arXiv - CS - Artificial Intelligence最新文献_第6页

An Argumentative Approach for Explaining Preemption in Soft-Constraint Based Norms 解释基于软约束的规范中优先权的论证方法

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-06 DOI: arxiv-2409.04065

Wachara Fungwacharakorn, Kanae Tsushima, Hiroshi Hosobe, Hideaki Takeda, Ken Satoh

引用次数: 0

TRACE-cs: Trustworthy Reasoning for Contrastive Explanations in Course Scheduling Problems TRACE-cs：课程安排问题中对比解释的可信推理

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-05 DOI: arxiv-2409.03671

Stylianos Loukas Vasileiou, William Yeoh

引用次数: 0

HUMOS: Human Motion Model Conditioned on Body Shape HUMOS：以体形为条件的人体运动模型

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-05 DOI: arxiv-2409.03944

Shashank Tripathi, Omid Taheri, Christoph Lassner, Michael J. Black, Daniel Holden, Carsten Stoll

引用次数: 0

In Search of Trees: Decision-Tree Policy Synthesis for Black-Box Systems via Search 寻找树：通过搜索为黑盒系统合成决策树策略

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-05 DOI: arxiv-2409.03260

Emir Demirović, Christian Schilling, Anna Lukina

{"title":"In Search of Trees: Decision-Tree Policy Synthesis for Black-Box Systems via Search","authors":"Emir Demirović, Christian Schilling, Anna Lukina","doi":"arxiv-2409.03260","DOIUrl":"https://doi.org/arxiv-2409.03260","url":null,"abstract":"Decision trees, owing to their interpretability, are attractive as control\u0000policies for (dynamical) systems. Unfortunately, constructing, or synthesising,\u0000such policies is a challenging task. Previous approaches do so by imitating a\u0000neural-network policy, approximating a tabular policy obtained via formal\u0000synthesis, employing reinforcement learning, or modelling the problem as a\u0000mixed-integer linear program. However, these works may require access to a\u0000hard-to-obtain accurate policy or a formal model of the environment (within\u0000reach of formal synthesis), and may not provide guarantees on the quality or\u0000size of the final tree policy. In contrast, we present an approach to\u0000synthesise optimal decision-tree policies given a black-box environment and\u0000specification, and a discretisation of the tree predicates, where optimality is\u0000defined with respect to the number of steps to achieve the goal. Our approach\u0000is a specialised search algorithm which systematically explores the\u0000(exponentially large) space of decision trees under the given discretisation.\u0000The key component is a novel pruning mechanism that significantly reduces the\u0000search space. Our approach represents a conceptually novel way of synthesising\u0000small decision-tree policies with optimality guarantees even for black-box\u0000environments with black-box specifications.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"54 2 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142193864","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

TC-LLaVA: Rethinking the Transfer from Image to Video Understanding with Temporal Considerations TC-LaVA：重新思考从图像到视频的时空理解转换

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-05 DOI: arxiv-2409.03206

Mingze Gao, Jingyu Liu, Mingda Li, Jiangtao Xie, Qingbin Liu, Bo Zhao, Xi Chen, Hui Xiong

{"title":"TC-LLaVA: Rethinking the Transfer from Image to Video Understanding with Temporal Considerations","authors":"Mingze Gao, Jingyu Liu, Mingda Li, Jiangtao Xie, Qingbin Liu, Bo Zhao, Xi Chen, Hui Xiong","doi":"arxiv-2409.03206","DOIUrl":"https://doi.org/arxiv-2409.03206","url":null,"abstract":"Multimodal Large Language Models (MLLMs) have significantly improved\u0000performance across various image-language applications. Recently, there has\u0000been a growing interest in adapting image pre-trained MLLMs for video-related\u0000tasks. However, most efforts concentrate on enhancing the vision encoder and\u0000projector components, while the core part, Large Language Models (LLMs),\u0000remains comparatively under-explored. In this paper, we propose two strategies\u0000to enhance the model's capability in video understanding tasks by improving\u0000inter-layer attention computation in LLMs. Specifically, the first approach\u0000focuses on the enhancement of Rotary Position Embedding (RoPE) with\u0000Temporal-Aware Dual RoPE, which introduces temporal position information to\u0000strengthen the MLLM's temporal modeling capabilities while preserving the\u0000relative position relationships of both visual and text tokens. The second\u0000approach involves enhancing the Attention Mask with the Frame-wise Block Causal\u0000Attention Mask, a simple yet effective method that broadens visual token\u0000interactions within and across video frames while maintaining the causal\u0000inference mechanism. Based on these proposed methods, we adapt LLaVA for video\u0000understanding tasks, naming it Temporal-Considered LLaVA (TC-LLaVA). Our\u0000TC-LLaVA achieves new state-of-the-art performance across various video\u0000understanding benchmarks with only supervised fine-tuning (SFT) on\u0000video-related datasets.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"84 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142193867","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Harnessing LLMs for Cross-City OD Flow Prediction 利用 LLM 进行跨城市外径流量预测

