arXiv subjects

Shoji Nishimura

Publications and source records attributed to Shoji Nishimura.

5 recordsLinked to original sources

arXiv2026

HyEm: Query-Adaptive Hyperbolic Retrieval for Biomedical Ontologies via Euclidean Vector Indexing

Retrieval-augmented generation (RAG) for biomedical knowledge faces a hierarchy-aware ontology grounding challenge: resources like HPO, DO, and MeSH use deep ``is-a" taxonomies, yet production stacks rely on Euclidean embeddings and ANN indexes. While hyperbolic embeddings suit hierarchical representation, they face two barriers: (i) lack of native vector database support, and (ii) risk of underperforming on entity-centric queries where hierarchy is irrelevant. We present HyEm, a lightweight retrieval layer integrating hyperbolic ontology embeddings into existing Euclidean ANN infrastructure. HyEm learns radius-controlled hyperbolic embeddings, stores origin log-mapped vectors in standard Euclidean databases for candidate retrieval, then applies exact hyperbolic reranking. A query-adaptive gate outputs continuous mixing weights, combining Euclidean semantic similarity with hyperbolic hierarchy distance at reranking time. Our bi-Lipschitz analysis under radius constraints provides practical guidance for ANN oversampling and dimensionality.Experiments on biomedical ontology subsets demonstrate HyEm preserves 94-98% of Euclidean baseline performance on entity-centric queries while substantially improving hierarchy-navigation and mixed-intent queries, maintaining indexability at moderate oversampling.

cs.IR↗

arXiv2026

Statistical-Neural Interaction Networks for Interpretable Mixed-Type Data Imputation

Real-world tabular databases routinely combine continuous measurements and categorical records, yet missing entries are pervasive and can distort downstream analysis. We propose Statistical-Neural Interaction (SNI), an interpretable mixed-type imputation framework that couples correlation-derived statistical priors with neural feature attention through a Controllable-Prior Feature Attention (CPFA) module. CPFA learns head-wise prior-strength coefficients $\{\lambda_h\}$ that softly regularize attention toward the prior while allowing data-driven deviations when nonlinear patterns appear to be present in the data. Beyond imputation, SNI aggregates attention maps into a directed feature-dependency matrix that summarizes which variables the imputer relied on, without requiring post-hoc explainers. We evaluate SNI against six baselines (Mean/Mode, MICE, KNN, MissForest, GAIN, MIWAE) on six datasets spanning ICU monitoring, population surveys, socio-economic statistics, and engineering applications. Under MCAR/strict-MAR at 30\% missingness, SNI is generally competitive on continuous metrics but is often outperformed by accuracy-first baselines (MissForest, MIWAE) on categorical variables; in return, it provides intrinsic dependency diagnostics and explicit statistical-neural trade-off parameters. We additionally report MNAR stress tests (with a mask-aware variant) and discuss computational cost, limitations -- particularly for severely imbalanced categorical targets -- and deployment scenarios where interpretability may justify the trade-off.

cs.LG↗

arXiv2025

Auditable Unit-Aware Thresholds in Symbolic Regression via Logistic-Gated Operators

AI for health will only scale when models are not only accurate but also readable, auditable, and governable. Many clinical and public-health decisions hinge on numeric thresholds -- cut-points that trigger alarms, treatment, or follow-up -- yet most machine-learning systems bury those thresholds inside opaque scores or smooth response curves. We introduce logistic-gated operators (LGO) for symbolic regression, which promote thresholds to first-class, unit-aware parameters inside equations and map them back to physical units for direct comparison with guidelines. On public ICU and population-health cohorts (MIMIC-IV ICU, eICU, NHANES), LGO recovers clinically plausible gates on MAP, lactate, GCS, SpO2, BMI, fasting glucose, and waist circumference while remaining competitive with established scoring systems (AutoScore) and explainable boosting machines (EBM). The gates are sparse and selective: they appear when regime switching is supported by the data and are pruned on predominantly smooth tasks, yielding compact formulas that clinicians can inspect, stress-test, and revise. As a standalone symbolic model or a safety overlay on black-box systems, LGO helps translate observational data into auditable, unit-aware rules for medicine and other threshold-driven domains.

cs.LG↗

arXiv2024

Evolutionary Causal Discovery with Relative Impact Stratification for Interpretable Data Analysis

This study proposes Evolutionary Causal Discovery (ECD) for causal discovery that tailors response variables, predictor variables, and corresponding operators to research datasets. Utilizing genetic programming for variable relationship parsing, the method proceeds with the Relative Impact Stratification (RIS) algorithm to assess the relative impact of predictor variables on the response variable, facilitating expression simplification and enhancing the interpretability of variable relationships. ECD proposes an expression tree to visualize the RIS results, offering a differentiated depiction of unknown causal relationships compared to conventional causal discovery. The ECD method represents an evolution and augmentation of existing causal discovery methods, providing an interpretable approach for analyzing variable relationships in complex systems, particularly in healthcare settings with Electronic Health Record (EHR) data. Experiments on both synthetic and real-world EHR datasets demonstrate the efficacy of ECD in uncovering patterns and mechanisms among variables, maintaining high accuracy and stability across different noise levels. On the real-world EHR dataset, ECD reveals the intricate relationships between the response variable and other predictive variables, aligning with the results of structural equation modeling and shapley additive explanations analyses.

cs.LG↗

arXiv2020

Weakly-Supervised Multi-Person Action Recognition in 360$^{\circ}$ Videos

The recent development of commodity 360$^{\circ}$ cameras have enabled a single video to capture an entire scene, which endows promising potentials in surveillance scenarios. However, research in omnidirectional video analysis has lagged behind the hardware advances. In this work, we address the important problem of action recognition in top-view 360$^{\circ}$ videos. Due to the wide filed-of-view, 360$^{\circ}$ videos usually capture multiple people performing actions at the same time. Furthermore, the appearance of people are deformed. The proposed framework first transforms omnidirectional videos into panoramic videos, then it extracts spatial-temporal features using region-based 3D CNNs for action recognition. We propose a weakly-supervised method based on multi-instance multi-label learning, which trains the model to recognize and localize multiple actions in a video using only video-level action labels as supervision. We perform experiments to quantitatively validate the efficacy of the proposed method and qualitatively demonstrate action localization results. To enable research in this direction, we introduce 360Action, the first omnidirectional video dataset for multi-person action recognition.

cs.CV↗