Graph foundation models (GFMs) with global attention are increasingly used to represent mixed-integer linear programs (MILPs), aiming to capture structure beyond the locality of standard graph neural networks. We study their expressive power through graph isomorphism testing, asking which MILP instances they map to identical representations. We prove that a broad class of hierarchical graph transformers combining global linear attention, edge-weighted cross-attention, and bipartite message passing is bounded by the one-dimensional Weisfeiler-Leman (1-WL) test: under any parameter setting, 1-WL-equivalent MILP graphs receive identical graph embeddings. Our compositional proof shows that each architectural component is a symmetric multiset function and thus preserves 1-WL equivalence. We validate this characterization across ten diverse graph encoders, including Graphormer-, GraphGPS-, Set-Transformer-, and Gasse-style models. Across model capacities, graph scales, and pooling operators, every tested encoder maps 1-WL-equivalent non-isomorphic graph pairs to numerically identical embeddings. Consequently, graph invariants that vary within a 1-WL equivalence class cannot be recovered from these representations. We further show that expressiveness beyond 1-WL arises from input encoding rather than attention: random-walk positional encodings separate the constructed pairs, while additional constructions expose the limits of this remedy. These results characterize the expressive power of global-attention GFMs and provide an encoder-agnostic diagnostic for detecting 1-WL-induced representation equivalence.

Dynamic graph representation learning requires capturing both structural relationships and temporal evolution, yet existing approaches face a fundamental trade-off: attention-based methods achieve expressiveness at $O(T^2)$ complexity, while recurrent architectures suffer from gradient pathologies and dense state storage. Spiking neural networks offer event-driven efficiency but remain limited by sequential propagation, binary information loss, and local aggregation that misses global context. We propose ChronoSpike, an adaptive spiking graph neural network that integrates learnable LIF neurons with per-channel membrane dynamics, multi-head attentive spatial aggregation on continuous features, and a lightweight Transformer temporal encoder, enabling both fine-grained local modeling and long-range dependency capture with linear memory complexity $O(T \cdot d)$. On three large-scale benchmarks, ChronoSpike outperforms twelve state-of-the-art baselines by $2.0\%$ Macro-F1 and $2.4\%$ Micro-F1 while achieving $3-10\times$ faster training than recurrent methods with a constant 105K-parameter budget independent of graph size. We provide theoretical guarantees for membrane potential boundedness, gradient flow stability under contraction factor $ρ< 1$, and BIBO stability; interpretability analyses reveal heterogeneous temporal receptive fields and a learned primacy effect with $83-88\%$ sparsity.

Historical maps contain valuable, detailed survey data often unavailable elsewhere. Automatically extracting linear objects, such as fault lines, from scanned historical maps benefits diverse application areas, such as mining resource prediction. However, existing models encounter challenges in capturing adequate image context and spatial context. Insufficient image context leads to false detections by failing to distinguish desired linear objects from others with similar appearances. Meanwhile, insufficient spatial context hampers the accurate delineation of elongated, slender-shaped linear objects. This paper introduces the Linear Object Detection TRansformer (LDTR), which directly generates accurate vector graphs for linear objects from scanned map images. LDTR leverages multi-scale deformable attention to capture representative image context, reducing false detections. Furthermore, LDTR’s innovative N-hop connectivity component explicitly encourages interactions among nodes within an N-hop neighborhood, enabling the model to learn sufficient spatial context for generating graphs with accurate connectivity. Experiments show that LDTR improves detection precision by 6

Historical geologic maps contain rich geospatial information, such as rock units, faults, folds, and bedding planes, that is critical for assessing mineral resources essential to renewable energy, electric vehicles, and national security. However, digitizing maps remains a labor-intensive and time-consuming task. We present DIGMAPPER, a modular, scalable system developed in collaboration with the United States Geological Survey (USGS) to automate the digitization of geologic maps. DIGMAPPER features a fully dockerized, workflow-orchestrated architecture that integrates state-of-the-art deep learning models for map layout analysis, feature extraction, and georeferencing. To overcome challenges such as limited training data and complex visual content, our system employs innovative techniques, including in-context learning with large language models, synthetic data generation, and transformer-based models. Evaluations on over 100 annotated maps from the DARPA-USGS dataset demonstrate high accuracy across polygon, line, and point feature extraction, and reliable georeferencing performance. Deployed at USGS, DIGMAPPER significantly accelerates the creation of analysis-ready geospatial datasets, supporting national-scale critical mineral assessments and broader geoscientific applications.

Geospatial data are critical for urban planning and smart city applications, yet understanding and classifying geo-entities in diverse datasets remains challenging. Accurate representation and classification of geo-entities are essential for tasks such as geo-entity typing and linking, enabling better map understanding and informed decision-making. This paper presents a self-supervised learning approach to classify geo-entities by embedding their geometric, spatial, and semantic neighborhood contexts, creating robust representations for geo-entity typing. Using OpenStreetMap (OSM) data, our method links geo-referenced entities to Wikidata classes and OSM tags with high performance, achieving an..1 score of approximately 0.85. Beyond the technical contribution, our method addresses Responsible AI challenges, including transparency, and data standardization on the Web, aligning with sustainable smart city development.

