inference process is established to fulfill the checking. Finally, a prototype system is developed to validate the approach. It is concluded that the approach can be applied to check the quality of BIM models in IFC format automatically. This research contributes to the development of automatic compliance checking systems for the quality of BIM models against real standards.

Incorporating Tissue-Specific Gene Expression Data to Improve Chemical-Disease Inference of in Silico Toxicogenomics Methods.

In silico toxicogenomics methods are resource- and time-efficient approaches for inferring chemical-protein-disease associations with potential mechanism information for exploring toxicological effects. However, current in silico toxicogenomics systems make inferences based on only chemical-protein interactions without considering tissue-specific gene/protein expressions. As a result, inferred diseases could be overpredicted with false positives. In this work, six tissue-specific expression datasets of genes and proteins were collected from the Expression Atlas. Genes were then categorized into high, medium, and low expression levels in a tissue- and dataset-specific manner. Subsequently, the tissue-specific expression datasets were incorporated into the chemical-protein-disease inference process of our ChemDIS system by filtering out relatively low-expressed genes. By incorporating tissue-specific gene/protein expression data, the enrichment rate for chemical-disease inference was largely improved with up to 62.26% improvement. A case study of melamine showed the ability of the proposed method to identify more specific disease terms that are consistent with the literature. A user-friendly user interface was implemented in the ChemDIS system. The methodology is expected to be useful for chemical-disease inference and can be implemented for other in silico toxicogenomics tools.

3D Industrial anomaly detection via dual reconstruction network

Currently, 2D anomaly detection has demonstrated outstanding performance. However, 2D images limit the improvement of anomaly detection accuracy without utilizing depth information. Therefore, this paper proposes a Dual Reconstruction viAInpainting Network for 3D industrial anomaly detection (DRAIN). Firstly, we design a 3D reconstruction network using an encoder-decoder-based U-shaped network for processing RGB images and depth images. Subsequently, accurate anomaly segmentation is implemented through a 3D segmentation network. We introduce a lightweight MLP module to enhance segmentation performance to capture long-range dependencies in the reconstructed images. Furthermore, we propose a dual attention-based information entropy fusion module to expedite feature fusion in the inference process, aiming for enhanced deployment in the industry. Extensive experiments demonstrate that DRAIN achieves a 94.3% AUROC on the 3D anomaly detection dataset MVTec 3D-AD, surpassing other research methods.Graphical abstractOverall architecture for 3D industrial anomaly detection via dual reconstruction network

Convolutional neural network model compression method

In the field of deep learning, Convolutional Neural Networks (CNN) has become a focal point due to its multi-layered structure and wide application. The success of deep learning is due to the model has more layers and more parameters, which gives it a stronger nonlinear fitting ability. Traditionally, CNNs are primarily run on Central Processing Units (CPUs) and Graphics Processing Units (GPUs). However, CPUs have lower computational power, and GPUs consume a lot of energy. In contrast, Field-Programmable Gate Arrays (FPGAs) offer high parallelism, low power consumption, flexible programming, and rapid development cycles. These combined advantages make FPGAs more suitable for the forward inference processes of deep learning compared to other platforms. However, CNNs has the characteristics of parameter redundancy, the storage cost of deploying it on FPGA is too high. In order to apply CNNs to FPGA, we need to optimize CNNs for compression. Because after compression, This paper analyzes several specific cases of model compression for convolutional neural networks, and summarizes and compares the efficient methods of model compression. The results show that the model compression methods are mainly divided into the structure change of convolutional neural network and the quantization of parameters. The two compression methods can be cascaded at the same time to achieve a better optimization effect.

AssistDistil for Medical Image Segmentation

Deep learning models have demonstrated significant effectiveness in addressing intricate object segmentation and image classification tasks. Nevertheless, their widespread use is impeded by high computational demands, limiting their applicability on resource-constrained devices and in contexts like medical image segmentation. This paper proposes AssistDistil, a semi-knowledge distillation technique designed to facilitate the transfer of knowledge from a larger teacher network to a more compact student model. During the inference process, the student model works in conjunction with the teacher model by condensing the teacher model’s latent information into its own latent representation, thereby boosting its representational capacity. The effectiveness of the proposed approach is demonstrated for multiple case studies in medical image segmentation task of eye segmentation, skin lesion segmentation, and chest X-ray segmentation. Experimental results on the IIITD Cataract Surgery, HAM10000, PH2, Shenzhen and Montgomery chest X-ray datasets demonstrate the efficacy of the proposed approach both in terms of accuracy and computational cost. For example, in comparison to the AUNet-based teacher model, the proposed approach achieves a similar mIOU with only 0.5% of the model size. In the future, we plan to explore knowledge distillation approaches to improve the distillation process in case of large model capacity gap between teacher and student networks. With fewer parameters, we intend for the student model to attain performance comparable to that of the teacher model without additional assistance.

