Parallel Training Research Articles

Camera-Aware Recurrent Learning and Earth Mover’s Test-Time Adaption for Generalizable Person Re-Identification

Domain generalization in person re-identification (ReID) aims to design a generalizable model, which is trained under the supervision of a set of labeled source domains and can be directly deployed on unknown domains. Existing approaches simply treat each identity as a distinct class and ignore the differences among cameras. We argue that the camera information is crucial for learning discriminative representations, as people’s behavior usually varies between cameras. In this paper, we present Multi-Centroid Memory (MCM) to capture different camera information for each identity and Soft Triple Hard (ST-Hard) loss to align the information of the same identity across cameras. Furthermore, in contrast to the traditional approaches of training a single model using a parallel training mechanism, we propose the Recurrent Implicit Lifelong Learning (RILL) that feeds the source domains into the model in a continuous loop to train an expert for each domain. To make each expert further generalized to other source domains, during the training on the current domain, RILL adopts a style replay-based method to simulate the training of the previous domain, encouraging each domain’s expert to extract generalizable features. We also present Earth Mover’s Test-time Adaption (EMTA) to be used in conjunction with RILL, which enables source domains that are more similar to the test domain to play a more significant role in the test. This is achieved by our proposed Earth Mover’s Similarity (EMS), which helps model the similarities between the source and test domains. Extensive experiments on two evaluation protocols fully demonstrate our framework’s generalization and competitiveness.

Adaptively Placed Multi-Grid Scene Representation Networks for Large-Scale Data Visualization.

Scene representation networks (SRNs) have been recently proposed for compression and visualization of scientific data. However, state-of-the-art SRNs do not adapt the allocation of available network parameters to the complex features found in scientific data, leading to a loss in reconstruction quality. We address this shortcoming with an adaptively placed multi-grid SRN (APMGSRN) and propose a domain decomposition training and inference technique for accelerated parallel training on multi-GPU systems. We also release an open-source neural volume rendering application that allows plug-and-play rendering with any PyTorch-based SRN. Our proposed APMGSRN architecture uses multiple spatially adaptive feature grids that learn where to be placed within the domain to dynamically allocate more neural network resources where error is high in the volume, improving state-of-the-art reconstruction accuracy of SRNs for scientific data without requiring expensive octree refining, pruning, and traversal like previous adaptive models. In our domain decomposition approach for representing large-scale data, we train an set of APMGSRNs in parallel on separate bricks of the volume to reduce training time while avoiding overhead necessary for an out-of-core solution for volumes too large to fit in GPU memory. After training, the lightweight SRNs are used for realtime neural volume rendering in our open-source renderer, where arbitrary view angles and transfer functions can be explored. A copy of this paper, all code, all models used in our experiments, and all supplemental materials and videos are available at https://github.com/skywolf829/APMGSRN.

Demonstration-guided deep reinforcement learning for coordinated ramp metering and perimeter control in large scale networks

Effective traffic control methods have great potential in alleviating network congestion. Particularly, in an urban network consisting of heterogeneous roads (e.g., freeways and urban roads), how to integrate and coordinate control policies on different roads is a critical issue in large-scale networks. This study addresses this question from two aspects: modeling and control. From the modeling aspect, we formulate the hybrid traffic modeling in heterogeneous networks with the Asymmetric Cell Transmission Model (ACTM) for freeways and the generalized bathtub model for urban roads. For the control aspect, this study considers two representative control approaches: ramp metering for freeways and perimeter control for urban roads, and we aim to develop a deep reinforcement learning (DRL)-based coordinated control framework for large-scale networks. However, there are two significant challenges in the coordinated control in large-scale networks with DRL methods: non-stationary environment and large search space. To address both issues, we incorporate the demonstration to guide the DRL method for better convergence by introducing the concept of “teacher” and “student” models. The teacher models are traditional controllers that provide control demonstrations. For instance, ALINEA and Gating are two representative feedback controllers for ramp metering and perimeter control which can be “teacher” models. The student models are DRL methods, which learn from teachers and aim to surpass the teachers’ performance. Additionally, we develop a parallel training scheme to accelerate the proposed DRL method. To validate the proposed framework, we conduct two case studies in a small-scale network and a real-world large-scale traffic network in Hong Kong. Numerical results show that the proposed DRL method outperforms demonstrators as well as DRL methods, and the coordinated control is more effective than just controlling ramps or perimeters respectively. The research outcome reveals the great potential of combining traditional controllers with DRL for coordinated control in large-scale networks.