This paper presents a method for constructing a lightweight taxonomy of geospatial features using OpenStreetMap (OSM) data. Leveraging the OSM data model, our process mines frequent tags to efficiently produce a structured hierarchy, enriching the semantic representation of geo-features. This data-driven taxonomy supports various geospatial analysis applications. Accompanying the methodology, we release the source code of our tool and demonstrate its practical application with tailored taxonomies for California (US) and Greece, underscoring our approach’s adaptability and scalability.

Road inspection and cleaning are crucial to securing driving safety. Deploying a fleet of robots that run through a city can inspect and clean pavements without causing road closure. To achieve high coverage, one has to prevent robots from going through a road segment more than once. However, robots may need more than one visit to a particular road segment to inspect a defect. The uncertain success rate of defect inspection and the unknown maximum number of defects hinder the efficacy. Such uncertainty and constraints in objectives can also be seen in security patrolling, trip planning, and network maintenance. We target the problem of multi-agent scheduling with bounded objectives. The scheduling aims for maximum road network coverage while ensuring sufficient visits to particular road segments for defect identification of an uncertain subject, such as potholes and faded markings during road inspection or crimes and parking violations during security patrolling. We leverage an approximate bi-objective algorithm and propose a hierarchical circular route-planning algorithm. Our approach maximizes the road coverage among robots and decreases the search space when maximizing defect identification. Evaluation on a real-world dataset shows that our approach achieves the Pareto optimal among comparative methods, outperforming existing methods by at least one optimization objective.

Understanding the semantic structure of tabular data is essential for data integration and discovery. Specifically, the goal is to annotate columns in a tabular source with types and relationships between them using classes and predicates of a target ontology. Previous work that exploits the matches between entities in a knowledge graph and the table data does not perform well for tables with noisy or ambiguous data. A key reason for this poor performance is the limited amount of labeled data to train these methods. To address this problem, we propose a novel distant supervision approach that leverages existing Wikipedia tables and hyperlinks to automatically label tables with their semantic descriptions. Then, we use the labeled dataset to train neural network models to predict the semantic description of a new table. Our empirical evaluation shows that using the automatically labeled dataset provides approximately 5

Locating undiscovered deposits of critical minerals requires accurate geological data. However, most of the 100,000 historical geological maps of the United States Geological Survey (USGS) are in raster format. This hinders critical mineral assessment. We target the problem of extracting geological features represented as polygons from raster maps. We exploit the polygon metadata that provides information on the geological features, such as the map keys indicating how the polygon features are represented, to extract the features. We present a metadata-driven machine-learning approach that encodes the raster map and map key into a series of bitmaps and uses a convolutional model to learn to recognize the polygon features. We evaluated our approach on USGS geological maps; our approach achieves a median F1 score of 0.809 and outperforms state-of-the-art methods by 4.52%.

Private vehicles are a direct means to bring people from one place to their desired destinations. However, no omniscient dispatcher is handling the origin-destination of vehicles and the availability of stationary resources, such as parking spaces or charging stations. Competitive cruising for stationary resources leads to environmental pollution and is a waste of drivers' time. We focus on the problem of distributed stationary-resource searching with predefined destinations under a multi-agent scenario. It is a distributed route planning problem with global optimization objectives. We present a probabilistic approach to achieving indirect resource coordination and latent agent cooperation in a distributed manner. Our approach treats the estimated availability of stationary resources as a reference and guides each agent based on their preferences. We evaluate our approach on four real-world datasets. Our approach outperforms state-of-the-art methods by 5% in multi-criteria optimization.

Tables can contain various types of errors, including both syntactic and semantic errors. Semantic errors relate to the meaning of the data and can be detrimental for downstream applications. The existing approaches for semantic error detection use structured knowledge sources such as Wikidata and DBpedia, but the coverage of such sources is quite limited. There is much more information available in free text to validate the contents of tables. In this paper, we present a novel semantic-error-detection approach that exploits open-domain textual data to verify the semantic correctness of tables. Our approach leverages contrastive learning, table linearization, and pre-trained language models to implement the error detection process. We implement our approach in a system called SEED and show in the evaluation that it significantly outperforms the other competing approaches.

Historical maps provide rich information for researchers in many areas, including the social and natural sciences. These maps contain detailed documentation of a wide variety of natural and human-made features and their changes over time, such as changes in transportation networks or the decline of wetlands or forest areas. Analyzing changes over time in such maps can be labor-intensive for a scientist, even after the geographic features have been digitized and converted to a vector format. Knowledge Graphs (KGs) are the appropriate representations to store and link such data and support semantic and temporal querying to facilitate change analysis. KGs combine expressivity, interoperability, and standardization in the Semantic Web stack, thus providing a strong foundation for querying and analysis. In this paper, we present an automatic approach to convert vector geographic features extracted from multiple historical maps into contextualized spatio-temporal KGs. The resulting graphs can be easily queried and visualized to understand the changes in different regions over time. We evaluate our technique on railroad networks and wetland areas extracted from the United States Geological Survey (USGS) historical topographic maps for several regions over multiple map sheets and editions. We also demonstrate how the automatically constructed linked data (i.e., KGs) enable effective querying and visualization of changes over different points in time.

