Inference Mechanism Research Articles

Vapor–liquid phase equilibrium prediction for mixtures of binary systems using graph neural networks

AbstractVapor–liquid phase equilibrium (VLE) plays a crucial role in chemical process design, process equipment control, and experimental process simulation. However, experimental acquisition of VLE data is a challenging and complex task. As an alternative to experimentation, VLE data prediction offers great convenience and utility. In this article, an artificial intelligence network is proposed to predict the temperature and the vapor phase composition of binary mixtures. We constructed a graph neural network (GNN) and designed an uncertainty‐aware learning and inference mechanism (UALF) in the prediction process. The model was tested on both a self‐constructed dataset and a publicly available dataset. The results demonstrate that the proposed method effectively reveals the phase equilibrium properties of the target data. This work presents a novel approach for predicting vapor–liquid phase equilibrium in binary systems and proposes innovative ideas for investigating phase equilibrium mechanisms and principles.

An intelligent font generation system based on stroke inference, mitigating production labor and enhancing design experience

Traditionally, font design has relied on manual craftsmanship by designers, a time-consuming and labor-intensive process that can take over a year to complete a new font family. Despite advancements in computer vision and graphics enabling the automation of font generation, creating high-quality fonts meeting commercial standards remains a technical challenge. Current automatic font generation technologies have not fully met production demands, mainly due to their lack of focus on generating glyphs that can be decomposed into strokes and their ineffective post-processing interaction, particularly for Chinese fonts. This study presents an innovative system for intelligently generating Chinese character fonts. The system utilizes a stroke database created by professional designers and combines font images generated through style transfer learning to perform stroke inference for font generation. The system’s core lies in its unique stroke inference mechanism, accurately identifying and matching strokes within font images to efficiently align with standard stroke data in the database. This approach not only improves the precision of font generation but also streamlines subsequent processing steps. Compared to traditional font design systems, our system shows significant advantages in generating fonts suitable for commercial use. It not only aids designers in enhancing work efficiency but also has the potential to greatly increase the production efficiency of font libraries. Moreover, the system’s design is scalable, offering extensive application prospects for future expansion to other East Asian scripts like Japanese and Korean.

Patient-centric knowledge graphs: a survey of current methods, challenges, and applications

Patient-Centric Knowledge Graphs (PCKGs) represent an important shift in healthcare that focuses on individualized patient care by mapping the patient’s health information holistically and multi-dimensionally. PCKGs integrate various types of health data to provide healthcare professionals with a comprehensive understanding of a patient’s health, enabling more personalized and effective care. This literature review explores the methodologies, challenges, and opportunities associated with PCKGs, focusing on their role in integrating disparate healthcare data and enhancing patient care through a unified health perspective. In addition, this review also discusses the complexities of PCKG development, including ontology design, data integration techniques, knowledge extraction, and structured representation of knowledge. It highlights advanced techniques such as reasoning, semantic search, and inference mechanisms essential in constructing and evaluating PCKGs for actionable healthcare insights. We further explore the practical applications of PCKGs in personalized medicine, emphasizing their significance in improving disease prediction and formulating effective treatment plans. Overall, this review provides a foundational perspective on the current state-of-the-art and best practices of PCKGs, guiding future research and applications in this dynamic field.

SIMULATION OF FUZZY LOGIC CONTROLLER AS DIRECTION ADJUSTMENT FOR LINE FOLLOWING ROBOT

Line Follower is a robot that follows a line, and in it there is an Integrated Circuit that contains programming to run it. In this experiment, the line follower using fuzzy logic control provides another alternative in the control system. In fuzzy logic control, there is no need for a mathematical model of the system because fuzzy logic control works based on rules. The designed control is a one input, two output system and uses the Mamdani system as an inference mechanism. The output error in the program will increase when the robot turns to follow the trajectory, the error can occur due to the calculation of the membership function input and rules as a determinant of robot action when detecting the trajectory in front of it. Therefore, in this line follower experiment, the rule will be a reference in processing the input value of the distance and speed of the right and left motors, the rule also determines the direction of the robot using the help of a camera sensor.