An Active Inference Agent for Modeling Human Translation Processes.

This paper develops an outline for a hierarchically embedded architecture of an artificial agent that models human translation processes based on principles of active inference (AIF) and predictive processing (PP). AIF and PP posit that the mind constructs a model of the environment which guides behavior by continually generating and integrating predictions and sensory input. The proposed model of the translation agent consists of three processing strata: a sensorimotor layer, a cognitive layer, and a phenomenal layer. Each layer consists of a network of states and transitions that interact on different time scales. Following the AIF framework, states are conditioned on observations which may originate from the environment and/or the embedded processing layer, while transitions between states are conditioned on actions that implement plans to optimize goal-oriented behavior. The AIF agent aims at simulating the variation in translational behavior under various conditions and to facilitate investigating the underlying mental mechanisms. It provides a novel framework for generating and testing new hypotheses of the translating mind.

Category-agnostic semantic edge detection by measuring neural representation randomness

Edge detection plays a fundamental role in computer vision tasks and gains wide applications. In particular, semantic edge detection recently draws more attention due to the high demand for a fine-grained understanding of visual scenes. However, detecting high-level semantic edges hidden in visual scenes is quite challenging. Existing semantic edge detection methods focus on category-aware semantic edges and require elaborate category annotations. Instead, we first propose the category-agnostic semantic edge detection task without additional semantic category annotations. To achieve this goal, we propose to utilize only edge position annotations and leverage the information randomness of semantic edges. Specifically, we align semantic edge positions to the ground truth by maximizing randomness on edge regions and minimizing randomness on non-edge regions in the training process. In the inference process, we first obtain neural representations by the trained network, and then generate semantic edges by measuring neural randomness. We evaluate our method by comparisons with alternative methods on two well-known datasets: Cityscapes (Cordts et al., 2016) and SBD (Hariharan et al., 2014). The results demonstrate our superiority over the alternatives, which is more significant under weak annotations. We also provide comprehensive mechanism studies to verify the generalizability, rationality, and validity of our working mechanism.

TopoDoE: a design of experiment strategy for selection and refinement in ensembles of executable gene regulatory networks

BackgroundInference of Gene Regulatory Networks (GRNs) is a difficult and long-standing question in Systems Biology. Numerous approaches have been proposed with the latest methods exploring the richness of single-cell data. One of the current difficulties lies in the fact that many methods of GRN inference do not result in one proposed GRN but in a collection of plausible networks that need to be further refined. In this work, we present a Design of Experiment strategy to use as a second stage after the inference process. It is specifically fitted for identifying the next most informative experiment to perform for deciding between multiple network topologies, in the case where proposed GRNs are executable models. This strategy first performs a topological analysis to reduce the number of perturbations that need to be tested, then predicts the outcome of the retained perturbations by simulation of the GRNs and finally compares predictions with novel experimental data.ResultsWe apply this method to the results of our divide-and-conquer algorithm called WASABI, adapt its gene expression model to produce perturbations and compare our predictions with experimental results. We show that our networks were able to produce in silico predictions on the outcome of a gene knock-out, which were qualitatively validated for 48 out of 49 genes. Finally, we eliminate as many as two thirds of the candidate networks for which we could identify an incorrect topology, thus greatly improving the accuracy of our predictions.ConclusionThese results both confirm the inference accuracy of WASABI and show how executable gene expression models can be leveraged to further refine the topology of inferred GRNs. We hope this strategy will help systems biologists further explore their data and encourage the development of more executable GRN models.

