Selected work

Representative 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

Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics

May 06, 2026

LLMs are trained once, then deployed into a world that never stops changing. External memory compensates for this, but most systems manage it explicitly rather than letting it adapt on its own. Biological memory works differently: coupled multi-timescale dynamics make new associations immediately usable, strengthen what repetition confirms, and let the rest fade. We argue that external memory should follow a similar principle. In Memini, this view takes the form of an associative memory that organizes knowledge as a directed graph. Each edge carries two coupled internal variables, one fast and one slow, following the Benna-Fusi model of synaptic consolidation. From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics.

0 citationsRead paper

Processing through encoding: Quantum circuit approaches for point-wise multiplication and convolution

Dec 12, 2025

This paper addresses the low efficiency of pointwise multiplication and convolution operations in quantum signal processing with complex-valued functions. To this end, it proposes a novel “Processing through Encoding” paradigm. Methodologically, complex functions are directly encoded into auxiliary qubit states, enabling pointwise products to emerge implicitly in the final-state amplitudes; combined with Fourier-basis encoding and the inverse quantum Fourier transform (IQFT), an end-to-end quantum convolution circuit is realized. Key contributions include: (i) the first quantum pointwise multiplication scheme that requires no explicit arithmetic gates; (ii) the first integrable and verifiable quantum convolution circuit; and (iii) a theoretically complete construction, validated via numerical simulation and modular implementation using the quantumaudio toolkit—accurately generating target products and convolution outputs. This work establishes a new pathway for quantum signal processing.

0 citationsRead paper

Quantum Brush: A quantum computing-based tool for digital painting

Sep 01, 2025

This work addresses the absence of quantum-native interaction paradigms in digital art creation by proposing the first quantum painting framework tailored for Noisy Intermediate-Scale Quantum (NISQ) devices. Methodologically, it introduces four types of quantum brushes that encode user strokes in real time into parameterized quantum circuits; quantum superposition and entanglement are leveraged to generate visual textures provably infeasible to simulate classically. The framework integrates quantum state encoding, measurement-induced wavefunction collapse, and classical rendering interfaces to enable hardware-level real-time interaction. Contributions include: (1) the first end-to-end quantum painting deployment on real quantum hardware (IQM Sirius); (2) an open-source toolchain empirically validated for robust artistic output on noisy, intermediate-scale devices; and (3) a demonstration that quantum phenomena—such as interference and quantum correlations—can be systematically harnessed as novel aesthetic primitives, thereby expanding the frontier of quantum computing applications in creative AI.

0 citationsRead paper

A Principled Framework for Multi-View Contrastive Learning

Jul 09, 2025

Existing multi-view contrastive learning suffers from four key limitations: conflicting optimization objectives, incomplete modeling of view-sample interactions, coupled alignment-uniformity trade-offs, and difficulty in harnessing diversity gains across views. This paper proposes MVCL—the first unified framework for multi-view contrastive learning—introducing two novel loss functions: MV-InfoNCE and MV-DHEL. MVCL enables joint modeling across *all* pairwise view combinations and theoretically guarantees asymptotic decoupling of alignment and uniformity optimization. It natively supports extension to two or more modalities. Grounded in information theory, MVCL integrates multi-view augmentation with a decoupled optimization mechanism. Evaluated on four benchmarks including ImageNet1K, MVCL significantly outperforms state-of-the-art methods. Empirical results demonstrate consistent performance gains with increasing numbers of views, and effective mitigation of dimensional collapse beyond five views.

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

Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics

May 06, 2026

LLMs are trained once, then deployed into a world that never stops changing. External memory compensates for this, but most systems manage it explicitly rather than letting it adapt on its own. Biological memory works differently: coupled multi-timescale dynamics make new associations immediately usable, strengthen what repetition confirms, and let the rest fade. We argue that external memory should follow a similar principle. In Memini, this view takes the form of an associative memory that organizes knowledge as a directed graph. Each edge carries two coupled internal variables, one fast and one slow, following the Benna-Fusi model of synaptic consolidation. From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics.

0 citationsRead paper

Processing through encoding: Quantum circuit approaches for point-wise multiplication and convolution

Dec 12, 2025

This paper addresses the low efficiency of pointwise multiplication and convolution operations in quantum signal processing with complex-valued functions. To this end, it proposes a novel “Processing through Encoding” paradigm. Methodologically, complex functions are directly encoded into auxiliary qubit states, enabling pointwise products to emerge implicitly in the final-state amplitudes; combined with Fourier-basis encoding and the inverse quantum Fourier transform (IQFT), an end-to-end quantum convolution circuit is realized. Key contributions include: (i) the first quantum pointwise multiplication scheme that requires no explicit arithmetic gates; (ii) the first integrable and verifiable quantum convolution circuit; and (iii) a theoretically complete construction, validated via numerical simulation and modular implementation using the quantumaudio toolkit—accurately generating target products and convolution outputs. This work establishes a new pathway for quantum signal processing.

0 citationsRead paper

Quantum Brush: A quantum computing-based tool for digital painting

Sep 01, 2025

This work addresses the absence of quantum-native interaction paradigms in digital art creation by proposing the first quantum painting framework tailored for Noisy Intermediate-Scale Quantum (NISQ) devices. Methodologically, it introduces four types of quantum brushes that encode user strokes in real time into parameterized quantum circuits; quantum superposition and entanglement are leveraged to generate visual textures provably infeasible to simulate classically. The framework integrates quantum state encoding, measurement-induced wavefunction collapse, and classical rendering interfaces to enable hardware-level real-time interaction. Contributions include: (1) the first end-to-end quantum painting deployment on real quantum hardware (IQM Sirius); (2) an open-source toolchain empirically validated for robust artistic output on noisy, intermediate-scale devices; and (3) a demonstration that quantum phenomena—such as interference and quantum correlations—can be systematically harnessed as novel aesthetic primitives, thereby expanding the frontier of quantum computing applications in creative AI.

0 citationsRead paper

A Principled Framework for Multi-View Contrastive Learning

Jul 09, 2025

Existing multi-view contrastive learning suffers from four key limitations: conflicting optimization objectives, incomplete modeling of view-sample interactions, coupled alignment-uniformity trade-offs, and difficulty in harnessing diversity gains across views. This paper proposes MVCL—the first unified framework for multi-view contrastive learning—introducing two novel loss functions: MV-InfoNCE and MV-DHEL. MVCL enables joint modeling across *all* pairwise view combinations and theoretically guarantees asymptotic decoupling of alignment and uniformity optimization. It natively supports extension to two or more modalities. Grounded in information theory, MVCL integrates multi-view augmentation with a decoupled optimization mechanism. Evaluated on four benchmarks including ImageNet1K, MVCL significantly outperforms state-of-the-art methods. Empirical results demonstrate consistent performance gains with increasing numbers of views, and effective mitigation of dimensional collapse beyond five views.

0 citationsRead paper