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-05 DOI: arxiv-2409.03937

Chenyang Yu, Xinpeng Xie, Yan Huang, Chenxi Qiu

{"title":"Harnessing LLMs for Cross-City OD Flow Prediction","authors":"Chenyang Yu, Xinpeng Xie, Yan Huang, Chenxi Qiu","doi":"arxiv-2409.03937","DOIUrl":"https://doi.org/arxiv-2409.03937","url":null,"abstract":"Understanding and predicting Origin-Destination (OD) flows is crucial for\u0000urban planning and transportation management. Traditional OD prediction models,\u0000while effective within single cities, often face limitations when applied\u0000across different cities due to varied traffic conditions, urban layouts, and\u0000socio-economic factors. In this paper, by employing Large Language Models\u0000(LLMs), we introduce a new method for cross-city OD flow prediction. Our\u0000approach leverages the advanced semantic understanding and contextual learning\u0000capabilities of LLMs to bridge the gap between cities with different\u0000characteristics, providing a robust and adaptable solution for accurate OD flow\u0000prediction that can be transferred from one city to another. Our novel\u0000framework involves four major components: collecting OD training datasets from\u0000a source city, instruction-tuning the LLMs, predicting destination POIs in a\u0000target city, and identifying the locations that best match the predicted\u0000destination POIs. We introduce a new loss function that integrates POI\u0000semantics and trip distance during training. By extracting high-quality\u0000semantic features from human mobility and POI data, the model understands\u0000spatial and functional relationships within urban spaces and captures\u0000interactions between individuals and various POIs. Extensive experimental\u0000results demonstrate the superiority of our approach over the state-of-the-art\u0000learning-based methods in cross-city OD flow prediction.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"81 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142193861","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

InfraLib: Enabling Reinforcement Learning and Decision Making for Large Scale Infrastructure Management InfraLib：为大规模基础设施管理提供强化学习和决策支持

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-05 DOI: arxiv-2409.03167

Pranay Thangeda, Trevor S. Betz, Michael N. Grussing, Melkior Ornik

{"title":"InfraLib: Enabling Reinforcement Learning and Decision Making for Large Scale Infrastructure Management","authors":"Pranay Thangeda, Trevor S. Betz, Michael N. Grussing, Melkior Ornik","doi":"arxiv-2409.03167","DOIUrl":"https://doi.org/arxiv-2409.03167","url":null,"abstract":"Efficient management of infrastructure systems is crucial for economic\u0000stability, sustainability, and public safety. However, infrastructure\u0000management is challenging due to the vast scale of systems, stochastic\u0000deterioration of components, partial observability, and resource constraints.\u0000While data-driven approaches like reinforcement learning (RL) offer a promising\u0000avenue for optimizing management policies, their application to infrastructure\u0000has been limited by the lack of suitable simulation environments. We introduce\u0000InfraLib, a comprehensive framework for modeling and analyzing infrastructure\u0000management problems. InfraLib employs a hierarchical, stochastic approach to\u0000realistically model infrastructure systems and their deterioration. It supports\u0000practical functionality such as modeling component unavailability, cyclical\u0000budgets, and catastrophic failures. To facilitate research, InfraLib provides\u0000tools for expert data collection, simulation-driven analysis, and\u0000visualization. We demonstrate InfraLib's capabilities through case studies on a\u0000real-world road network and a synthetic benchmark with 100,000 components.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"37 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142193873","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Sequential Decision-Making Model for Perimeter Identification 周界识别的顺序决策模型

