Task Model Research Articles

Informative relationship multi-task learning: Exploring pairwise contribution across tasks’ sharing knowledge

Multi-task learning is a machine learning paradigm, that aims to leverage useful domain information to help improve the generalization performance of all tasks. Learning the relationships of tasks helps to identify the latent tasks’ associations and access a better performance. However, most of the existing methods hardly pay attention to the determination of knowledge interaction among tasks and instead concentrate solely on certain aspects of task affinity. This compulsory similarity among all tasks leads to deficiencies in both task diversity and model robustness. To address this issue, we emphasize the task relationships within mutual information interaction. We propose a regularized framework from an informative perspective to quantify pairwise contributions during the knowledge-sharing stage, meanwhile utilizing an exclusive Lasso to identify the exclusive characteristics of tasks. An efficient optimization algorithm is developed to solve the proposed objective function. Detailed theoretical analyses and extensive experiments on both synthetic and real-world datasets are provided to demonstrate the effectiveness of our proposed method.

On a Method for Calculating Bending and Shear Vibrations of a Porous Piezoelement in the Low-Frequency Region

Introduction. Devices for collecting and storing energy from the external environment are low-power sources of electric energy that are actively used. The autonomous devices for monitoring the damaged condition of various structures include them as well. The working element of these devices is a piezoelectric generator (PEG) — a converter of mechanical energy into electrical energy. The design of PEG is associated with the preliminary construction of their mathematical and computer models, with the help of which the calculation and optimization of structures is carried out. One of the ways to model and calculate PEG is to develop approximate calculation methods based on applied theories. The applied theories for calculating bending vibrations of multilayer piezoactive plates are known and previously developed in the literature. However, in the scientific literature there is not enough information about bending and shear vibrations as a tool for improving the efficiency of engineering calculations of the described structures. The objective of this work was to develop an applied method for calculating bending and shear vibrations of piezoceramic plates, including porous ones.Materials and Methods. Piezoceramics PZT-4, including porous ones, were used as the piezoactive material of the plate. When using porous ceramics, the rigidity of the structure decreased to a greater extent than the piezoelectric modules, which made it possible to obtain a more effective PEG under mechanical action. The mathematical formulation was carried out within the framework of the linear theory of electroelasticity with plate polarization in thickness. The sides of the plate were electrodated, the right side was fixed, and a smooth contact in the vertical wall was set on the left side. Steady-state vibrations of the plate were caused by pressure on the front surfaces of the plate or the difference in electrical potentials at the electrodes. To calculate the characteristics of PEG, the authors proposed an applied theory based on hypotheses about the distribution of characteristics of the stress-strain state and the electric field.Results. Transverse vibrations of a piezoceramic plate in the low-frequency region (below the first bending-shear resonance) were studied. Due to the fact that the mathematical formulation was considered within the framework of the linear theory of elasticity, the problem was divided into the sum of two. The first one took into account the mechanical effect: a distributed load and a transverse force at the left end acted on the front surfaces of the plate, and the potentials at the electrodes were zero. In the second task, there were no mechanical loads, but the potential difference was set at the electrodes. Based on hypotheses about the distribution of deformations, mechanical stresses and electric potential, both problems were reduced to a system of ordinary differential equations and boundary conditions. Comparison with the results of calculations by the finite element method in the ACELAN package showed the adequacy of the proposed applied theory in the low-frequency region.Discussion and Conclusion. Since the formulation of the problem was considered in the linear theory of electroelasticity, and the low-frequency region was studied, the work succeeded in dividing the problem of bending-shear vibrations of a porous piezoceramic plate into two: bending — with mechanical action at zero potentials, and shear — when setting the potential difference and zero mechanical action. The corresponding hypotheses about bending and shear were used. Two systems of ordinary differential equations and boundary conditions, which were solved analytically without the use of “heavy” finite element packages, were constructed. To compare the results and confirm the adequacy of the proposed method, the finite element modeling of such tasks was carried out in a specialized ACELAN package. The comparison showed that the error in determining displacements and electric potential when using this approach, in the case of setting mechanical loads and potential differences, did not exceed 6%. The method developed in the paper can be applied in the design of piezoelectric generators for energy storage in the low-frequency region.