Early-Exit Deep Neural Network - A Comprehensive Survey

Deep neural networks (DNNs) typically have a single exit point that makes predictions by running the entire stack of neural layers. Since not all inputs require the same amount of computation to reach a confident prediction, recent research has focused on incorporating multiple ”exits” into the conventional DNN architecture. Early-exit DNNs are multi-exit neural networks that attach many side branches to the conventional DNN, enabling inference to stop early at intermediate points. This approach offers several advantages, including speeding up the inference process, mitigating the vanishing gradients problems, reducing overfitting and overthinking tendencies. It also supports DNN partitioning across devices and is ideal for multi-tier computation platforms such as edge computing. This paper decomposes the early-exit DNN architecture and reviews the recent advances in the field. The study explores its benefits, designs, training strategies, and adaptive inference mechanisms. Various design challenges, application scenarios, and future directions are also extensively discussed.

Explainable AI based Predictions for Workpiece Quality

It is well known that sound emissions of drilling and milling machines can be used to predict process and work piece quality. Deep Learning models have successfully been applied for that task. However, artificial neural networks are perceived as black boxes as the inference mechanisms of the network are not transparent. In this paper we present a Convolutional Neural Network trained on audio data with which process and product quality of a milling process can be controlled. The model was trained on 10-second audio snippets recorded from a milling machine that were converted into spectrograms. The trained model classifies the drilling process and predicts the expected workpiece quality. Additionally, rough estimates of the center rough and average roughness depth values are predicted. Moreover, we apply the Explainable AI methods that produce explanations by highlighting sections of the spectrogram that were relevant for the prediction. The spectrograms enriched with the explanations provide insights into the decision making process of the deep learning model in a human interpretable form. Patterns of how and why of a prediction become visible. Areas considered relevant by the model can be compared to the expectations of the human user. Through this explanation component, the model is not only able to make predictions but also to increase trust in the system which is an important aspect of acceptance of AI based quality assurance.

Meta-Reinforcement Learning in Nonstationary and Nonparametric Environments.

Recent state-of-the-art artificial agents lack the ability to adapt rapidly to new tasks, as they are trained exclusively for specific objectives and require massive amounts of interaction to learn new skills. Meta-reinforcement learning (meta-RL) addresses this challenge by leveraging knowledge learned from training tasks to perform well in previously unseen tasks. However, current meta-RL approaches limit themselves to narrow parametric and stationary task distributions, ignoring qualitative differences and nonstationary changes between tasks that occur in the real world. In this article, we introduce a Task-Inference-based meta-RL algorithm using explicitly parameterized Gaussian variational autoencoders (VAEs) and gated Recurrent units (TIGR), designed for nonparametric and nonstationary environments. We employ a generative model involving a VAE to capture the multimodality of the tasks. We decouple the policy training from the task-inference learning and efficiently train the inference mechanism on the basis of an unsupervised reconstruction objective. We establish a zero-shot adaptation procedure to enable the agent to adapt to nonstationary task changes. We provide a benchmark with qualitatively distinct tasks based on the half-cheetah environment and demonstrate the superior performance of TIGR compared with state-of-the-art meta-RL approaches in terms of sample efficiency (three to ten times faster), asymptotic performance, and applicability in nonparametric and nonstationary environments with zero-shot adaptation. Videos can be viewed at https://videoviewsite.wixsite.com/tigr.