To Exit or Not to Exit: Cost-Effective Early-Exit Architecture Based on Markov Decision Process

Recently, studies on early-exit mechanisms have emerged to reduce the computational cost during the inference process of deep learning models. However, most existing early-exit architectures simply determine early exiting based only on a target confidence level in the prediction, without any consideration of the computational cost. Such an early-exit criterion fails to balance accuracy and cost, making it difficult to use in various environments. To address this problem, we propose a novel, cost-effective early-exit architecture in which an early-exit criterion is designed based on the Markov decision process (MDP). Since the early-exit decisions within an early-exit model are sequential, we model them as an MDP problem to maximize accuracy as much as possible while minimizing the computational cost. Then, we develop a cost-effective early-exit algorithm using reinforcement learning that solves the MDP problem. For each input sample, the algorithm dynamically makes early-exit decisions considering the relative importance of accuracy and computational cost in a given environment, thereby balancing the trade-off between accuracy and cost regardless of the environment. Consequently, it can be used in various environments, even in a resource-constrained environment. Through extensive experiments, we demonstrate that our proposed architecture can effectively balance the trade-off in different environments, while the existing architectures fail to do so since they focus only on reducing their cost while preventing the degradation of accuracy.

BagFormer: Better cross-modal retrieval via bag-wise interaction

In the field of cross-modal retrieval, single encoder models tend to perform better than dual encoder models, but they suffer from high latency and low throughput. In this paper, we propose a dual encoder model called BagFormer that utilizes bag-wise late interaction mechanism to improve re-rank performance without sacrificing latency and throughput. BagFormer achieves this by employing a bagging layer, which facilitates the transformation of text to an appropriate granularity. This not only mitigates the issue of modal granularity mismatch but also enables the integration of entity knowledge into the model. Our experiments have shown that BagFormerViT-B outperforms the traditional dual-encoder model CLIPViT-B by 7.97% in zero-shot settings. Under fine-tuned conditions, BagFormerViT-B demonstrates an even more significant improvement of 17.98% over CLIPViT-B. Moreover, BagFormer not only matches the performance of cutting-edge single-encoder models in cross-modal retrieval tasks but also provides efficient inference processes characterized by lower latency and higher throughput. Compared to single-encoder models, BagFormer can achieve a speedup ratio of 38.14 when re-ranking individual candidates. Code and models are available at github.com/howard-hou/BagFormer.

Perception and adaptation of receptive prosody in autistic adolescents

A fundamental aspect of language processing is inferring others’ minds from subtle variations in speech. The same word or sentence can often convey different meanings depending on its tempo, timing, and intonation–features often referred to as prosody. Although autistic children and adults are known to experience difficulty in making such inferences, the science remains unclear as to why. We hypothesize that detail-oriented perception in autism may interfere with the inference process if it lacks the adaptivity required to cope with the variability ubiquitous in human speech. Using a novel prosodic continuum that shifts the sentence meaning gradiently from a statement (e.g., “It’s raining”) to a question (e.g., “It’s raining?”), we have investigated the perception and adaptation of receptive prosody in autistic adolescents and two groups of non-autistic controls. Autistic adolescents showed attenuated adaptivity in categorizing prosody, whereas they were equivalent to controls in terms of discrimination accuracy. Combined with recent findings in segmental (e.g., phoneme) recognition, the current results provide the basis for an emerging research framework for attenuated flexibility and reduced influence of contextual feedback as a possible source of deficits that hinder linguistic and social communication in autism.

GeoAI Reproducibility and Replicability: A Computational and Spatial Perspective

GeoAI has emerged as an exciting interdisciplinary research area that combines spatial theories and data with cutting-edge AI models to address geospatial problems in a novel, data-driven manner. Although GeoAI research has flourished in the GIScience literature, its reproducibility and replicability (R&R), fundamental principles that determine the reusability, reliability, and scientific rigor of research findings, have rarely been discussed. This article aims to provide an in-depth analysis of this topic from both computational and spatial perspectives. We first categorize the major goals for reproducing GeoAI research, namely, validation (repeatability), learning and adapting the method for solving a similar or new problem (reproducibility), and examining the generalizability of the research findings (replicability). Each of these goals requires different levels of understanding of GeoAI, as well as different methods to ensure its success. We then discuss the factors that might cause the lack of R&R in GeoAI research, with an emphasis on (1) the selection and use of training data; (2) the uncertainty that resides in the GeoAI model design, training, deployment, and inference processes; and more important (3) the inherent spatial heterogeneity of geospatial data and processes. We use a deep learning-based image analysis task as an example to demonstrate the results’ uncertainty and spatial variance caused by different factors. The findings reiterate the importance of knowledge sharing, as well as the generation of a “replicability map” that incorporates spatial autocorrelation and spatial heterogeneity into consideration in quantifying the spatial replicability of GeoAI research.