The MATERIAL framework: Modeling and AuTomatic code Generation of Edge Real-TIme AppLications under the QNX RTOS

Modern edge real-time automotive applications are becoming more complex, dynamic, and distributed, moving away from conventional static operating environments to support advanced driving assistance and autonomous driving functionalities. This shift necessitates formulating more complex task models to represent the evolving nature of these applications aptly. Modeling of real-time automotive systems is typically performed leveraging Architectural Languages (ALs) such as Amalthea, which are commonly used by the industry to describe the characteristics of processing platforms, operating systems, and tasks. However, these architectural languages are originally derived for classical automotive applications and need to evolve to meet the needs of next-generation applications. This paper proposes an automatic framework for the modeling and automatic code generation of dynamic automotive applications under the QNX RTOS. To this end, we extend Amalthea to describe chains of communicating tasks with multiple operating modes and to consider the QNX’s reservation-based scheduler, called APS, which allows providing temporal isolation between applications co-located on the same hardware platform. Finally, an evaluation is presented to compare different implementation alternatives under QNX that are automatically generated by our code generation framework.

PREPARATION OF FUTURE TEACHERS FOR TRAINING PHYSICS IN INSTITUTIONS OF PROFESSIONAL PRE-HIGHER EDUCATION AND VOCATIONAL EDUCATION

The article highlights relevant aspects of training future teachers of physics in institutions of professional pre-higher education, vocational education and modelling of practice-oriented tasks. The authors justified the importance of modernizing the content of relevant programmes of the "Physics" discipline, considering the profession of students at professional colleges and lyceums. It is noted that studying physics contributes to the formation of subject competencies of students, the ability to analyse, systematize, generalize information, overcome difficulties of adaptation to professional activity. Based on the analysis of scientific sources, the target, content, operational-activity, diagnostic-resultative blocks of the structural-functional model of training future teachers for teaching physics in institutions of professional pre-higher education and vocational education were selected. The target block is based on the socio-economic demand of society for qualified specialists. The content block contains pedagogical, psychological, methodical, and technological knowledge. Gnostic (research), design, organizational, and communication skills are highlighted in the structure of the operational-activity block. The diagnostic-resultative block will help the physics teacher to make corrections in the process of professional training of college students and vocational-technical lyceums. The importance of practice-oriented tasks in the process of learning physics for students of professional higher education and professional and technical education is indicated. Differences between practice-oriented tasks and other physical tasks are determined. The structure (name of the task, individually meaningful cognitive question, information, task involving work with information) and the algorithm of preparation for creating tasks of practical content in physics are presented. The methodology of teaching simulation of practice-oriented tasks in the process of studying the educational component "Methodology of teaching physics in a specialized school" is illustrated.

Iterative flexibility compensation based high precision trajectory tracking for industrial manipulators

High precision industrial manipulators are widely demanded in precision machining and advanced manufacturing. To realize high precision trajectory tracking in task space, feedback plus model based feedforward control strategies are widely used in modern industrial manipulators. Furthermore, the accuracy of feedforward model determines the accuracy of trajectory tacking to a great extent, especially for elastic manipulators. In traditional feedforward controllers, the flexibilities out of the rotational plane are often neglected and thus the end-effector tracking performance can be degraded a lot. Therefore, an iterative flexibility compensation based feedforward controller is proposed to improve the end-effector trajectory tracking accuracy of industrial manipulators in this paper. First, the flexible deformations in all directions are described accurately by using the extended flexible joint model (EFJM) for industrial manipulators. Second, a feedforward controller is designed by using the dynamic inversion of EFJM. In the feedforward controller, an iterative flexibility compensation scheme is designed to transform the reference Cartesian trajectory into joint space and thus the limitation of the nonminimum phase characteristics of EFJM can be avoided. Moreover, the flexible deformations are predicted and compensated based on EFJM. After that, the feedforward torque is derived in joint space by using the efficient feedforward torque computation algorithm for EFJM. Simulation and experimental results demonstrate the effectiveness of the proposed method.