Seismic data interpolation with recurrent inference mechanism

Abstract Seismic data interpolation is a significant procedure in seismic data processing as the highly complete data is extremely useful for the subsequent imaging and interpretation workflows. Recently, numerous deep learning based techniques have been utilized to help improve the reconstruction quality. Supervised learning approaches are able to predict the results in a second, but they depend heavily on the training data, which therefore limits the generalized applications. In contrast, unsupervised learning methods are applicable to any data, but they need much more computational time and prior knowledge to be implemented. Especially, the recovery of aliased and consecutively missing data is incredibly challenging. To solve this problem, we propose a novel framework for Seismic Interpolation using Recurrent Inference Mechanism(SIRIM). Integrating the advantages of supervised and unsupervised learning paradigms, we build a specific convolutional recurrent inference network such that it is able to learn the dynamic priors when plugged in the physics-informed iterative algorithm, as well as directly reconstruct various types of incomplete data. We validate SIRIM in comparison with some traditional and learning-based methods. The performance on synthetic and field data illustrates the effectiveness and robustness of our proposed approach which outperforms baseline methods in terms of aliased and consecutively missing data.

Structured light 3D shape measurement for translucent media base on deep Bayesian inference

Traditional structured light technique faces challenges in measuring translucent media due to low fringe modulation and strong random noise caused by subsurface scattering, thereby significantly reducing phase quality. In addition, the difficulty in obtaining ground truth makes it hard to assess reliability even though obtaining measured results. Here, we proposed a 3D measurement method for translucent media base on deep Bayesian inference to achieve both fringe enhancement and phase uncertainty evaluation. Specifically, a deep network incorporated with quatuor-branch residual block is developed to significantly enhance the fringe modulation and signal-to-noise ratio (SNR) for accurate phase recovery. Subsequently, a Bayesian inference mechanism is established for probabilistic statistics, which allows for the optimization of fringe output and provides uncertainty self-evaluation based on Monte Carlo (MC) sampling. Furthermore, by incorporating both numerical and physical constraints into the supervised learning, the network can effectively mitigate phase-shifted errors in the final results. The proposed method shows high efficiency and flexibility since it requires no additional patterns or hardware setup. Experiments validate the feasibility of the proposed method.

Multisensory perceptual and causal inference is largely preserved in medicated post-acute individuals with schizophrenia.

Hallucinations and perceptual abnormalities in psychosis are thought to arise from imbalanced integration of prior information and sensory inputs. We combined psychophysics, Bayesian modeling, and electroencephalography (EEG) to investigate potential changes in perceptual and causal inference in response to audiovisual flash-beep sequences in medicated individuals with schizophrenia who exhibited limited psychotic symptoms. Seventeen participants with schizophrenia and 23 healthy controls reported either the number of flashes or the number of beeps of audiovisual sequences that varied in their audiovisual numeric disparity across trials. Both groups balanced sensory integration and segregation in line with Bayesian causal inference rather than resorting to simpler heuristics. Both also showed comparable weighting of prior information regarding the signals' causal structure, although the schizophrenia group slightly overweighted prior information about the number of flashes or beeps. At the neural level, both groups computed Bayesian causal inference through dynamic encoding of independent estimates of the flash and beep counts, followed by estimates that flexibly combine audiovisual inputs. Our results demonstrate that the core neurocomputational mechanisms for audiovisual perceptual and causal inference in number estimation tasks are largely preserved in our limited sample of medicated post-acute individuals with schizophrenia. Future research should explore whether these findings generalize to unmedicated patients with acute psychotic symptoms.

