Many industrial plant processes, involving the control of variables such as temperature, pressure, and flowrates, often experience time delays between control actions and the resulting responses of these variables, posingchallenges for maintaining stable and precise control operations. Control methods for dealing with such responsedelays have been the subject of research for many years, but their effective implementation has been hindered byvarious issues such as computational cost, safety considerations, and the gap between simulators and real plants.In this paper, an approach is presented to address the challenge of response delays, where the control modelis initially trained using the operating data to incorporate response delays, and then the control model learned inthis manner is employed to perform control operations. Transparency in training the control model is ensured bythe inherent methodology of the proposed approach, which enables learning solely from past operating data. As aresult, the need for repetitive trial-and-error exercises on physical plants to fine-tune the control model is eliminated,mitigating potential safety and efficiency concerns associated with real-world experimentation. To evaluate the effectivenessof the proposed approach, the steam temperature control system of a real waste-to-energy plant is chosenas the target process, where both learning and actual control operations are conducted. As a result of actual operation,the proposed approach demonstrates the equivalent or superior control performance compared with the currentcontrol system. Furthermore, it is highlighted that the proposed approach effectively handles the time delay, enablingproactive control actions.

In this paper, we propose a communication mechanism for multi-agent reinforcement learning that selectively sends the information needed to solve a task in a cooperative task. We address the question of “what” information should be sent to other agents from observations with information on multiple attributes, taking into account communication with other agents. The proposed method introduces a mechanism called Selective messenger, which combines the decomposition and organization of information by the deep generative model VAE and the selection of information by the Attention mechanism, as a mechanism for sending communication. Experiments show that the proposed method performs better than conventional communication methods in tasks where images with information on multiple attributes are observations and past communication needs to be referenced. In addition, the proposed method shows minimal performance variation compared to other methods when the message size for communication changes. This suggests that the our method sends important information more compactly. In the experiment to evaluate the effectiveness of the Selective messenger mechanism, we verified the efficacy of both the VAE and Attention components. The results show that the VAE component successfully performed disentanglement on the observations, and is capable of organizing and representing the information of each attribute contained within these observations. Furthermore, it show that the Attention component is able to appropriately select information while taking into account messages sent from other agents.

Entity Linking (EL) is a technology that links mentions (context-dependent token sequences that may refer to specific entities) in a text to corresponding entities in a knowledge base. It serves as a foundational technology in knowledge processing and natural language processing. Most research on EL focuses on English, and there is little research that targets Japanese. While research on multilingual EL models, which can be used for languages other than English, including Japanese, is advancing, EL that considers the characteristics of individual languages has not been fully achieved. Additionally, recent EL research utilizes word embeddings and knowledge graph embeddings in the construction of EL models, and building an EL model for a specific language requires language-specific adaptations. Pointer Network has the feature of selecting elements of the output sequence from the input sequence. By using Pointer Network, it may be possible to perform EL that is not included in the training data. In this study, we proposed a Japanese EL model based on the Pointer Network (Japanese PNEL model) forWikidata. We evaluated the accuracy of the Japanese PNEL model and analyzed effective features in Japanese EL and English EL.In addition, we have constructed a Japanese EL evaluation dataset by machine translating the English EL evaluation datasets WebQSP, SimpleQuestions, and LC-QuAD2 into Japanese.From the results of comparison experiments with existing multilingual models, we confirmed that the Japanese PNEL model outperformed existing multilingual models in terms of F1 score in evaluation experiments targeting WebQSPs that were machine-translated into Japanese.We analyzed more effective features through experiments on the impact of the difference in the number of dimensions and models for word embedding and knowledge graph embedding on EL accuracy. The results of the ablation study showed that knowledge graph embedding is also as effective in Japanese as in English.

The paper discusses the challenges of detecting deception in text communication, where non-verbal cues like tone of voice and facial expressions are absent, making it more difficult than in-person communication. It addresses the issue of scarcity in labeled corpora for lie detection research. The study uses data from a Werewolf BBS (bulletin board system), which, while extensive and useful, might be too specific to the Werewolf game context in terms of vocabulary and expressions. To address this, the research filters data from a general domain, X’s posting data, using only common vocabulary to enhance generalizability. Additionally, a new dataset for a murder mystery game was created to further validate the model trained on Werewolf BBS logs for cross-domain applicability. Experiments in lie detection were conducted using several different learning methods, with the BERT+GRU hierarchical model achieving an F1 score of 0.70 before filtering and 0.68 after, demonstrating the feasibility of lie detection in general domains.