Competency model development: The backbone of successful stealth assessments

AbstractBackgroundStealth assessment is a learning analytics method, which leverages the collection and analysis of learners' interaction data to make real‐time inferences about their learning. Employed in digital learning environments, stealth assessment helps researchers, educators, and teachers evaluate learners' competencies and customize the learning experience to their specific needs. This adaptability is closely intertwined with theories related to learning, engagement, and motivation. The foundation of stealth assessment rests on evidence‐cantered design (ECD), consisting of four core models: the Competency Model (CM), Evidence Model, Task Model, and Assembly Model.ObjectiveThe first step in designing a stealth assessment entails producing operational definitions of the constructs to be assessed. The CM establishes a framework of latent variables representing the target constructs, as well as their interrelations. When developing the CM, assessment designers must produce clear descriptions of the claims associated with the latent variables and their states, as well as sketch out how the competencies can be measured using assessment tasks. As the designers elaborate on the assessment model, the CM definitions need to be revisited to make sure they work with the scope and constraints of the assessment. Although this is the first step, problems at this stage may result in an assessment that does not meet the intended purpose. The objective of this paper is to elucidate the necessary steps for CM development and to highlight potential challenges in the process, along with strategies for addressing them, particularly for designers without much formal assessment experience.MethodThis paper is a methodological exposition, showcasing five examples of CM development. Specifically, we conducted a qualitative retrospective analysis of the CM development procedure, wherein participants unfamiliar with ECD applied the framework and showcased their work. In a stealth assessment course, four groups of students (novice stealth assessment designers) engaged in developing stealth assessments for challenging‐to‐measure constructs across four distinct projects. During their CM development process, we observed various activities to pinpoint areas of difficulty.ResultsThis paper presents five illustrative examples, including one for assessing physics understanding and four for the development of CMs for four complex competencies: (1) systems thinking, (2) online information credibility evaluation, (3) computational thinking, and (4) collaborative creativity. Each example represents a case in CM development, offering valuable insights.ConclusionThe paper concludes by discussing several guidelines derived from the examples discussed. Emphasizing the importance of dedicating ample time to fine‐tune CMs can significantly enhance the accuracy of assessments related to learners' knowledge and skills. It underscores the significance of qualitative phases in crafting comprehensive stealth assessments, such as CMs, alongside the quantitative statistical modeling and technical aspects of these assessments.

Synergizing Unsupervised and Supervised Learning: A Hybrid Approach for Accurate Natural Language Task Modeling

While supervised learning models have shown remarkable performance in various natural language processing (NLP) tasks, their success heavily relies on the availability of large-scale labeled datasets, which can be costly and time-consuming to obtain. Conversely, unsupervised learning techniques can leverage abundant unlabeled text data to learn rich representations, but they do not directly optimize for specific NLP tasks. This paper presents a novel hybrid approach that synergizes unsupervised and supervised learning to improve the accuracy of NLP task modeling. While supervised models excel at specific tasks, they rely on large labeled datasets. Unsupervised techniques can learn rich representations from abundant unlabeled text but don't directly optimize for tasks. Our methodology integrates an unsupervised module that learns representations from unlabeled corpora (e.g., language models, word embeddings) and a supervised module that leverages these representations to enhance task-specific models [4]. We evaluate our approach on text classification and named entity recognition (NER), demonstrating consistent performance gains over supervised baselines. For text classification, contextual word embeddings from a language model pretrain a recurrent or transformer-based classifier. For NER, word embeddings initialize a BiLSTM sequence labeler. By synergizing techniques, our hybrid approach achieves SOTA results on benchmark datasets, paving the way for more data-efficient and robust NLP systems.

Stand in surgeon's shoes: virtual reality cross-training to enhance teamwork in surgery.

Teamwork in surgery depends on a shared mental model of success, i.e., a common understanding of objectives in the operating room. A shared model leads to increased engagement among team members and is associated with fewer complications and overall better outcomes for patients. However, clinical training typically focuses on role-specific skills, leaving individuals to acquire a shared model indirectly through on-the-job experience. We investigate whether virtual reality (VR) cross-training, i.elet@tokeneonedotexposure to other roles, can enhance a shared mental model for non-surgeons more directly. Our study focuses on X-ray guided pelvic trauma surgery, a procedure where successful communication depends on the shared model between the surgeon and a C-arm technologist. We present a VR environment supporting both roles and evaluate a cross-training curriculum in which non-surgeons swap roles with the surgeon. Exposure to the surgical task resulted in higher engagement with the C-arm technologist role in VR, as measured by the mental demand and effort expended by participants ( ). It also has a significant effect on non-surgeon's mental model of the overall task; novice participants' estimation of the mental demand and effort required for the surgeon's task increases after training, while their perception of overall performance decreases ( ), indicating a gap in understanding based solely on observation. This phenomenon was also present for a professional C-arm technologist. Until now, VR applications for clinical training have focused on virtualizing existing curricula. We demonstrate how novel approaches which are not possible outside of a virtual environment, such as role swapping, may enhance the shared mental model of surgical teams by contextualizing each individual's role within the overall task in a time- and cost-efficient manner. As workflows grow increasingly sophisticated, we see VR curricula as being able to directly foster a shared model for success, ultimately benefiting patient outcomes through more effective teamwork in surgery.