Adversarial Variational Autoencoders to Extend and Improve Generative Model

Generative artificial intelligence (GenAI) has been advancing with many notable achievements like ChatGPT and Bard. The deep generative model (DGM) is a branch of GenAI, which is preeminent in generating raster data such as image and sound due to the strong role of deep neural networks (DNNs) in inference and recognition. The built-in inference mechanism of DNN, which simulates and aims at synaptic plasticity of the human neuron network, fosters the generation ability of DGM, which produces surprising results with the support of statistical flexibility. Two popular approaches in DGM are the variational autoencoder (VAE) and generative adversarial network (GAN). Both VAE and GAN have their own strong points although they share and imply the underlying theory of statistics as well as significant complex via hidden layers of DNN when DNN becomes effective encoding/decoding functions without concrete specifications. This research unifies VAE and GAN into a consistent and consolidated model called the adversarial variational autoencoder (AVA) in which the VAE and GAN complement each other; for instance, the VAE is a good data generator by encoding data via the excellent ideology of Kullback–Leibler divergence and the GAN is a significantly important method to assess the reliability of data as to whether it is real or fake. In other words, the AVA aims to improve the accuracy of generative models; besides, the AVA extends the function of simple generative models. In methodology, this research focuses on the combination of applied mathematical concepts and skillful techniques of computer programming in order to implement and solve complicated problems as simply as possible.

Ontology-Driven Automated Reasoning About Property Crimes

AbstractThe classification of police reports according to the typification of the criminal act described in them is not an easy task. The reports are written in natural language and often present missing, imprecise, or even inconsistent information, or lack sufficient details to make a clear decision. Focusing on property crimes, the aim of this work is to assist judges in this classification process by automatically extracting information from police reports and producing a list of possible classifications of crimes accompanied by a degree of confidence in each of them. The work follows the design science research methodology, developing a tool as an artifact. The proposal uses information extraction techniques to obtain the data from the reports, guided by an ontology developed for the Spanish legal system on property crimes. Probabilistic inference mechanisms are used to select the set of articles of the law that could apply to a given case, even when the evidence does not allow an unambiguous identification. The proposal has been empirically validated in a real environment with judges and prosecutors. The results show that the proposal is feasible and usable, and could be effective in assisting judges to classify property crime reports.

An Enhanced Adaptive Medical Diagnostic Model Driven By Genetic–Neuro-Fuzzy Algorithms

This research aims at developing adaptive medical diagnostic system driven by genetic, neural network and fuzzy logic algorithms for effective diagnosis of tuberculosis. Medical diagnosis for tuberculosis disease is a complex decision process that involves a lot of vagueness and uncertainty management, since the disease has multiple symptoms. The use of several algorithms has been explored in clinical diagnosis models for diagnosing and prescribing therapy for tuberculosis, but has challenges in transforming the vagueness and uncertainty of multiple symptoms due to the noisy nature of the disease data and over fitting of the models, this had led to the challenge of accurate classification, training, optimization, diagnosis and prescription of therapy. Object Oriented Analysis and Design (OOAD) and System Structured Analysis and Design (SSAD) methodologies were adopted in the research. The methodology involved obtaining four hundred and thirty (430) clinical data from patients with tuberculosis records at the Tuberculosis and Leprosy Hospital, Eku, Delta State. The obtained data were pre-processed using missing values imputation and numeric data encoding methods. Thereafter, a genetic algorithm was applied to the processed data to select six relevant features from the initial number of 16 features (Cough, Night sweats, Fever, Systolic Blood Pressure, Difficulty in Breathing, Loss of appetite, Sputum, Chills, Loss of pleasure, Immune Suppression, Chest Pain, Lack of concentration, Irritation, Loss of energy, Lymph Node Enlargement, Body Mass Index). The reduced dataset was split into training and test sets. The ANFIS (Adaptive Neuro-Fuzzy Inference System) was applied to perform diagnosis in which it was trained and validated on training and test sets respectively. The ANFIS was driven by Mamdani’s inference mechanism with sixty-four (64) generated rules with confidence and support scores of above 10% and 15% respectively. The developed model performance on the test sets gave an accuracy of 90% precision, sensitivity, and specificity of 100%. The results showed that all the attributes of tuberculosis contributed to the degree of the diseases based on their respective weights. Conclusively, the system accurately classified, trained attributes, and predicts the severity of tuberculosis.