This study investigates the diversity of image sets generated from textual descriptions by Text-to-image (T2I) models. Understanding this diversity is crucial for evaluating the relationship between textual inputs and the resultant images, which could significantly streamline the analysis of T2I models and minimize the human effort involved in generating desired images. Yet, this exploration raises two primary research questions: Firstly, the most appropriate metric for evaluating diversity, in terms of its correlation with human subjective assessments, remains unidentified. Secondly, the capability to predict the diversity of images produced by T2I models―encompassing both the feasibility and precision of such predictions―is yet to be established. To address these inquiries, our research was structured around two experiments. The first experiment aimed to identify the diversity evaluation metric that most closely aligns with human judgment by comparing the correlations between several metrics and human evaluations. The second experiment focused on developing a model capable of predicting the diversity of generated image sets, testing both its practicality and the accuracy of its predictions. Our findings reveal that the Vendi Score exhibits the highest correlation with human subjective evaluations, suggesting its effectiveness as a diversity metric. Additionally, while constructing a model to predict diversity proved feasible, the accuracy of these predictions was found to be significantly influenced by the specific objects described in the input texts. This highlights the nuanced relationship between textual descriptions and the diversity of generated images, suggesting avenues for further research to enhance the predictive model’s accuracy.

With the increasing trading volume in the Japanese CFD (Contract for Difference) market, the strategy of market making in CFDs has become a significant issue. Unlike other quote-driven markets where market makers face the dilemma of maximizing profits and managing inventory risk, CFD market makers can hedge their risks using the underlying market. Therefore, effectively incorporating hedging strategies into market making becomes a crucial aspect of CFD market strategies. This study aimed to analyze the optimal market-making strategies in the CFD market. A multi-agent simulation model of a CFD market was constructed, simulating the environment of a prominent Japanese CFD exchange, ”Click kabu 365”. This model included an index market as the underlying asset market, along with investor agents and market maker agents. Through the analysis of market makers in the CFD market, we proposed a new market-making strategy that dynamically adjusts hedging intensity based on market volatility and trends, in addition to using the spread strategy from previous research. To evaluate the effectiveness of the proposed strategy, we conducted simulations under different market conditions. The first set of experiments compared the performance of different strategies in separate market environments. The results showed that the proposed strategy outperformed others in terms of profitability and risk management. The second set of experiments considered competition among market makers within the same market environment, confirming that the proposed strategy maintained superior overall performance, though its advantage was less pronounced in this competitive scenario. The study concludes that the proposed market-making strategy effectively balances inventory risk management and hedging costs. Additionally, it highlights the potential risk of presenting competitive quotes in adverse market conditions when combining spread and hedging strategies, which should be considered in practical applications.

Echo chamber refers to a state in which individuals with similar preferences on social media construct closednetworks and consume similar information reflectively. This is one of the causes to prevalent fake news and conspiracyon social media. Previous research has shown that echo chamber is prevented in communication where there aremultiple topics. However, even though communication across multiple topics is common in daily social media use,echo chamber is still occurring on various topics. In this paper, we suggest the presence of agenda-setting as a potentialbackground factor for occurrence of echo chamber from communication on multiple topics. Agenda-setting is theconcept that topics frequently present to users are regarded as significant agendas. In this study, we experimented withmulti-agent simulations to investigate the communication among agents who assess the significance of specific topicsbased on the frequency of encountering them. Our result shows that if a specific topic is recognized as significant bymany users, users construct closed networks as echo chamber on the opinion of significant topic. We also conductedanother experiment to verify the effect of agenda setting by mass media coverage. In this experiment, a specific topiccame to be perceived as significant by many users, but echo chambers did not occur regarding significant topic.

Kerosene delivery is a service that regularly supplies kerosene to household tanks to prevent them from runningdry. Delivery companies create kerosene delivery plans to maintain tank levels and implement deliveries based onthese plans. We formulate the Kerosene Delivery Planning Problem (KRP) and propose a solution method that utilizesa problem partitioning approach through the rolling horizon approach, combined with a hybrid method named HAIRthat integrates tabu search and integer programming. From experiments on parameter search, it was confirmed that thevalue ofMaxIter for HAIR should be set as a priority. Furthermore, in experiments targeting a single season, it wasobserved that the evaluation value of the proposed method improves by about 10% compared to the evaluation valuecalculated from actual delivery records. It was also confirmed that the parameters of the rolling horizon approach,specifically W and S, should be prioritized.

In dialogue systems, it is especially important to ask questions that can be answered with either “yes,” “no,” or“I don’t know” intentions (YES-NO questions) in order to confirm the user’s intentions and status, and to accuratelyinterpret the intentions of the user’s answers. In this study, we aimed to perform highly accurate intention estimationfor generic topics, and to be able to determine unknowns as well. Specifically, we created a question answering corpus(Japanese yes-no question-and-answer pairs), designed multiple intention estimators using a large-scale languagemodel, and compared and evaluated their accuracy. As a result, the GPT model with an additional all-combininglayer (Fine-tuned GPT) showed the highest estimation accuracy, achieving 91% accuracy. On the other hand, whenprompt programming was performed using GPT-4 (Few-shot learned GPT-4), we observed a possibility that there wasa tendency to judge as unknown responses for which intention estimation was difficult. The results of this study areexpected to provide valuable guidelines for future research and practical use, as they suggest a policy for selecting anestimation method and tuning a model in the intention estimation task.

