With the increasing concerns about privacy and data regulations, federated learning (FL) has been emerging as a solution to train machine learning models collaboratively with non-exchangeable data from multiple clients. As a result of data locality, data is usually not identically or independently (non-IID) distributed across clients, and the non-IID property has long been the key challenge in FL. Furthermore, in real-world cross-silo scenarios, it is ubiquitous that clients are organizations owning private data from multiple domains internally, which exacerbates the non-IID issue. For example, in healthcare applications, each client (hospital) gathers data from patients with heterogeneous demographics. While previous works have made efforts to address the non-IID challenge across clients by assuming various relations among client-level data distributions and enabling personalized models at the client level, they ignore the internal data heterogeneity within each client or require explicit data domain indicators, which are hardly accessible in real-world data. Here, we propose Sample-Level Prototypical Federated Learning (SL-PFL) to bridge the gap. SL-PFL incorporates prototypical learning under the FL framework and provides a fine-grained personalized model for each data sample instead of learning one uniform model for all samples of each client. Meanwhile, it can be trained using data without ground-truth domain indicators. Experimental results demonstrate that our proposed method with sample-level personalized models outperforms existing FL methods with a global model or client-level personalized models on various real-world regression and classification tasks from weather, computer vision, and healthcare applications.

The rise of Large Visual-Language Models (LVLMs) has unlocked new possibilities for seamlessly integrating visual and textual information. However, their ability to interpret cartographic maps remains largely unexplored. In this paper, we introduce CartoMapQA, a benchmark specifically designed to evaluate LVLMs' understanding of cartographic maps through question-answering tasks. The dataset includes over 2000 samples, each composed of a cartographic map, a question (with open-ended or multiple-choice answers), and a ground-truth answer. These tasks span key low-, mid- and high-level map interpretation skills, including symbol recognition, embedded information extraction, scale interpretation, and route-based reasoning. Our evaluation of both open-source and proprietary LVLMs reveals persistent challenges: models frequently struggle with map-specific semantics, exhibit limited geospatial reasoning, and are prone to Optical Character Recognition (OCR)-related errors. By isolating these weaknesses, CartoMapQA offers a valuable tool for guiding future improvements in LVLM architectures. Ultimately, it supports the development of models better equipped for real-world applications that depend on robust and reliable map understanding, such as navigation, geographic search, and urban planning. Our source code and data are openly available to the research community at: https://github.com/ungquanghuy-kddi/CartoMapQA.git

Multivariate long-term time series forecasting (MLTSF), applicable across various domains, has gained increasing research attention. Channel-independent (CI) models, including Linear and Transformer-based architectures, have recently achieved state-of-the-art (SOTA) performance for MLTSF. Notably, Linear models can deliver satisfactory forecasting performance even with just a single linear projection layer. However, we identify a limitation in this architecture: a single linear projection struggles to adequately capture the inter- and intra-variate heterogeneity in temporal patterns. Similarly, any complex models like Transformer-based models that use a single projection layer to generate final predictions, may face capacity bottlenecks. To overcome this, we propose the Mixture of Projection Experts (MoPE), which replaces the single linear projection with multiple projection branches, and employs a gate network to dynamically assign weights to each branch based on the input data. We applied MoPE to multiple SOTA models and evaluated it on nine real-world datasets. Results show that MoPE boosts forecasting accuracy by an average of 9.59%, demonstrating its effectiveness in mitigating the limitations of a single projection layer. Additionally, our experiments demonstrate that integrating our proposal into CI models enhances their generalization to unseen variates. Interpretability analysis also reveals MoPE's ability to disentangle different temporal patterns. Overall, our paper establishes MoPE as an effective solution for MLTSF tasks.

Existing research on task incremental learning in continual learning has primarily focused on preventing catastrophic forgetting (CF). Although several techniques have achieved learning with no CF, they attain it by letting each task monopolize a sub-network in a shared network, which seriously limits knowledge transfer (KT) and causes over-consumption of the network capacity, i.e., as more tasks are learned, the performance deteriorates. The goal of this paper is threefold: (1) overcoming CF, (2) encouraging KT, and (3) tackling the capacity problem. A novel technique (called SPG) is proposed that soft-masks (partially blocks) parameter updating in training based on the importance of each parameter to old tasks. Each task still uses the full network, i.e., no monopoly of any part of the network by any task, which enables maximum KT and reduction in capacity usage. To our knowledge, this is the first work that soft-masks a model at the parameter-level for continual learning. Extensive experiments demonstrate the effectiveness of SPG in achieving all three objectives. More notably, it attains significant transfer of knowledge not only among similar tasks (with shared knowledge) but also among dissimilar tasks (with little shared knowledge) while mitigating CF.