Different mechanisms of contextual inference govern associatively learned and sensory-evoked postural responses

Human standing balance relies on the continuous monitoring and integration of sensory signals to infer our body's motion and orientation within the environment. However, when sensory information is no longer contextually relevant to balancing the body (e.g., when sensory and motor signals are incongruent), sensory-evoked balance responses are rapidly suppressed, much earlier than any conscious perception of changes in balance control. Here, we used a robotic balance simulator to assess whether associatively learned postural responses are similarly modulated by sensorimotor incongruence and contextual relevance to postural control. Twenty-nine participants in three groups were classically conditioned to generate postural responses to whole-body perturbations when presented with an initially neutral sound cue. During catch and extinction trials, participants received only the auditory stimulus but in different sensorimotor states corresponding to their group: 1) during normal active balance, 2) while immobilized, and 3) throughout periods where the computer subtly removed active control over balance. In the balancing and immobilized states, conditioned responses were either evoked or suppressed, respectively, according to the (in)ability to control movement. Following the immobilized state, conditioned responses were renewed when balance was restored, indicating that conditioning was retained but only expressed when contextually relevant. In contrast, conditioned responses persisted in the computer-controlled state even though there was no causal relationship between motor and sensory signals. These findings suggest that mechanisms responsible for sensory-evoked and conditioned postural responses do not share a single, central contextual inference and assessment of their relevance to postural control, and may instead operate in parallel.

How to reward animals based on their subjective percepts: A Bayesian approach to online estimation of perceptual biases.

Elucidating the neural basis of perceptual biases, such as those produced by visual illusions, can provide powerful insights into the neural mechanisms of perceptual inference. However, studying the subjective percepts of animals poses a fundamental challenge: unlike human participants, animals cannot be verbally instructed to report what they see, hear, or feel. Instead, they must be trained to perform a task for reward, and researchers must infer from their responses what the animal perceived. However, animals' responses are shaped by reward feedback, thus raising the major concern that the reward regimen may alter the animal's decision strategy or even intrinsic perceptual biases. We developed a method that estimates perceptual bias during task performance and then computes the reward for each trial based on the evolving estimate of the animal's perceptual bias. Our approach makes use of multiple stimulus contexts to dissociate perceptual biases from decision-related biases. Starting with an informative prior, our Bayesian method updates a posterior over the perceptual bias after each trial. The prior can be specified based on data from past sessions, thus reducing the variability of the online estimates and allowing it to converge to a stable estimate over a small number of trials. After validating our method on synthetic data, we apply it to estimate perceptual biases of monkeys in a motion direction discrimination task in which varying background optic flow induces robust perceptual biases. This method overcomes an important challenge to understanding the neural basis of subjective percepts.

Per-Instance Algorithm Configuration in Homogeneous Instance Spaces: A Use Case in Reconfigurable Assembly Systems

The physical capabilities of a reconfigurable assembly system (RAS) increase the agility and responsiveness of the system in highly volatile market conditions. However, achieving optimal RAS utilization entails solving complex optimization problems effectively and efficiently. These optimizations often define homogenous sets of problem instances. While algorithm configuration in such homogeneous contexts traditionally adopts a “one-size-fits-all” approach, recent studies have shown the potential of per-instance algorithm configuration (PIAC) methods in these settings. In this work, we evaluate and compare the performance of different PIAC methods in this context, namely Hydra—a state-of-the-art PIAC method—and a simpler case-based reasoning (CBR) approach. We evaluate the impact of the tuning time budget and/or the number of unique problem instances used for training on each of the method’s performance and robustness. Our experiments show that whilst Hydra fails to improve upon the default algorithm configuration, the CBR method can lead to 16% performance increase using as few as 100 training instances. Following these findings, we evaluate Hydra’s methodology when applied to homogenous instance spaces. This analysis shows the limitations of Hydra’s inference mechanisms in these settings and showcases the advantages of distance-based approaches used in CBR.

Researching the CNN Collaborative Inference Mechanism for Heterogeneous Edge Devices.