Amygdala-related electrical fingerprint is modulated with neurofeedback training and correlates with deep-brain activation: proof-of-concept in borderline personality disorder.

The modulation of brain circuits of emotion is a promising pathway to treat borderline personality disorder (BPD). Precise and scalable approaches have yet to be established. Two studies investigating the amygdala-related electrical fingerprint (Amyg-EFP) in BPD are presented: one study addressing the deep-brain correlates of Amyg-EFP, and a second study investigating neurofeedback (NF) as a means to improve brain self-regulation. Study 1 combined electroencephalography (EEG) and simultaneous functional magnetic resonance imaging to investigate the replicability of Amyg-EFP-related brain activation found in the reference dataset (N = 24 healthy subjects, 8 female; re-analysis of published data) in the replication dataset (N = 16 female individuals with BPD). In the replication dataset, we additionally explored how the Amyg-EFP would map to neural circuits defined by the research domain criteria. Study 2 investigated a 10-session Amyg-EFP NF training in parallel to a 12-weeks residential dialectical behavior therapy (DBT) program. Fifteen patients with BPD completed the training, N = 15 matched patients served as DBT-only controls. Study 1 replicated previous findings and showed significant amygdala blood oxygenation level dependent activation in a whole-brain regression analysis with the Amyg-EFP. Neurocircuitry activation (negative affect, salience, and cognitive control) was correlated with the Amyg-EFP signal. Study 2 showed Amyg-EFP modulation with NF training, but patients received reversed feedback for technical reasons, which limited interpretation of results. Recorded via scalp EEG, the Amyg-EFP picks up brain activation of high relevance for emotion. Administering Amyg-EFP NF in addition to standardized BPD treatment was shown to be feasible. Clinical utility remains to be investigated.

PARTNERING TO IMPROVE CARE: STRENGTHENING THE FAMILY CAREGIVER AND HOME CARE WORKER RELATIONSHIP

Abstract Family caregivers and home care workers provide essential daily care for millions of older adults and individuals with disabilities across the United States—and together they constitute the backbone of home and community-based services (HCBS). In many cases, these unpaid and paid caregivers work closely together and share common knowledge, skills, experiences, and challenges. Yet the overlap and relationship between them remains relatively understudied. This session will describe the importance of strengthening the family caregiver/home care worker dyad and discuss promising developments in research, policy, and practice. We will then highlight a pilot training intervention designed to improve the shared knowledge and interpersonal competencies of family caregivers and home care workers. The intervention— Partnering to Improve Care: Direct Care Workers and Families—was developed by PHI (a national nonprofit research, public education, and consulting organization) in partnership with the Florida Pioneer Network, Home Instead, and Florida State University through the North and Central Florida Geriatrics Workforce Enhancement Partnership (GWEP). Informed by the PHI Coaching Approach®, the intervention consists of five parallel online training modules for family caregivers and for home care workers that utilize adult learner-centered teaching methods and reflect input from home care stakeholders. The modules portray common scenarios that family caregivers and home care workers face and aim to build communication and relationship skills that facilitate collaborative problem-solving and the provision of person-centered care. We will present key findings on the feasibility and acceptability of this intervention, discuss next steps in this research, and highlight broader policy and practice recommendations.

Dynamic Corrected Split Federated Learning With Homomorphic Encryption for U-Shaped Medical Image Networks.