Optimal Scheme Models with Two Kinds of Errors

In this paper, we propose stochastic scheme models with two kinds of errors for a computer system. A highly reliable computer system needs greatly to be backed up successfully in case of computer system crashes due to hardware failures and malicious software infections. We aim to determine appropriate backup times in computer processing of constant and random tasks. For this purpose, the trade-off between backup times and data losses due to errors is considered. The total execution times until the computing processing completes are obtained, and optimal backup policies to minimize them are derived analytically. Furthermore, we take up extended models of random tasks with [Formula: see text] sub-tasks and analyze them.

HAMMF: Hierarchical attention-based multi-task and multi-modal fusion model for computer-aided diagnosis of Alzheimer’s disease

Alzheimer’s disease (AD) is a progressive neurodegenerative condition, and early intervention can help slow its progression. However, integrating multi-dimensional information and deep convolutional networks increases the model parameters, affecting diagnosis accuracy and efficiency and hindering clinical diagnostic model deployment. Multi-modal neuroimaging can offer more precise diagnostic results, while multi-task modeling of classification and regression tasks can enhance the performance and stability of AD diagnosis. This study proposes a Hierarchical Attention-based Multi-task Multi-modal Fusion model (HAMMF) that leverages multi-modal neuroimaging data to concurrently learn AD classification tasks, cognitive score regression, and age regression tasks using attention-based techniques. Firstly, we preprocess MRI and PET image data to obtain two modal data, each containing distinct information. Next, we incorporate a novel Contextual Hierarchical Attention Module (CHAM) to aggregate multi-modal features. This module employs channel and spatial attention to extract fine-grained pathological features from unimodal image data across various dimensions. Using these attention mechanisms, the Transformer can effectively capture correlated features of multi-modal inputs. Lastly, we adopt multi-task learning in our model to investigate the influence of different variables on diagnosis, with a primary classification task and a secondary regression task for optimal multi-task prediction performance. Our experiments utilized MRI and PET images from 720 subjects in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. The results show that our proposed model achieves an overall accuracy of 93.15% for AD/NC recognition, and the visualization results demonstrate its strong pathological feature recognition performance.

Large environmental changes reduce valence-dependent belief updating

When updating beliefs, humans tend to integrate more desirable information than undesirable information. In stable environments (low uncertainty and high predictability), this asymmetry favors motivation towards action and perceived self-efficacy. However, in changing environments (high uncertainty and low predictability), this process can lead to risk underestimation and increase unwanted costs. Here, we examine how people (n = 388) integrate threatening information during an abrupt environmental change (mandatory quarantine during the COVID-19 pandemic). Given that anxiety levels are associated with the magnitude of the updating belief asymmetry; we explore its relationship during this particular context. We report a significant reduction in asymmetrical belief updating during a large environmental change as individuals integrated desirable and undesirable information to the same extent. Moreover, this result was supported by computational modeling of the belief update task. However, we found that the reduction in asymmetrical belief updating was not homogeneous among people with different levels of Trait-anxiety. Individuals with higher levels of Trait-anxiety maintained a valence-dependent updating, as it occurs in stable environments. On the other hand, updating behavior was not associated with acute anxiety (State-Anxiety), health concerns (Health-Anxiety), or having positive expectations (Trait-Optimism). These results suggest that highly uncertain environments can generate adaptive changes in information integration. At the same time, it reveals the vulnerabilities of individuals with higher levels of anxiety to adapt the way they learn.

Temporal validity reassessment: commonsense reasoning about information obsoleteness

It is useful for machines to know whether text information remains valid or not for various applications including text comprehension, story understanding, temporal information retrieval, and user state tracking on microblogs as well as via chatbot conversations. This kind of inference is still difficult for current models, including also large language models, as it requires temporal commonsense knowledge and reasoning. We approach in this paper the task of Temporal Validity Reassessment, inspired by traditional natural language reasoning to determine the updates of the temporal validity of text content. The task requires judgment whether actions expressed in a sentence are still ongoing or rather completed, hence, whether the sentence still remains valid or has become obsolete, given the presence of context in the form of a supplementary content such as a follow-up sentence. We first construct our own dataset for this task and train several machine learning models. Then we propose an effective method for learning information from an external knowledge base that gives information regarding temporal commonsense knowledge. Using our prepared dataset, we introduce a machine learning model that incorporates the information from the knowledge base and demonstrate that incorporating external knowledge generally improves the results. We also experiment with different embedding types to represent temporal commonsense knowledge as well as with data augmentation methods to increase the size of our dataset.

