东京大学

Large language models have exhibited impressive performance across a broad range of downstream tasks in natural language processing. However, how a language model predicts the next token and generates content is not generally understandable to humans. Furthermore, these models often make errors in prediction and reasoning, known as hallucinations. These errors underscore the urgent need to better understand and interpret the intricate inner workings of language models and how they generate predictive outputs. Motivated by this gap, this paper investigates local explainability and mechanistic interpretability in Transformer-based large language models to foster trust in them. In this regard, our paper aims to make three key contributions. First, we present a review of local explainability and mechanistic interpretability approaches, along with insights from relevant studies in the literature. Furthermore, we describe experimental studies on explainability and reasoning with large language models in two critical domains – healthcare and autonomous driving – and analyze the trust implications of such explanations for explanation receivers. Finally, we summarize the current unaddressed issues in the evolving landscape of LLM explainability and outline opportunities, critical challenges, and future directions for generating human-aligned, trustworthy LLM explanations.

The Kullback–Leibler divergence together with exponential families establishes the foundation of information geometry and is widely generalized. Among the generalization, we focus on the (h,τ)-divergence and (h,τ)-exponential families. We present a sufficient condition for the (h,τ)-divergence to induce a Hessian structure on an (h,τ)-exponential family. We also define the (h,τ)-dependence of random variables and prove a kind of the law of large numbers.

Topic discovery is a fundamental technique for text mining that identifies abstract topics within large document collections. A recent approach to topic discovery is to cluster document or sentence embeddings, typically obtained from pre-trained language models, and represent each cluster as a topic. Despite their strong empirical performance, the design principles linking the geometry of embedding spaces to human-interpretable topic organization remain unclear. Clarifying the relationship between these geometric structures and topic interpretability is therefore a key challenge in topic discovery. In this study, we propose TopiCLEAR (Topic discovery by CLustering Embeddings with Adaptive dimensionality Reduction), a simple framework that integrates document embeddings with iterative clustering based on adaptive dimensionality reduction. TopiCLEAR is guided by the hypothesis that human-interpretable topics correspond to low-dimensional geometric structures in embedding spaces and leverages adaptive dimensionality reduction to identify them. We evaluate topic quality using a document-level approach that combines quantitative evaluation based on human-labeled data with qualitative assessment in the absence of ground-truth labels. Experiments on four benchmark datasets, covering both formal and informal texts, show that TopiCLEAR consistently achieves strong agreement with human annotations, particularly for short and informal texts. Furthermore, a case study on Twitter data demonstrates that TopiCLEAR produces more interpretable topics than Latent Dirichlet Allocation (LDA), recovering both human-annotated topic structure and coherent sub-topic structure. These results highlight the effectiveness of clustering documents in a low-dimensional topic space for topic discovery.

Living in a changing climate, human society now faces more frequent and severe natural disasters than before. As a consequence, rapid disaster response during the “Golden 72 Hours” of search and rescue becomes a vital humanitarian necessity and community concern. However, traditional disaster damage analysis routinely fails to generalize across distinct urban morphologies and new disaster events. Effective damage mapping typically requires exhaustive and time-consuming manual data annotation. To address this issue, we introduce Smart Transfer, a novel Geospatial Artificial Intelligence (GeoAI) framework, leveraging state-of-the-art vision Foundation Models (FMs) for rapid building damage mapping with post-earthquake Very High Resolution (VHR) imagery. Specifically, we design two novel model transfer strategies: first, Pixel-wise Clustering (PC), ensuring robust prototype-level global feature alignment; second, a Distance-Penalized Triplet (DPT), integrating patch-level spatial autocorrelation patterns by assigning stronger penalties to semantically inconsistent yet spatially adjacent patches. Extensive experiments and ablations on the recent 2023 Türkiye-Syria earthquake show promising performance in multiple cross-region transfer settings, namely Leave One Domain Out (LODO) and Specific Source Domain Combination (SSDC), achieving an average F1-score of 0.64 and mIoU of 0.69. Moreover, Smart Transfer provides a scalable, automated GeoAI solution to accelerate building damage mapping and support rapid disaster response, offering new opportunities to enhance disaster resilience in climate-vulnerable regions and communities. The data and code are publicly available at https://github.com/ai4city-hkust/SmartTransfer.

A three-dimensional numerical model is developed to predict the electrochemical performance of polymer electrolyte fuel cell cathodes using catalyst-layer microstructures reconstructed from focused ion beam scanning electron microscopy images. Proton conduction, gas diffusion, and electrochemical reactions are solved using the lattice Boltzmann method (LBM). To reduce the computational cost while retaining local microstructural information, an anisotropic sub-grid scale (SGS) model is introduced. In this model, direction-dependent effective diffusivities and conductivities are evaluated from the reconstructed microstructures and assigned to each coarse computational cell, allowing the coarse-grid simulation to retain the local anisotropic transport characteristics of the catalyst layer. The model is validated under various operating conditions. The calculated cathode overpotentials agree well with the experimental results, with errors of less than 8%. The overpotential breakdown also shows physically reasonable trends. The validated model is then applied to cathodes with different ionomer distributions. The results show that increasing the ionomer content near the membrane side improves cathode performance. This trend is further confirmed by experiments using double-layer cathode electrodes. The proposed framework provides a practical method for connecting microstructural information with cathode performance prediction, which can support future catalyst layer design.