Selected work

Representative Papers

SampoNLP: A Self-Referential Toolkit for Morphological Analysis of Subword Tokenizers

Jan 08, 2026arXiv.org

This study addresses the challenge of evaluating subword tokenizers for morphologically rich languages like those in the Uralic family, where high-quality morpheme lexicons are scarce. To circumvent reliance on corpora, the authors propose a method grounded in the Minimum Description Length (MDL) principle, incorporating a self-referential atomicity scoring mechanism that leverages intraword structural cues to filter out compounds and construct high-purity morpheme lexicons. The work delivers the first empirically supported morpheme resources for Finnish, Hungarian, and Estonian, introduces an Integrated Performance Score (IPS) to balance coverage against over-segmentation, and employs elbow-point analysis to recommend optimal BPE vocabulary sizes. Findings highlight the limitations of standard BPE in highly agglutinative languages. The accompanying SampoNLP toolkit and resources are publicly released.

1 citationsRead paper

CAS: A Causal Attribution Score for Local and Global Explainable Artificial Intelligence

Aug 12, 2026

This study addresses the challenge that existing predictive explanation methods conflate predictive importance with drivers of heterogeneous causal effects, thereby failing to attribute outcomes to genuine intervention effects. To resolve this, the authors propose a causal Shapley attribution framework grounded in intervention coalition games, integrated with DoubleML causal estimation to construct a local-to-global Causal Attribution Score (CAS). This approach explicitly disentangles causal effect modification from predictive contribution without altering the Shapley value formulation, enabling interpretable analysis of heterogeneous treatment effects. Empirical results demonstrate strong performance: on synthetic data, the method achieves a mean absolute error as low as 0.107; on real-world datasets—401(k) and Pennsylvania reemployment—it identifies key causal effect modifiers via Feature-CAS that markedly differ from those highlighted by conventional SHAP, substantially outperforming current baselines.

0 citationsRead paper

IoXT: The Internet of Explainable Things. Why Explainability in IoT Requires a New System-Level Paradigm and Protocol Design

Aug 09, 2026

This work addresses the gap in existing explainable AI methods, which struggle to cover the end-to-end decision pipeline—from sensing to actuation—in Internet of Things (IoT) systems. The paper proposes a system-level explainability paradigm that embeds the “design-as-explanation” principle throughout the full-stack architecture of intelligent IoT, spanning perception, communication, intelligence, decision-making, and actuation. Key contributions include the formulation of cross-layer provenance, bidirectional traceability, and orphan-free execution as foundational principles; the introduction of a class of Explainable Telemetry Protocols (XTP); and the development of technical mechanisms—such as timestamped provenance graphs, cross-layer coverage analysis, streaming evidence processing, and protocol semantic continuity—to establish system conditions and a compliance framework that ensure persistent alignment between model explanations and sensor inputs, communication history, decisions, and actuation outcomes.

0 citationsRead paper

Improved Quantum Algorithms for Subset Sum and $k$-SUM

Aug 07, 2026

This work addresses the worst-case quantum complexity bottlenecks of Subset Sum and its parameterized variant, the k-SUM problem. The authors propose a novel quantum search framework that integrates quantum walks, block reduction, and a refined modular-7 classification scheme. By leveraging structured subproblem decomposition and an improved search strategy, they achieve the first faster worst-case quantum algorithm for instances where \(k \equiv 3\) or \(6 \pmod{7}\), attaining a time complexity of \(\tilde{O}(n^{\Psi_k})\). Extending this approach to the general Subset Sum problem yields a quantum algorithm with complexity \(O^*(2^{2n/7})\), surpassing the previous best-known bound of \(O^*(2^{n/3})\).

0 citationsRead paper

Cultural Awareness is Represented but Not Decoded: Tracing Mythological Knowledge across 18 Open-Source LLMs

Aug 03, 2026

This study addresses the challenge that open-source large language models struggle to accurately recognize and generate deity names from non-Western mythologies, such as Finnish, Slavic, Egyptian, or Chinese traditions. By employing linear probing, logit lens analysis, activation patching, and output extraction across 18 open-source models, the work systematically investigates how cross-cultural mythological knowledge is internally represented and decoded. The findings reveal that cultural bias predominantly arises during the decoding phase rather than in internal representations, and that the language of the prompt exerts a gating effect on decoding accuracy. Residual stream representations effectively distinguish entities across cultures, substantially outperforming baselines relying solely on name strings. The paper contributes the first entity-level probe–output decomposition framework and a benchmark for cross-cultural mythological entity annotation.