Design and Implementation of Simulation Training System for Ground Maintenance Pannel of a Certain Type of Aircraft

According to the design requirements of ground maintenance pannel simulation training system of a certain type of aircraft, a set of training system with the functions of ground handling practice, aircraft maintenance operation training and PMA interactive operation training is developed. The system design fully considers the requirements of generalization and modularization design. The function division of each module is clear, the signal interface is clear, and one of the modules can be replaced quickly to facilitate the positioning and elimination of failure. It has the dual expansibility of training task and training model, and can continuously add training task package, including procedure training, troubleshooting guidance training, etc., and can adapt to change in aircraft state.

Peer-mediated Social Signals Alter Risk Tolerance in Teenage Boys Depending on their Peers.

During early adolescence, peer influences play a crucial role in shaping learning and decision preferences. When teens observe what their peers are doing, they can learn and change their behavior, especially when they are taking risks. Our study incorporated an economical behavioral task and computational modeling framework to examine whether and how early male adolescents' risk attitudes change when they see information about their peers' choices. We recruited 38 middle school male students aged 12-15 years. The experiment consisted of three sessions: The first and third sessions were designed to evaluate the risk attitude of the participants. In the second session, participants were asked to guess the choices made by their peers, and then, the computer gave them feedback on the correctness of their predictions. Each participant was randomly assigned to risk-taking or risk-averse peers. Our results revealed that teenagers who predicted risk-averse peers exhibited significant declines in their risk attitudes during the last session. On the other hand, participants with risk-seeking peers exhibited a significantly higher level of risk attitudes after predicting their peers. The data showed that these peer-biased changes in risk attitudes are proportional to the gap between teens and their peers' risk perspectives. Results showed that their perspectives aligned closer after receiving the information, and approximately a third of the gap was eliminated. Here, by combining choice data and computational modeling, we demonstrate that risky behavior is contagious among male adolescents. According to our data, peer-biased risk contagion, a socially motivated and deliberate process, is associated with social distance in teens. There's no causal directionality here, but we could speculate that peer influence goes hand-in-hand with social integration as an adaptive process.

Generative artificial intelligence for distributed learning to enhance smart grid communication

Machine learning models are the backbone of smart grid optimization, but their effectiveness hinges on access to vast amounts of training data. However, smart grids face critical communication bottlenecks due to the ever-increasing volume of data from distributed sensors. This paper introduces a novel approach leveraging Generative Artificial Intelligence (GenAI), specifically a type of pre-trained Foundation Model (FM) architecture suitable for time series data due to its efficiency and privacy-preserving properties. These GenAI models are distributed to agents, or data holders, empowering them to fine-tune the foundation model with their local datasets. By fine-tuning the foundation model, the updated model can produce synthetic data that mirrors real-world grid conditions. The server aggregates fine-tuned model from all agents and then generates synthetic data which considers all data collected in the grid. This synthetic data can be used to train global machine learning models for specific tasks like anomaly detection and energy optimization. Then, the trained task models are distributed to agents in the grid to leverage them. The paper highlights the advantages of GenAI for smart grid communication, including reduced communication burden, enhanced privacy through anonymized data transmission, and improved efficiency and scalability. By enabling a distributed and intelligent communication architecture, GenAI introduces a novel way for a more secure, efficient, and sustainable energy future.

A Critical Review of Human Jaw Biomechanical Modeling

The human jaw is a complex biomechanical system involving different anatomical components and an articulated muscular system devoted to its dynamical activation. The numerous actions exerted by the mandible, such as talking, eating or chewing, make its biomechanical comprehension absolutely indispensable. To date, even if research on this topic has achieved interesting outcomes using in vitro testing, thanks to the development of new apparatus and methods capable of performing more and more realistic experiments, theoretical modeling is still worthy of investigation. In light of this, nowadays, the Finite Element Method (FEM) approach constitutes certainly the most common tool adopted to investigate particular issues concerning stress–strain characterization of the human jaw. In addition, kinematics analyses, both direct and inverse, are also diffuse and reported in the literature. This manuscript aimed to propose a critical review of the most recurrent biomechanical models of the human mandible to give readers a comprehensive overview on the topic. In light of this, the numerical approaches, providing interesting outcomes, such as muscular activation profiles, condylar forces and stress–strain fields for the human oral cavity, are mainly differentiated between according to the joint degrees of freedom, the analytical descriptions of the muscular forces, the boundary conditions imposed, the kind of task and mandible anatomical structure modeling.