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-04 DOI: arxiv-2409.02549

Ayal Taitler

引用次数: 0

Decision Transformer for Enhancing Neural Local Search on the Job Shop Scheduling Problem 在工作车间调度问题上增强神经局部搜索的决策变换器

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-04 DOI: arxiv-2409.02697

Constantin Waubert de Puiseau, Fabian Wolz, Merlin Montag, Jannik Peters, Hasan Tercan, Tobias Meisen

{"title":"Decision Transformer for Enhancing Neural Local Search on the Job Shop Scheduling Problem","authors":"Constantin Waubert de Puiseau, Fabian Wolz, Merlin Montag, Jannik Peters, Hasan Tercan, Tobias Meisen","doi":"arxiv-2409.02697","DOIUrl":"https://doi.org/arxiv-2409.02697","url":null,"abstract":"The job shop scheduling problem (JSSP) and its solution algorithms have been\u0000of enduring interest in both academia and industry for decades. In recent\u0000years, machine learning (ML) is playing an increasingly important role in\u0000advancing existing and building new heuristic solutions for the JSSP, aiming to\u0000find better solutions in shorter computation times. In this paper we build on\u0000top of a state-of-the-art deep reinforcement learning (DRL) agent, called\u0000Neural Local Search (NLS), which can efficiently and effectively control a\u0000large local neighborhood search on the JSSP. In particular, we develop a method\u0000for training the decision transformer (DT) algorithm on search trajectories\u0000taken by a trained NLS agent to further improve upon the learned\u0000decision-making sequences. Our experiments show that the DT successfully learns\u0000local search strategies that are different and, in many cases, more effective\u0000than those of the NLS agent itself. In terms of the tradeoff between solution\u0000quality and acceptable computational time needed for the search, the DT is\u0000particularly superior in application scenarios where longer computational times\u0000are acceptable. In this case, it makes up for the longer inference times\u0000required per search step, which are caused by the larger neural network\u0000architecture, through better quality decisions per step. Thereby, the DT\u0000achieves state-of-the-art results for solving the JSSP with ML-enhanced search.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"45 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142193869","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Creating a Gen-AI based Track and Trace Assistant MVP (SuperTracy) for PostNL 为 PostNL 创建基于 Gen-AI 的跟踪与追踪助理 MVP (SuperTracy)

arXiv - CS - Artificial Intelligence Pub Date : 2024-09-04 DOI: arxiv-2409.02711

Mohammad Reshadati

{"title":"Creating a Gen-AI based Track and Trace Assistant MVP (SuperTracy) for PostNL","authors":"Mohammad Reshadati","doi":"arxiv-2409.02711","DOIUrl":"https://doi.org/arxiv-2409.02711","url":null,"abstract":"The developments in the field of generative AI has brought a lot of\u0000opportunities for companies, for instance to improve efficiency in customer\u0000service and automating tasks. PostNL, the biggest parcel and E-commerce\u0000corporation of the Netherlands wants to use generative AI to enhance the\u0000communication around track and trace of parcels. During the internship a\u0000Minimal Viable Product (MVP) is created to showcase the value of using\u0000generative AI technologies, to enhance parcel tracking, analyzing the parcel's\u0000journey and being able to communicate about it in an easy to understand manner.\u0000The primary goal was to develop an in-house LLM-based system, reducing\u0000dependency on external platforms and establishing the feasibility of a\u0000dedicated generative AI team within the company. This multi-agent LLM based\u0000system aimed to construct parcel journey stories and identify logistical\u0000disruptions with heightened efficiency and accuracy. The research involved\u0000deploying a sophisticated AI-driven communication system, employing\u0000Retrieval-Augmented Generation (RAG) for enhanced response precision, and\u0000optimizing large language models (LLMs) tailored to domain specific tasks. The MVP successfully implemented a multi-agent open-source LLM system, called\u0000SuperTracy. SuperTracy is capable of autonomously managing a broad spectrum of\u0000user inquiries and improving internal knowledge handling. Results and\u0000evaluation demonstrated technological innovation and feasibility, notably in\u0000communication about the track and trace of a parcel, which exceeded initial\u0000expectations. These advancements highlight the potential of AI-driven solutions\u0000in logistics, suggesting many opportunities for further refinement and broader\u0000implementation within PostNL operational framework.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"53 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142193871","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0