A major issue with real-time monitoring is to collect complete data. Hardware or software failures, network issues or, more frequently, time delays can disrupt such a collection. This results in having two versions of the same information: one in real-time but with potentially missing data, and the another, albeit complete, is delayed. Many works have studied how to handle missing data for classification and prediction. However, to the best of our knowledge, they do not consider how to leverage the delayed complete data to assist in learning the representation of real-time available data with missing values. This is despite the fact that the delayed complete data contain all the information (e.g., periodicities and trends). In this paper, we propose a framework to enhance the representation learning of the real-time available data by aligning the representation of past real-time but with missing data to that of past delayed but complete data. We test both a distance metric and contrastive learning to achieve this alignment. We implement our framework on a Transformer-based model and experiment it on three datasets. The efficiency of our solution is evaluated against seven baselines and considering four distinct patterns of missing data. Our experiments show that this proposal has a significant improvement in prediction accuracy (5.21% on average) over the baselines.

Knowledge graphs (KGs) are useful information sources to make machine learning efficient with human knowledge. Since KGs are often incomplete, KG completion has become an important problem to complete missing facts in KGs. Whereas most of the KG completion methods are conducted on a single KG, multiple KGs can be effective to enrich embedding space for KG completion. However, most of the recent studies have concentrated on entity alignment prediction and ignored KG-invariant semantics in multiple KGs that can improve the completion performance. In this paper, we propose a new multiple KG embedding method composed of intra-KG and inter-KG regularization to introduce KG-invariant semantics into KG embedding space using aligned entities between related KGs. The intra-KG regularization adjusts local distance between aligned and not-aligned entities using contrastive loss, while the inter-KG regularization globally correlates aligned entity embeddings between KGs using multi-view loss. Our experimental results demonstrate that our proposed method combining both regularization terms largely outperforms existing baselines in the KG completion task.

Spatiotemporal data aggregated over regions or time windows at various resolutions demonstrate heterogeneous patterns and dynamics in each resolution. Meanwhile, the multi-resolution characteristic provides rich contextual information, which is critical for effective long-sequence forecasting. The importance of such inter-resolution information is more significant in practical cases, where fine-grained data is usually collected via approaches with lower costs but also lower qualities compared to those for coarse-grained data. However, existing works focus on uni-resolution data and cannot be directly applied to fully utilize the aforementioned extra information in multi-resolution data. In this work, we propose Spatiotemporal Koopman Multi-Resolution Network (ST-KMRN), a physics-informed learning framework for long-sequence forecasting from multi-resolution spatiotemporal data. Our method jointly models data aggregated in multiple resolutions and captures the inter-resolution dynamics with the self-attention mechanism. We also propose downsampling and upsampling modules among resolutions to further strengthen the connections among data of multiple resolutions. Moreover, we enhance the modeling of intra-resolution dynamics with physics-informed modules based on Koopman theory. Experimental results demonstrate that our proposed approach achieves the best performance on the long-sequence forecasting tasks compared to baselines without a specific design for multi-resolution data.

In a world where the growing concern about climate change becomes more apparent, electricity industry has its role to play in slowing down such an evolution. In fact, tools and services could be provided to consumers in order to better manage their ever-increasing consumption. But, while waiting for such a complex solution to be deployed, a first step would be to better forecast current consumption. Such an improvement would enable to generate electricity accordingly, while renewable solutions are not presently optimal – i.e., intermittent, long-term storage issue, etc.Electricity Load Demand (ELD) fluctuations do not solely depend on temporal factors, they also have a spatial contribution that is often underestimated. Indeed, it is possible to map different ELD profiles with their Land Use (LU) information. For example, ELD in residential areas fits with residents’ commuting style, while in industrial areas, it mostly correlates with working hours. This paper aims to investigate such a relation. Considering ELD data on a 500m square grid and LU data on a 100m square grid, different 500m cell labeling has been defined. In order to assess if cells with similar labels have similar ELD, they have been grouped under the same forecast model. Several types of models (Multilayer Perceptron, Recurrent Neural Network) have been used to compare their performance and efficiency. This study confirms that, all models considered, some labels are more difficult to forecast than others. Such associations can reduce by over 30% the prediction error compare to a per cell scenario. Additional investigations would be needed to further reduce prediction error and to help models better seize the land specificity of each grid-cell. External data also affects ELD, and pairing them with an optimal LU labeling could be a promising solution.