ДОСЛІДЖЕННЯ ІНТЕЛЕКТУАЛЬНИХ МОДЕЛЕЙ КЛАСИФІКАЦІЇ НЕВИЗНАЧЕНИХ ДАНИХ З ВИМОГАМИ ДОСТОВІРНОСТІ РЕЗУЛЬТАТІВ

The report contains the results of research and development of intellectual models for managing complex systems under conditions of data uncertainty based on classification procedures, which ensure reliable solution of tasks considering the assessment of the boundary dimensionality of models. The possibilities of improving Hamming neural networks for data classification in formats of fuzzy variables and certainty factor CF(A) are investigated. The features of the mathematical model of classification tasks based on a set of feature templates are determined. A software complex of information technology for assignment/selection of performers, as well as determination of authorship of Ukrainian-language works based on classification of sets of templates with certain fuzzy features, is presented. The software complex utilizes the reduction and Cohen's kappa statistics procedures proposed in the report.

Learning from doctoral supervisors’ and candidates’ reflections on a supervisory model

ABSTRACT Professional development for doctoral supervisors differs between higher education institutions (HEIs) across the globe from non-existent support to one off workshops, to mandatory programmes. Communities of practice programmes encourage supervisors to reflect on case studies and conceptual models but there is limited research which explores the learning from supervisors’ and candidates’ reflections. Using interviews with supervisors and candidates from one HEI in England, this paper explores their reflections on a model of supervisor roles and tasks. The model aided the reflections of supervisors and candidates in terms of their own preferences, but also how they responded to candidate needs or when working in supervision teams. There is complexity and fluidity in terms of the roles offered in a single meeting and a suggestion that more pastoral roles need to be added. The model could be used for discussions between supervisors and candidates or within professional development programmes.

Task allocation algorithm for distributed large data stream group computing in the era of digital intelligence

In the era of digital intelligence, this paper studies the task allocation algorithm of distributed large data stream group computing, and reasonably allocates the task of group computing to meet the needs of massive computing and analysis of distributed large data stream. According to the idea of swarm intelligence perception and crowdsourcing platform, the task allocation model of distributed large data stream group computing is constructed to realize the task allocation of group computing. A distributed large data stream group computing task model and a user model are constructed, user attributes are initialized by using the accuracy of the answers submitted by users, the possibility that users can participate in the group computing task is predicted by a logistic regression algorithm, so that user candidate sequences participating in the computing task can be obtained, and the accuracy of the user’s real topics and corresponding topics can be grasped by capturing the candidate users’ real topics and evaluating the accuracy algorithm. Select the users who meet the subject area, update the candidate user sequence, and filter the users again on the basis of fully considering the factors such as information gain, user integrity and cost, so as to get the final user sequence and complete the task allocation of group computing. Experiments show that this method can solve the problem of distributed large data flow group computing task allocation, achieve high accuracy, reduce the cost, and effectively improve the information gain.

TMHD: Twin-Bridge Scheduling of Multi-Heterogeneous Dependent Tasks for Edge Computing

As an efficient computing paradigm, Mobile Edge Computing (MEC) is essential in assisting mobile devices with real-time complex tasks such as big data analytics. In MEC, application tasks consist of multiple dependent subtasks, and the way to process tasks ensuring lower response latency through efficient scheduling orders is of relevant importance. Most existing research on scheduling sorts the dependent tasks out by simple topological ordering, which leads to idle and wasteful resources. For such, we propose a Twin-Bridge Scheduling of Multi-Heterogeneous Dependent Tasks for Edge Computing (TMHD), with two state embedding layers for multi-dependent tasks that reconstruct the model of dependent tasks to find the optimal scheduling paths and concurrently make the best offloading decisions. Considering the limitations of local edge-end model training, we upload the trained models to the cloud periodically for local model aggregation update, speeding up the model convergence whilst guaranteeing the training accuracy and demonstrating the convergence of multi-model aggregation. Compared with existing dependent task processing methods, TMHD reduces the average task processing time on four different datasets by 18.68 ms, 71.79 ms, 70.99 ms, and 64.32 ms, which are shown to be effective in reducing the response time of the device task.