An Efficient Sparse Recovery STAP Algorithm for Airborne Bistatic Radars Based on Atomic Selection under the Bayesian Framework

The traditional sparse recovery (SR) space-time adaptive processing (STAP) algorithms are greatly affected by grid mismatch, leading to poor performance in airborne bistatic radar clutter suppression. In order to address this issue, this paper proposes an SR STAP algorithm for airborne bistatic radars based on atomic selection under the Bayesian framework. This method adopts the idea of atomic selection for the process of Bayesian inference, continuously evaluating the contribution of atoms to the likelihood function to add or remove atoms, and then using the selected atoms to estimate the clutter support subspace and perform sparse recovery in the clutter support subspace. Due to the inherent sparsity of clutter signals, performing sparse recovery in the clutter support subspace avoids using a massive number of atoms from an overcomplete space-time dictionary, thereby greatly improving computational efficiency. In airborne bistatic radar scenarios where significant grid mismatch exists, this method can mitigate the performance degradation caused by grid mismatch by encrypting grid points. Since the sparse recovery is performed in the clutter support subspace, encrypting grid points does not lead to excessive computational burden. Additionally, this method integrates out the noise term under a new hierarchical Bayesian model, preventing the adverse effects caused by inaccurate noise power estimation during iterations in the traditional SR STAP algorithms, further enhancing its performance. Our simulation results demonstrate the high efficiency and superior clutter suppression performance and target detection performance of this method.

Temporal-Quality Ensemble Technique for Handling Image Blur in Packaging Defect Inspection.

Despite achieving numerous successes with surface defect inspection based on deep learning, the industry still faces challenges in conducting packaging defect inspections that include critical information such as ingredient lists. In particular, while previous achievements primarily focus on defect inspection in high-quality images, they do not consider defect inspection in low-quality images such as those containing image blur. To address this issue, we proposed a noble inference technique named temporal-quality ensemble (TQE), which combines temporal and quality weights. Temporal weighting assigns weights to input images by considering the timing in relation to the observed image. Quality weight prioritizes high-quality images to ensure the inference process emphasizes clear and reliable input images. These two weights improve both the accuracy and reliability of the inference process of low-quality images. In addition, to experimentally evaluate the general applicability of TQE, we adopt widely used convolutional neural networks (CNNs) such as ResNet-34, EfficientNet, ECAEfficientNet, GoogLeNet, and ShuffleNetV2 as the backbone network. In conclusion, considering cases where at least one low-quality image is included, TQE has an F1-score approximately 17.64% to 22.41% higher than using single CNN models and about 1.86% to 2.06% higher than an average voting ensemble.

Explainability analysis: An in-depth comparison between Fuzzy Cognitive Maps and LAMDA

Currently, it has become very relevant that machine learning techniques can provide an explanation of the results they generate, which is even more relevant in certain domains, called critical, such as health and energy, among others. For this reason, several explainability methods have been generated in the literature. Some of them are Local Interpretable Model-Agnostic Explanations (LIME) and Feature Importance, which have been used in a wide range of problems. The objective of this work is to analyze the explainability capacity of the Learning Algorithm for Multivariate Data Analysis (LAMDA) and Fuzzy Cognitive Maps (FCM), which have been used with great interest due to their interpretability, management of uncertainty during the inference process, simplicity in its use, among other things. For the development of this work, data from two critical domains have been considered, health (COVID-19 and Dengue datasets) and energy (energy price dataset), for which prediction/classification models have been developed using LAMDA and FCM techniques. Afterward, two explainability techniques were used to analyze the explainability provided by each method, one based on the LIME method and the other on the feature importance method, the latter adapted to our work by being based on the permutation of values. Finally, the work proposes two new explainability methods, one based on causal inference and the other on the degrees of membership of the variables to the classes. The latter, in particular, allows doing an explainability analysis by class. The new explainability methods reproduce the results of well-known explainability methods in the literature such as LIME and feature importance, with less execution cost, and also, with an explainability analysis by class. This work opens the doors to new work on class explainability. Furthermore, we see that machine learning approaches based on causal or fuzzy relationships are quite self-explanatory, but specific explainability methods such as those we propose in the work allow us to study particular aspects, such as highly important variables, which general explainability methods do not allow us to do, such as LIME and feature importance.