In recent years, entertainment content, such as movies, music, and anime, has been gaining attention due tothe stay-at-home demand caused by the expansion of COVID-19. In the content domain, research in the field ofknowledge representation is primarily concerned with accurately describing metadata. Therefore, different knowledgerepresentations are required for applications in downstream tasks. In this study, we aim to clarify effectiveknowledge representation for predicting users’ latent preferences through a case study of an anime recommendationtask. We developed hypotheses from both quantitative and qualitative aspects on how to represent work knowledgeto improve recommendation performance, and verified them by changing the structure of knowledge representationaccording to the hypothesis. Initially, we constructed a Knowledge Graph (KG) by integrating domain-specific andgeneral-purpose data sources through the process of entity matching and imposing constraints on the properties.Subsequently, we constructed multiple KGs by varying the knowledge configuration. Specifically, we changed thecomposition of the data sources considered in the KG construction or excluded a triplet associated with an arbitraryproperty. After that, we fed the constructed KGs into the graph neural network recommender model and comparedthe recommendation performance. As a result, it was shown that the recommendation performance based on the KGcomposed of multiple data sources was the best, thus supporting the hypothesis from a quantitative aspect. Next,an ablation study on the properties revealed that knowledge characterizing the work itself contributed to the recommendationperformance, thus supporting the hypothesis from a qualitative aspect. Furthermore, we constructed atext-based KG by generating a new vocabulary from the “synopsis” text. It can describe the work’s storyline andworldview in more detail. We take it as an input to a Large Language Model (LLM) and extend the existing metadatabasedKG. The results showed that the KG considering both metadata and text had the best overall recommendationperformance, again confirming the hypothesis.

Icons are utilized to visually elucidate functions and objects, and are frequently employed on websites. Forweb developers with limited design skills, producing a substantial volume of icons poses a challenge. Consequently,there is a demand for automated methods to generate icons. The process of creating icons involves two stages: linedrawing and coloring. Recent research on automating the coloring process through deep learning has been active.However, existing methods fail to recognize the entities represented by line drawings, leading to inaccuracies inreflecting the target’s structure during the coloring process. A notable limitation of these methods is their inability toaccurately color line drawings of icons with hollow structures, such as donuts. In this study, we introduce a method forcoloring hollowed-out icons, ensuring adherence to the original line drawings. Specifically, we employ a large-scalepre-trained model to generate icons from line drawings, supplemented by reference images and textual descriptions.To facilitate experimental implementation, we compiled a dataset comprising gradient style icons and their associatedtexts. The results of our experimental evaluation indicate that our method surpasses existing approaches in morethan one metrics. Furthermore, the outcomes of the coloring experiments reveal that our method faces challenges incoloring efficiency and highlight the limitations of incorporating text as an auxiliary input for the coloring process.

During the COVID-19 pandemic, many hospitals faced a surge in patient numbers, leading to fatalities dueto delayed treatment, intensifying the importance of exploring dynamic task sharing. Using the pandemic as an illustration,the task involves treating patients and utilizing limited resources like beds and medical staff. It is crucialto commence this task within a specified timeframe to avoid adverse outcomes. The scheduling of tasks in advanceproves impossible due to the dynamic and asynchronous nature of task creation. The uneven distribution ofresources and tasks among agents underscores the essential need for transferring tasks from busy to less occupiedagents to maintain a well-balanced workload. However, decentralized patient reallocation is deemed necessary, giventhe independent operation of each hospital, emphasizing the pivotal role of negotiation and consensus among agentsin facilitating successful task transfers. In this context, we systematically developed, expanded, and evaluated sixgeneral-purpose decentralized algorithms tailored for emergency task share negotiation. The primary goal was tominimize the number of tasks not initiated within a specified time limit, particularly when the cost of non-executionwas significant. Our findings identified the CSRN (Continuous or Single Random Negotiation) algorithm as the mosteffective, characterized by its consistent negotiation with the same agent that had previously accepted a task transferbefore engaging with a randomly selected agent. Moreover, our observations indicated that negotiating with numerousagents for individual task transfers proved less effective overall. This distinctive comparative study not only providesvaluable insights for optimizing the sharing of patient treatment tasks but also holds relevance for analogous scenarioscharacterized by elevated costs associated with unexecuted tasks. In general, our study provides practical suggestionsfor negotiating strategies related to task transfers among autonomous agents in emergency scenarios characterized bydynamic and unevenly distributed workloads.

In Japan, declining birthrates have necessitated compact cities and reorganization of public transportation,particularly bus routes. Effective transportation planning requires Origin-Destination (OD) data, which is challengingto collect due to high costs. This study develops a system to estimate bus passenger OD from videos captured onroute buses, integrating object detection, tracking, metric learning, and assignment algorithms. Videos of passengersboarding and alighting were recorded, and human bounding boxes were annotated. Model selection and training wereconducted to assess accuracy and identify improvement strategies. The research contributes to cost-effective OD datacollection, offering scalable solutions for public transportation systems. It demonstrates how advanced tracking andmatching algorithms can be applied in urban mobility analytics, potentially optimizing public transport networks.This study enhances public transport efficiency, providing a framework for applying advanced technologies in realworldscenarios and extending methodologies for accurate, cost-effective data collection across various fields.