0 citationsRead paper

Recent publications

Latest Papers

CAS: A Causal Attribution Score for Local and Global Explainable Artificial Intelligence

Aug 12, 2026

This study addresses the challenge that existing predictive explanation methods conflate predictive importance with drivers of heterogeneous causal effects, thereby failing to attribute outcomes to genuine intervention effects. To resolve this, the authors propose a causal Shapley attribution framework grounded in intervention coalition games, integrated with DoubleML causal estimation to construct a local-to-global Causal Attribution Score (CAS). This approach explicitly disentangles causal effect modification from predictive contribution without altering the Shapley value formulation, enabling interpretable analysis of heterogeneous treatment effects. Empirical results demonstrate strong performance: on synthetic data, the method achieves a mean absolute error as low as 0.107; on real-world datasets—401(k) and Pennsylvania reemployment—it identifies key causal effect modifiers via Feature-CAS that markedly differ from those highlighted by conventional SHAP, substantially outperforming current baselines.

0 citationsRead paper

IoXT: The Internet of Explainable Things. Why Explainability in IoT Requires a New System-Level Paradigm and Protocol Design

Aug 09, 2026

This work addresses the gap in existing explainable AI methods, which struggle to cover the end-to-end decision pipeline—from sensing to actuation—in Internet of Things (IoT) systems. The paper proposes a system-level explainability paradigm that embeds the “design-as-explanation” principle throughout the full-stack architecture of intelligent IoT, spanning perception, communication, intelligence, decision-making, and actuation. Key contributions include the formulation of cross-layer provenance, bidirectional traceability, and orphan-free execution as foundational principles; the introduction of a class of Explainable Telemetry Protocols (XTP); and the development of technical mechanisms—such as timestamped provenance graphs, cross-layer coverage analysis, streaming evidence processing, and protocol semantic continuity—to establish system conditions and a compliance framework that ensure persistent alignment between model explanations and sensor inputs, communication history, decisions, and actuation outcomes.

0 citationsRead paper

Improved Quantum Algorithms for Subset Sum and $k$-SUM

Aug 07, 2026

This work addresses the worst-case quantum complexity bottlenecks of Subset Sum and its parameterized variant, the k-SUM problem. The authors propose a novel quantum search framework that integrates quantum walks, block reduction, and a refined modular-7 classification scheme. By leveraging structured subproblem decomposition and an improved search strategy, they achieve the first faster worst-case quantum algorithm for instances where \(k \equiv 3\) or \(6 \pmod{7}\), attaining a time complexity of \(\tilde{O}(n^{\Psi_k})\). Extending this approach to the general Subset Sum problem yields a quantum algorithm with complexity \(O^*(2^{2n/7})\), surpassing the previous best-known bound of \(O^*(2^{n/3})\).

0 citationsRead paper

Cultural Awareness is Represented but Not Decoded: Tracing Mythological Knowledge across 18 Open-Source LLMs

Aug 03, 2026

This study addresses the challenge that open-source large language models struggle to accurately recognize and generate deity names from non-Western mythologies, such as Finnish, Slavic, Egyptian, or Chinese traditions. By employing linear probing, logit lens analysis, activation patching, and output extraction across 18 open-source models, the work systematically investigates how cross-cultural mythological knowledge is internally represented and decoded. The findings reveal that cultural bias predominantly arises during the decoding phase rather than in internal representations, and that the language of the prompt exerts a gating effect on decoding accuracy. Residual stream representations effectively distinguish entities across cultures, substantially outperforming baselines relying solely on name strings. The paper contributes the first entity-level probe–output decomposition framework and a benchmark for cross-cultural mythological entity annotation.

0 citationsRead paper

Anatomy of a Sound Neural Reasoner: One-Shot Amortization, First-Pass Poisoning, and Search Inertness in Clue-Rich Completion

Jul 21, 2026

This work demonstrates that neural reasoners such as the Lattice Deduction Transformer, despite appearing to perform search and backtracking on constraint-rich tasks like Sudoku, actually rely solely on a single forward pass for prediction, with their apparent “search” serving only to reduce computational redundancy. The study introduces the concept of “first-pass poisoning,” showing that solution accuracy is governed by model calibration and symmetry handling. It further reveals that constraint-graph attention mechanisms substantially outperform positional encodings and proposes two effective interventions: symmetry-aware data augmentation and test-time ensembling. By integrating digit permutation augmentation, symmetric ensembling, and CoLT’s branching heuristics with shared unsatisfiable core techniques, the model achieves a dramatic improvement—from under 1% to 96.5 ± 0.3%—in accuracy on a 9×9 Sudoku symmetric holdout set, attaining 100% success on the most challenging instances.

0 citationsRead paper