U-shaped networks have become prevalent in various medical image tasks such as segmentation, and restoration. However, most existing U-shaped networks rely on centralized learning which raises privacy concerns. To address these issues, federated learning (FL) and split learning (SL) have been proposed. However, achieving a balance between the local computational cost, model privacy, and parallel training remains a challenge. In this articler, we propose a novel hybrid learning paradigm called Dynamic Corrected Split Federated Learning (DC-SFL) for U-shaped medical image networks. To preserve data privacy, including the input, model parameters, label and output simultaneously, we propose to split the network into three parts hosted by different parties. We propose a Dynamic Weight Correction Strategy (DWCS) to stabilize the training process and avoid the model drift problem due to data heterogeneity. To further enhance privacy protection and establish a trustworthy distributed learning paradigm, we propose to introduce additively homomorphic encryption into the aggregation process of client-side model, which helps prevent potential collusion between parties and provides a better privacy guarantee for our proposed method. The proposed DC-SFL is evaluated on various medical image tasks, and the experimental results demonstrate its effectiveness. In comparison with state-of-the-art distributed learning methods, our method achieves competitive performance.

Learning From Images: Proactive Caching With Parallel Convolutional Neural Networks

With the continuous trend of data explosion, delivering packets from data servers to end users causes increased stress on both the fronthaul and backhaul traffic of mobile networks. To mitigate this problem, caching popular content closer to the end-users has emerged as an effective method for reducing network congestion and improving user experience. To find the optimal locations for content caching, many conventional approaches construct various Mixed Integer Linear Programming (MILP) models. However, such methods may fail to support online decision making due to the inherent curse of dimensionality. In this paper, a novel framework for proactive caching is proposed. This framework merges model-based optimization with data-driven techniques by transforming an optimization problem into a grayscale image. For parallel training and simple design purposes, the proposed MILP model is first decomposed into a number of sub-problems and, then, Convolutional Neural Networks (CNNs) are trained to predict content caching locations of these sub-problems. Furthermore, since the MILP model decomposition neglects the network resources (such as caching space and link bandwidth) competition among sub-problems, the CNNs' outputs have the risk to be infeasible solutions. Therefore, two algorithms are provided: the first uses predictions from CNNs as an extra constraint to reduce the number of decision variables; the second employs CNNs' outputs to accelerate local search. Numerical results show that the proposed scheme can reduce 71.6% computation time, whose computation time reaches around 28.9 ms, with only 0.8% additional performance cost compared to the MILP solution, which provides high quality decision making in pseudo real-time.

Survey on Cloud Computing Integrated with Artificial Intelligence

The combination of cloud computing with Artificial Intelligence (AI) has become a viable approach in recent years to handle the increasing processing needs of AI applications. This article offers a thorough analysis of distributed and scalable artificial intelligence frameworks that use cloud computing to improve the effectiveness and performance of Deep Learning (DL). First, we provide a summary of common cloud services and AI frameworks, emphasizing the advantages and disadvantages of each. The important topics of data management and storage in cloud-based AI systems are then covered in detail, including feature engineering, data pretreatment, privacy, and security. Next, we investigate the distributed and parallel training methods for AI models, with particular emphasis on cloud-based training topologies, communication mechanisms, and model splitting. On the next chapters, we examine auto-scaling, load balancing, resource allocation, and performance benchmarking as optimization methodologies for AI workloads on the cloud. We also cover serverless deployment alternatives, containerization, best practices for monitoring, and the deployment and servicing of AI models on the cloud. We give a comprehensive study of costs, optimization methodologies, and case studies exhibiting successful installations to guarantee the cost-effectiveness of cloud-based AI systems. In conclusion, we provide an overview of the study's main conclusions, go through the difficulties and restrictions associated with cloud-based artificial intelligence, and point out new directions for future research in this area. For academics and practitioners trying to use cloud computing to create scalable, effective, and affordable artificial intelligence systems, this article will be a great resource.