1 Historical maps capture past landscapes' natural and anthropogenic features, with geohistorical data from periods before the 1970s (before the Landsat program's launch) primarily found, barring a few exceptions, only on printed map sheets. In the past decade, numerous maps have been digitized and made publicly accessible. This chapter overviews cutting-edge AI methods and systems for processing historical maps to generate valuable data, insights, and knowledge. Individual sections highlight our recently published research findings across various domains, including the semantic web, big data, data mining, machine learning, document understanding, natural language processing, remote sensing, and geographic information systems. 1 Contact: Yao-Yi Chiang. Email: yaoyi@umn.edu. All other authors are listed in alphabetical order.

Building semantic descriptions of tables is a vital step in data integration. However, this task is expensive and time-consuming as users often need to examine the table data, its metadata, and ontologies to find the most appropriate description. In this paper, we present SAND, a tool for creating semantic descriptions semi-automatically. SAND makes it easy to integrate with semantic modeling systems to predict or suggest semantic descriptions to the users, as well as to use different knowledge graphs (KGs). Besides its modeling capabilities, SAND is equipped with browsing/querying tools to enable users to explore data in the table and discover how it is often modeled in KGs.

Competition among companies is a crucial aspect of the economic landscape and is often key to differentiating business success and failure. Understanding the core business functions of both public and private companies is essential for regulators, investors, and researchers. Prior work has characterized competition using inflexible industry classifications, financial metrics with limited scope, or using text analytics that has narrow coverage or uses proprietary signals. Most important, the focus has been on public companies due to a dearth of data on private companies. We present a new approach to characterize competition among companies, and to construct a competitor network using company Web pages. Our approach uses state-of-the-art representation learning techniques to generate robust representations of companies in a high-dimensional vector space. We use these learned representations and a range of algorithms to accurately identify the competitor network. We evaluate the competitor network against three downstream applications: predicting profitability, determining industry classification codes, and predicting self-identified competitors. Notably, our methods improve over prior state-of-the-art models by as much as 30% when using clean data. Furthermore, our models match the state-of-the-art performance of prior models required corpora of curated regulatory filings while only using much noisier Web text from company web pages. Most importantly, our approach is not limited to a small number of public companies, and can exploit Web content that covers millions of publicly traded and privately held companies.

Transportation infrastructure, such as road or railroad networks, represent a fundamental component of our civilization. For sustainable planning and informed decision making, a thorough understanding of the long-term evolution of transportation infrastructure such as road networks is crucial. However, spatially explicit, multi-temporal road network data covering large spatial extents are scarce and rarely available prior to the 2000s. Herein, we propose a framework that employs increasingly available scanned and georeferenced historical map series to reconstruct past road networks, by integrating abundant, contemporary road network data and color information extracted from historical maps. Specifically, our method uses contemporary road segments as analytical units and extracts historical roads by inferring their existence in historical map series based on image processing and clustering techniques. We tested our method on over 300,000 road segments representing more than 50,000 km of the road network in the United States, extending across three study areas that cover 42 historical topographic map sheets dated between 1890 and 1950. We evaluated our approach by comparison to other historical datasets and against manually created reference data, achieving F-1 scores of up to 0.95, and showed that the extracted road network statistics are highly plausible over time, i.e., following general growth patterns. We demonstrated that contemporary geospatial data integrated with information extracted from historical map series open up new avenues for the quantitative analysis of long-term urbanization processes and landscape changes far beyond the era of operational remote sensing and digital cartography.

Many historical map sheets are publicly available for studies that require long-term historical geographic data. The cartographic design of these maps includes a combination of map symbols and text labels. Automatically reading text labels from map images could greatly speed up the map interpretation and helps generate rich metadata describing the map content. Many text detection algorithms have been proposed to locate text regions in map images automatically, but most of the algorithms are trained on out-of-domain datasets (e.g., scenic images). Training data determines the quality of machine learning models, and manually annotating text regions in map images is labor-extensive and time-consuming. On the other hand, existing geographic data sources, such as Open-StreetMap (OSM), contain machine-readable map layers, which allow us to separate out the text layer and obtain text label annotations easily. However, the cartographic styles between OSM map tiles and historical maps are significantly different. This paper proposes a method to automatically generate an unlimited amount of annotated historical map images for training text detection models. We use a style transfer model to convert contemporary map images into historical style and place text labels upon them. We show that the state-of-the-art text detection models (e.g., PSENet) can benefit from the synthetic historical maps and achieve significant improvement for historical map text detection.