Convolutional Neural Networks (CNNs) have been widely applied in various edge computing devices based on intelligent sensors. However, due to the high computational demands of CNN tasks, the limited computing resources of edge intelligent terminal devices, and significant architectural differences among these devices, it is challenging for edge devices to independently execute inference tasks locally. Collaborative inference among edge terminal devices can effectively utilize idle computing and storage resources and optimize latency characteristics, thus significantly addressing the challenges posed by the computational intensity of CNNs. This paper targets efficient collaborative execution of CNN inference tasks among heterogeneous and resource-constrained edge terminal devices. We propose a pre-partitioning deployment method for CNNs based on critical operator layers, and optimize the system bottleneck latency during pipeline parallelism using data compression, queuing, and "micro-shifting" techniques. Experimental results demonstrate that our method achieves significant acceleration in CNN inference within heterogeneous environments, improving performance by 71.6% compared to existing popular frameworks.

An integrated knowledge and data model for adaptive diagnosis of lubricant conditions

Lubricant condition diagnosis often encounters conflicting conclusions due to the reverse degradation of indicators and coupling failures. To address this issue, a knowledge-guided adaptive diagnosis (KG-ADLC) method is developed with expert knowledge. Based on lubricant indicators in three attributes, a confidence-driven failure inference mechanism is developed by integrating rule-based reasoning with evidence theory. Furthermore, the KG-ADLC model is established that employs the inference mechanism to guide the synchronous evaluation of lubricant failures. For verification, the developed model is tested with the samples from the simulation experiment and authentic aero-engines. Experimental results reveal that the constructed model can accurately identify six types of lubricant conditions. Notably, the average accuracy has been improved from 74.8 % to 93.7 % when compared to existing methods.

Data-driven ergonomic assessment of construction workers

Automating ergonomic assessment improves both objectivity and cost-effectiveness. However, existing ergonomic scales remain susceptible to inaccuracy and insensitivity when assessing diversified construction activities. This susceptibility stems from the unreliable ranges, sharp boundaries, and oversimplified binary rules within conventional joint-level assessment rules, coupled with the restricted transformation rules to integer input/output. This paper presents a data-driven ergonomic assessment method grounded in statistical data from Construction Motion Data Library (CML) dataset, comprising: 1) Developing joint-level scoring models to primarily improve accuracy through leveraging Heuristics Gaussian Cloud Transformation (H-GCT), and 2) Designing fuzzy inference mechanism to enhance sensitivity through retaining risk information during transformation across levels. Subsequently, a three-step validation confirms accurate risk identification, sensitive risk feedback, and positive correlation with previous methods, contributing to occupational health and safety management. Future research could enhance the method by expanding the pose dataset, recruiting more participants, and refining the 3D pose estimation.

An AI-Based Classification and Recommendation System for Digital Libraries

The immense volume of online content linked to digital libraries has given emergence to the advancement of screening and recommendation systems. A recommendation system is vitally important in both academic institutions and elibraries to assist professors, instructors, students, and researchers in finding appropriate sources of information. Distributed or collaborative screening is the most common method used in current recommendation systems. However, collaborative approaches cannot promote library repositories, including unrated or unpurchased electronic information. Thus, this paper deals with the automated classification and recommendation of a multiclass corpus found in virtual repositories (cloud databases). In various stages, Neuro-Fuzzy (NF) and Support Vector Machine (SVM) techniques are used as the base classifiers for the categorization of the essential subjects (contents). Later, a high-level ensemble learning strategy is utilized to recommend appropriate subjects from the available multiclass corpus. The methods use a CoC (Coherence of Content)-based inference mechanism to extract and filter the critical components before beginning the recommendation process. Experiments demonstrated that a recommended approach based on detailed conceptual descriptions instead of a handful of phrases/words might help academic and research communities to find relevant sources. Observing the results over a period of months shows that the suggested method increases user comfort, proving the system’s acceptability to users in this way. In addition, compared to previous models, the accuracy in categorizing therequisite subjects is more than 97.16 per cent.