The need for electricity has never been more important these days. In order to achieve balance between generation and distribution - as well as schedule operations accordingly-high-accuracy load demand predictions are mandatory. But our society is currently undergoing modifications in electricity consumption allocation. We are witnessing a fast shift from office-to home- based working style. As a consequence, Electricity load demand prediction models are in need for additional data in order to quickly adapt to these modifications and maintain efficient predictions accuracy. The rising popularity of "tracking" devices and alike-applications opens up to a new type of multi-modal investigations. The availability of associated location data enables researcher to study mobility routine and patterns in order to cross it with other data. Electricity consumption is one domain impacted by people's mobility behavior (commuting, telework, etc.) as people are not "plugged" onto the power grid while moving. This paper presents a study on population staying patterns and how it can relate to electricity load demand. Time-series data providing the number of people staying in a given area has been used within a Deep Learning model in order to enhance electricity load demand predictions at the provider level. It unveils the potential usage of such dynamics data, while setting the foundations for more complex studies.

The landscape of power grids is gradually changing. The growing number of electrical appliances as well as the outbreak of Electric Vehicles (EVs) is increasing the need for electricity. As a consequence, high-accuracy consumption predictions are necessary in order to both schedule and plan production and operations accordingly. The emergence of connected “tracking” devices opens up new data-sets into both Internet-of-Things and Big Data worlds. It provides information on human dynamics (people mobility behavior) and with it several opportunities. Electricity consumption is impacted by people movements as while moving they are not “connected” to the power grid. Therefore, predicting such movement patterns and volume could help electricity providers improve their own consumption predictions. This paper presents a system and methods used to predict people movement behaviors as well as detect any anomaly. A scoring system is used to both evaluate the dynamics predictions and raise alerts when the computed score surpasses established thresholds. This proposal is tested over a scenario using available datasets and demonstrates that modifications in people movement behavior is affecting the consumption profile.

With the popularization of smart sensors, detecting activities of daily living (ADL) from sensor readings has attracted many interests in both the academic and the industrial societies. A majority of research works on this topic focus on data of high sampling rate, however, most existing smart sensor deployments support sampling rate much lower than 1 Hz. We are interested in the possibility of inferring ADLs from solely coarse home-level gross power consumptions. In this paper, we first tentatively adopt a layered hidden Markov model (LHMM) in the hope to uncover the association between ADLs and power consumption data. We conduct an exploratory data analysis with this preliminary model on a real-world dataset, and based on the findings from this exploratory study, we propose to infer ADLs from low frequency power consumption data using a hierarchical Dirichlet process hidden markov model (HDP-HMM). We perform experiments on the same dataset, and demonstrate that with sensor readings of 1/180 Hz and 1/900 Hz granularities, HDP-HMM outperforms comparative models and ADLs such as "Entertaining" and "Not at home" can be captured with high accuracy.

This research proposes a novel method to present "thumbnails" of episodes of digitized comics, in order to improve the efficiency of comic search. Comic episode thumbnails are generated based on image analysis technologies developed especially for comic images. Namely, the following procedures are developed for our system: automatic comic frame segmentation, text balloon extraction, and a linear regression based model to calculate the importance score of each extracted frame. The system then selects frames from each episode with high importance score, and aligns the selected frames to create the episode thumbnail, which is presented to the system user as a compact preview of the episode. User experiments conducted with actual Japanese comic images prove that the proposed method significantly decreases the time necessary to search for specific episodes from a large scaled comic data collection.

We discuss the design issues on lightweight and FIPA compliant agent platform for Java-enabled mobile phones and describe the design of such agent platform. This platform changes Java-enabled mobile phones to ubiquitous terminals by providing place for agent applications. Combined with location services, it can be used for various ubiquitous services. We also show the performance comparison of the prototype with LEAP, another lightweight agent platform.

OSI Management model defines only peer to peer communication process between a managing application (AP) and a managed AP. Thus, when more than one managed AP stores portions of Management Information Base (MIB) distributively, the managing AP needs a sophisticated distributed processing function. This paper proposes an agent, called Management System Agent (MSA) which provides location transparency and schema transparency of MIB for managing APs, to ease the realization of managing APs. Location transparency hides the distribution of MIB from managing APs. Schema transparency also enables managing APs to invoke management operation on unified schema. This paper demonstrates the effectiveness of the MSA model through applying it to implementation of Customer Network Management (CNM) system focusing X.25 public data networks.