NeutronStream: A Dynamic GNN Training Framework with Sliding Window for Graph Streams

Existing Graph Neural Network (GNN) training frameworks have been designed to help developers easily create performant GNN implementations. However, most existing GNN frameworks assume that the input graphs are static, but ignore that most real-world graphs are constantly evolving. Though many dynamic GNN models have emerged to learn from evolving graphs, the training process of these dynamic GNNs is dramatically different from traditional GNNs in that it captures both the spatial and temporal dependencies of graph updates. This poses new challenges for designing dynamic GNN training frameworks. First, the traditional batched training method fails to capture real-time structural evolution information. Second, the time-dependent nature makes parallel training hard to design. Third, it lacks system supports for users to efficiently implement dynamic GNNs. In this paper, we present NeutronStream, a framework for training dynamic GNN models. NeutronStream abstracts the input dynamic graph into a chronologically updated stream of events and processes the stream with an optimized sliding window to incrementally capture the spatial-temporal dependencies of events. Furthermore, NeutronStream provides a parallel execution engine to tackle the sequential event processing challenge to achieve high performance. NeutronStream also integrates a built-in graph storage structure that supports dynamic updates and provides a set of easy-to-use APIs that allow users to express their dynamic GNNs. Our experimental results demonstrate that, compared to state-of-the-art dynamic GNN implementations, NeutronStream achieves speedups ranging from 1.48X to 5.87X and an average accuracy improvement of 3.97%.

Semi-supervised learning and bidirectional decoding for effective grammar correction in low-resource scenarios

The correction of grammatical errors in natural language processing is a crucial task as it aims to enhance the accuracy and intelligibility of written language. However, developing a grammatical error correction (GEC) framework for low-resource languages presents significant challenges due to the lack of available training data. This article proposes a novel GEC framework for low-resource languages, using Arabic as a case study. To generate more training data, we propose a semi-supervised confusion method called the equal distribution of synthetic errors (EDSE), which generates a wide range of parallel training data. Additionally, this article addresses two limitations of the classical seq2seq GEC model, which are unbalanced outputs due to the unidirectional decoder and exposure bias during inference. To overcome these limitations, we apply a knowledge distillation technique from neural machine translation. This method utilizes two decoders, a forward decoder right-to-left and a backward decoder left-to-right, and measures their agreement using Kullback-Leibler divergence as a regularization term. The experimental results on two benchmarks demonstrate that our proposed framework outperforms the Transformer baseline and two widely used bidirectional decoding techniques, namely asynchronous and synchronous bidirectional decoding. Furthermore, the proposed framework reported the highest F1 score, and generating synthetic data using the equal distribution technique for syntactic errors resulted in a significant improvement in performance. These findings demonstrate the effectiveness of the proposed framework for improving grammatical error correction for low-resource languages, particularly for the Arabic language.

A novel Congestion Control algorithm based on inverse reinforcement learning with parallel training

With the growing impact of the Internet, computer network communication has become an essential component for various industries. Congestion Control (CC) algorithms serve as the backbone of network communication, significantly affecting network quality. However, designing a CC algorithm that performs optimally across diverse network environments presents substantial challenges. The task of redesigning and optimizing CC algorithms for specific network environments demands both expert experience and a substantial workforce. In this paper, we proposed an inverse reinforcement learning (IRL) algorithm that can use expert data to guide the self-optimization of the CC model in specific network environments. To enhance model training efficiency, we propose a parallel training framework and a visualization analysis tool, enabling distributed training and real-time control level analysis. In the experiments, we assess the performance of 16 algorithms across 3 network scenarios using Pantheon. Our IRL model achieves the optimal level of network performance in satellite network, enhancing throughput by 10%–23%. For delay performance, it ranks 2nd in wired network and achieves a 21%–67% improvement over traditional TCP algorithms in regular network.

Morpheme-Based Neural Machine Translation Models for Low-Resource Fusion Languages

Neural approaches, which are currently state-of-the-art in many areas, have contributed significantly to the exciting advancements in machine translation. However, Neural Machine Translation (NMT) requires a substantial quantity and good quality parallel training data to train the best model. A large amount of training data, in turn, increases the underlying vocabulary exponentially. Therefore, several proposed methods have been devised for relatively limited vocabulary due to constraints of computing resources such as system memory. Encoding words as sequences of subword units for so-called open-vocabulary translation is an effective strategy for solving this problem. However, the conventional methods for splitting words into subwords focus on statistics-based approaches that mainly conform to agglutinative languages. In these languages, the morphemes have relatively clean boundaries. These methods still need to be thoroughly investigated for their applicability to fusion languages, which is the main focus of this article. Phonological and orthographic processes alter the borders of constituent morphemes of a word in fusion languages. Therefore, it makes it difficult to distinguish the actual morphemes that carry syntactic or semantic information from the word’s surface form, the form of the word as it appears in the text. We, thus, resorted to a word segmentation method that segments words by restoring the altered morphemes. We also compared conventional and morpheme-based NMT subword models. We could prove that morpheme-based models outperform conventional subword models on a benchmark dataset.