A dual-head output network attack detection and classification approach for multi-energy systems

In today’s digital age, multi-energy systems (MES) have become an indispensable part of the social infrastructure, providing people with diversified energy support such as electricity, gas, water and so on. However, with the increasing popularity and networking of MES, the network security threats they face are becoming more and more serious, especially the threat of network attacks. This makes it essential to detect attacks on MES and precisely classify attack types in order to establish effective defense strategies. In this paper, a Dual-Head output network attack detection and classification method based on parallel CNN-BiLSTM network is proposed. The method adopts a parallel structure and can process different aspects of information at the same time, speeding up the training and inference process of the whole network, making the system respond more quickly to potential network attacks, and improving real-time and efficiency. The multi-model fusion structure can give full play to the advantages of CNN and BiLSTM in processing different types of data, so that the system can capture attack characteristics more comprehensively in many aspects, and improve the overall detection and classification performance. The dual-head output not only improves the system’s ability to accurately detect attacks, but also can effectively classify different types of attacks in detail, which helps to formulate more targeted defense strategies. In addition, in order to effectively evaluate our proposed method, the network traffic data required for the experiment were collected in an environment very similar to the actual operating environment of a multi-energy system. Finally, the experiment verifies that our method can not only realize effective detection of network attacks, but also accurately classify different types of attacks.

A joint communication and computation design for semantic wireless communication with probability graph

In this paper, the problem of joint communication and computation design for probability graph-based semantic communication over wireless networks is investigated. In the considered model, the base station (BS) extracts the compressed small-sized semantic data by removing redundant information based on the shared knowledge base between the transceivers. In particular, the knowledge base is represented as a probability graph, which summarizes the statistic relations of massive knowledge graphs. On the user side, the compressed information is accurately inferred on the basis of the same probability graph as the BS. Although this approach brings additional computation resource consumption for semantic information extraction, it effectively reduces communication resource consumption through the transmission of small-sized data. Both the communication and computation cost models are derived based on the inference process of the probability graph. Based on the formulated models, the problem of joint communication and computation resource allocation is proposed to minimize the total energy consumption of the network considering both latency and power limitations. To solve this problem, the closed-form solution of the transmission power is obtained with fixed semantic compression ratio. Then, an effective linear search-based algorithm is proposed to obtain the optimal solution of the considered problem with low complexity. Simulation results demonstrate the effectiveness of the proposed system compared with the conventional non-semantic schemes.

FPGA-based tiling scheme and on-chip memory scheduling scheme for multi-branch semantic segmentation neural network accelerator

Abstract According to the characteristics of a multi-branch semantic segmentation neural network with a large amount of intermediate data during the inference process, we reduce the requirement of off-chip memory access and the redundant calculation due to tiling through better blocking scheme, better on-chip memory partitioning scheme, and on-chip memory scheduling scheme, improving the energy efficiency to 77.4 GOPS/W.

FP‐Flow: Feature pyramid flow model for fabric defect detection

AbstractFabric defect detection is a crucial aspect of fabric production. At present, deep learning detection methods mostly rely on supervised learning. To tackle this issue, this study proposes an unsupervised fabric defect detection approach based‐on normalising flow. The method only needs to train the mapping of the feature probability distribution of defect‐free samples to a Gaussian distribution. In the inference process, the location of defects can be determined by testing the distance between the probability distribution of image features and the estimated distribution. To adapt to the complex background and various minor defects of the fabric, a feature pyramid structure is adopted. Moreover, considering the gradient vanishing and network degradation caused by deep layers during training, a residual structure is incorporated into the model. Experimental results demonstrate that the feature pyramid flow model outperforms other methods in defect detection across multiple datasets, with an average score rate of 98.7% and 100% for pixel‐level area under the curve (AUC) receiver operating characteristic and image‐level AUC, respectively, compared to an average score rate of 91.1% and 85.4% for other methods.

Implementation of Tsukamoto Fuzzy Logic to Determine Sheep Livestock Population Based on Gender and Age Category in Aceh Tamiang Regency

This study implements the Tsukamoto Fuzzy Logic method to determine the sheep livestock population based on gender and age categories in Aceh Tamiang Regency. The population data of male and female sheep are categorized into three age groups: young, adolescent, and adult. The processes of fuzzification, fuzzy inference, and defuzzification are used to model the data and produce more accurate population predictions. Fuzzy rules are applied to consider the combination of membership levels of each age category and gender. The analysis results indicate that the sheep livestock population in Aceh Tamiang Regency is 5553,69007. The Tsukamoto Fuzzy Logic method has proven effective in handling uncertainty and variability in livestock population data, providing flexibility in complex data-driven decision-making. This study makes a significant contribution to the utilization of fuzzy logic methods for planning and managing livestock populations and can serve as a reference for policymakers in the livestock sector.