Metacognitive Strategy Training Improves Decision-Making Abilities in Amnestic Mild Cognitive Impairment.

Mild cognitive impairment (MCI) is associated with deficits in decision-making, which is of utmost importance for daily functioning. Despite evidence of declined decision-making abilities, research on decision-making interventions for MCI is scarce. As metacognition seems to play an important role in decision-making, the present study's aim was to examine whether a metacognitive strategy training can improve MCI patients' decision-making abilities. Older adults-patients of a day care center, diagnosed with amnestic MCI (n = 55) were randomly allocated in two groups, which were matched in gender, age and educational level. Τhe experimental group (n = 27, 18 women, mean age = 70.63, mean years of education = 13.44) received the metacognitive strategy training in parallel with the cognitive and physical training programs of the day care center, and the active control group (n = 28, 21 women, mean age = 70.86, mean years of education = 13.71) received only the cognitive and physical training of the center. The metacognitive strategy training included three online meeting sessions that took place once per week. The basis of the intervention was using analytical thinking, by answering four metacognitive-strategic questions, to make decisions about everyday situations. To examine the efficacy of the training, the ability to make decisions about everyday decision-making situations and the ability to apply decision rules were measured. Both groups participated in a pre-test session and a post-test session, while the experimental group also participated in a follow-up session, one month after the post-test session. The results showed that the experimental group improved its ability to decide, based on analytical thinking, about economic and healthcare-related everyday decision-making situations after they received the metacognitive strategy training. This improvement was maintained one month later. However, the ability to apply decision rules, which requires high cognitive effort, did not improve. In conclusion, it is important that some aspects of the analytical decision-making ability of amnestic MCI patients were improved due to the present metacognitive intervention.

Activities of Fedor Sannikov as a Supervisor of Schools in Ekaterinburg (1747—1751)

This article examines, for the first time, the activities of Fedor Sannikov as the superintendent of Yekaterinburg schools, based on archival documents. It is noted that he had to fulfill his duties in a difficult environment, as tensions arose between him and the administration. As a result, his proposals were likely intentionally delayed or rejected. Despite the unfavorable attitude and even resistance from his superiors, Sannikov made every effort to modernize the schools. In particular, he insisted on reintroducing parallel vocational training for students, as introduced by V. N. Tatishchev according to a document from 1736. Sannikov repeatedly made proposals to increase the number of teachers in the arithmetic school, where he taught over 100 students. It has been established that under the supervision of the superintendent, there were more than 270-280 students in Yekaterinburg schools by the mid-18th century, and Sannikov was aware of the financial situation, abilities, and academic achievements of each of them. In the conditions of rising costs in the late 1740s, he proposed training students in crafts and allowing them to work in workshops in the second half of the day, as the students’ government stipend was not enough for clothing, shoes, food, and many had to rely on charity. Sannikov also paid attention to the education of noble children whose parents were not inclined to send them to schools.

Curriculum Based Reinforcement Learning of Grid Topology Controllers to Prevent Thermal Cascading

This paper describes how domain knowledge of power system operators can be integrated into reinforcement learning (RL) frameworks to effectively learn agents that control the grid's topology to prevent thermal cascading. Typical RL-based topology controllers fail to perform well due to the large search/optimization space. Here, we propose an actor-critic-based agent to address the problem's combinatorial nature and train the agent using the RL environment developed by RTE, the French TSO. To address the challenge of the large optimization space, a curriculum-based approach with reward tuning is incorporated into the training procedure by modifying the environment using network physics for enhanced agent learning. Further, a parallel training approach on multiple scenarios is employed to avoid biasing the agent to a few scenarios and make it robust to the natural variability in grid operations. Without these modifications to the training procedure, the RL agent failed for most test scenarios, illustrating the importance of properly integrating domain knowledge of physical systems for real-world RL learning. The agent was tested by RTE for the 2019 learning to run the power network challenge and was awarded the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$2^{nd}$</tex-math></inline-formula> place in accuracy and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$1^{st}$</tex-math></inline-formula> place in speed. The developed code is open-sourced for public use. Analysis of a simple system proves the enhancement in training RL-agents using the curriculum.

Research on Choice Preference of Parallel Trains for High-Speed Rail Heterogeneous Passengers

The passenger capacity utilization of parallel trains on China’s high-speed railway (HSR) varies greatly, which seriously affects the revenue and passenger service quality of railroad companies. One of the key factors contributing to this issue is the uniformity of fares. To address this, China’s railway enterprises are implementing differential pricing reforms for parallel trains. This paper presents a two-step method that combines latent class analysis (LCA) and random parameter logit (RPL) models to estimate the preferences of diverse passengers for parallel trains and quantify them using willingness to pay. To illustrate this, we conducted revealed preference and stated preference questionnaire surveys at Nanning East Railway Station, China, focusing on the Nanning-Guangzhou HSR route. These surveys gathered data on passengers’ personal and travel attributes. LCA was carried out according to passenger attributes, and passengers were divided into three categories according to the goodness-of-fit measures. Through the questionnaire data, the travel time and departure time period of each category of the passenger were converted into costs, and the absolute value of each influencing factor of passengers for parallel trains was obtained. Considering the limited rational choice behavior of passengers, the perceived value of each influencing factor of passengers on parallel trains was calculated through prospect theory. Finally, an RPL model for parallel train choice of heterogeneous passengers was constructed, and the parameters of each category of the passenger were estimated based on the absolute value and perceived value of the influencing factors.

1170 An Experience of Arranging the UK’s First Free Robotic Surgery Workshop for Medical Students and Doctors Early in Their Careers

Abstract Aim Training opportunities for Robotic Surgery training are very competitive and expensive in the era of parallel training with consultants also in a learning phase. We share our experience of providing a free Robotic Surgery workshop to medical students and junior doctors under the umbrella of Association of Surgeons in training (ASiT). Method The course was organised as a collaboration between ASiT and Intuitive Surgical at Leicester General Hospital Educational centre. Dedicated opportunities were designed for device training, theory, and DaVinci XI console and simulator stations. Training was provided by a Robotic Surgery Fellow and Technical team from Intuitive on voluntary basis. No fees were charged for the course. Results 48 Delegates applied for registration. 23 were shortlisted into four groups. 13 medical students and 10 doctors were in attendance. Each Delegate was provided a dedicated time of at least 20 minutes on each station. Delegates had an opportunity to use the Console and Simulator for training exercises. Feedback from the course was resoundingly positive. To our knowledge this was the 1st free Robotic Surgery course in UK providing hands on experience to medical students and junior doctors. Conclusions Our experience of organizing a “Free Robotic Surgery Workshop” may be utilised nationally to increase training opportunities for medical students and junior doctors.

538 Audit of Robotic Activities at a Tertiary Centre, an Attempt to Identify training Opportunities

Abstract Aim Training in Robotic surgery is still in the development phase. The training opportunities are competitive with consultants and trainees in the learning phase at many centres. We aimed to analyse the use of robotic activity at a tertiary centre across different specialities to find out parallel training opportunities. Method Data was collected retrospectively from a prospectively maintained Electronic Theatre record over an eight-month period. We recorded number of days surgical robots are in use per month, number of operations per day and operating times for three subspecialties. Data was analysed using SPSS version 26 Results The results show the robot was used on average 14.6 days per month. which equated to an average of 20.25 operations per month with an average operating time of 214.43 minutes per operation. The robot was used by Urology for on average 10 days per month, Gynaecology 3.38, and General surgery 1.5 days. The use depends on days allocated to specialities and ITU bed availability. Urology performed on average 13.75 operations per month compared to 5 for gynaecology and 1.5 for general surgery. Average case finishing time for single operation was 15:29, for two operations in a day was 17:36. Conclusions Given the robot was used less than half a month on average for operations, we can increase training opportunities by utilising them for training activities on non-operating days. More studies are needed at national level to help in creating a hub and spoke model by using Robots in non-operating times.

Integrated advance assessment of power system transient voltage and transient angle stability based on two-stage ensemble spatio-temporal graph neural network

Transient angle stability (TAS) and transient voltage stability (TVS) are important bases for safe operation of power system. With the replacement of a high percentage of new energy, the voltage and power angle problems are closely coupled, so an integrated assessment method that takes into account the rapidity, accuracy and stability is urgently needed to deal with the system risks. In order to solve this problem, firstly, we propose a multi-task spatio-temporal graph neural network model based on two-stage ensemble learning (TSEL-MT-STGNN) for integrated assessment of TAS&TVS. In the first stage of this model, the topological spatial features of the system are extracted through multi-type graph neural networks, and the obtained knowledge is input to the gated recurrent unit (GRU) for further temporal feature extraction, so as to achieve spatio-temporal information fusion. In this stage, TAS and TVS were combined through the multi-task framework to achieve integrated assessment under the shared features. In the second stage, meta-learner multi-layer perceptron (MLP) is used to fuse the output results of multi-type spatio-temporal graph neural network. Secondly, an integrated assessment method is proposed based on the model, which enables end-to-end assessment, and gives the stability result and safety margin of the system in the current state without knowing the fault clearing time. Thirdly, in order to ensure the timeliness of the assessment method, a multi-graph parallel training scheme is proposed to accelerate the training process. By introducing time constrained loss function and dynamic weight training, the response time is greatly reduced on the basis of maintaining the existing accuracy, thus achieving advance assessment. Finally, the improved New England 39-bus system is taken as an example for verification and analysis. The results show that the proposed method can achieve high-precision advance assessment of TAS&TVS, and has strong robustness and anti-noise capability.© 2017 Elsevier Inc. All rights reserved.

INCREASING THE EFFICIENCY OF CARGO AND PASSENGER TRANSPORTATION ON LINES WITH PARALLEL RUNS

Purpose. The purpose of the work is to determine measures to improve logistics operations with the specialization of directions for cargo and passenger transportation by transferring part of the transit cargo flow to parallel routes. Solving the problem of choosing a rational distribution of freight and passenger transportation depending on the specialization of directions will allow to optimize the use of throughput and transportation capacity and investments in infrastructure development. Method. Optimization of the transport process during mixed transportation involves the possibility of separating cargo traffic from passenger traffic. The proposed method involves solving the problem of choosing a rational distribution of cargo and passenger transportation depending on the specialization of directions. The idea of such demarcation is based on the fact that on the railway network it is possible to distinguish stations between which there are several parallel train routes. The research was carried out on the railway network connecting Lviv and Kyiv. When solving the problem, various indicators were considered: the length of the route, the time of movement, the mechanical operation of the locomotive, the load on the track, and others. In addition, the boundary conditions were taken into account: the volume of passenger traffic, cargo traffic, the carrying capacity of sections and the permissible speed of train movement. Results. A mathematical model of the redistribution of train flows on the railway network has been developed, which ensures the rational operation of the railway track with minimal costs for the reconstruction of the railway. Scientific novelty. It has been established that in the case of traffic separation, the timing of repairs may be different due to lower traffic intensity, which allows you to focus on specific types of repair work, and not on all types of work at the same time. Although the general principles of the periodicity of repairs may remain the same in both cases, periodicity schemes may differ depending on many factors, in particular, the intensity of operation, the technical condition of the vehicle, the timing of repairs and the availability of specialized personnel. Practical significance. With the application of the new approaches proposed by the authors to solving the tasks, schemes for the rational distribution of cargo and passenger transportation at the closed landfills of the regional branches of the South-Western and Lviv Railways were established, which will allow determining measures to improve logistics operations and reduce operating costs for the transportation of cargo and passengers.