arXiv subjects

Yuki Ito

Publications and source records attributed to Yuki Ito.

15 recordsLinked to original sources

arXiv2026

KBBQ: A Predictive Noise Law and the Limits of Spectrum Flattening in FP4 Quantization

We develop a second-order theory of quantization noise in matrix multiplication in which the quantization format is characterized by the variance it assigns to each element. The constant variance profile of integer quantization recovers existing integer-noise theory, while the multiplicative profile of floating-point rounding reduces the data dependence to a scalar, the participation factor $\kappa$, yielding a closed-form signal-to-noise-ratio law. The resulting functional also admits a closed-form upper bound $\kappa^{*}$ that no function-preserving linear transform can exceed and that is attained by a recent state-of-the-art method. Building on this analysis, we introduce KBBQ (\textbf{K}appa-\textbf{B}raked \textbf{B}lockwise \textbf{Q}uantization), which parameterizes the extent to which a transform approaches this ceiling. At W4A4, across four base models and two FP4 formats, KBBQ outperforms the prior state of the art without additional deployment-time computation.

cs.LG↗

arXiv2026

Teaching Diffusion to Speculate Left-to-Right

Large language models (LLMs) achieve remarkable performance across a wide range of tasks, but their autoregressive decoding process incurs substantial inference costs due to inherently sequential token generation. Speculative decoding addresses this bottleneck by employing a lightweight draft model to propose multiple future tokens that are subsequently verified in parallel by a larger target model. Recent work has demonstrated that diffusion language models are well suited for this setting, as they can generate entire blocks of draft tokens in parallel and thereby alleviate the sequential constraints of autoregressive drafting. A subtlety of this regime is that block-diffusion drafters generate tokens bidirectionally within a block, whereas verification is performed by an autoregressive target model that evaluates tokens in a strictly left-to-right manner, leaving a gap between the symmetric training-time objective and the asymmetric verification-time reward. In this work, we offer an empirical analysis of three training-time interventions that narrow this gap: token positional weighting, a first-error focal loss that targets the position that breaks the accepted prefix within each block, and a chain loss term that substitutes a differentiable surrogate for the expected accepted length. The three interventions act along orthogonal axes (position, block-conditional first error, joint prefix) and compose additively; they are likewise orthogonal to test-time alignment mechanisms such as multi-draft self-selection, with which they can in principle be combined. Across four target models and six reasoning, code, and dialogue benchmarks, the three interventions raise accepted draft length by 21-76% per benchmark over a position-uniform baseline, without adding additional forward passes and without changing the inference pipeline or the rejection-sampling exactness contract.

cs.CL↗

arXiv2026

Quantum secure blind decryption with two users

We propose two types of protocols for quantum secure blind decryption, involving two users and servers. User 1 holds the encrypted ciphertext. The servers store several indexed keys including the key encrypting the ciphertext. User 2 aims to obtain the decrypted text. The protocols are designed to preserve the following types of secrecy: Users ensure the secrecy of the text from the servers. Servers maintain the secrecy of the keys from the users. Our protocols enable User 2 to obtain the decrypted text while preserving these secrecy requirements. Additionally, the second protocol ensures the secrecy of the key index to identify the key encrypting the ciphertext from the servers, and the second protocol requires two non-commuting servers. Furthermore, we analyze the secrecy of the second protocol under post-attack scenarios, where the two servers communicates with each other after the completion of the protocol. We show that our quantum protocol satisfies the secrecy under these attacks, whereas its classical counterpart fails to do so.

quant-ph↗

arXiv2026

Polynomial time constructive decision algorithm for multivariable quantum signal processing

Quantum signal processing (QSP) and quantum singular value transformation (QSVT) have provided a unified framework for understanding many quantum algorithms, including factorization, matrix inversion, and Hamiltonian simulation. As a multivariable version of QSP, multivariable quantum signal processing (M-QSP) is proposed. M-QSP interleaves signal operators corresponding to each variable with signal processing operators, which provides an efficient means to perform multivariable polynomial transformations. However, the necessary and sufficient condition for what types of polynomials can be constructed by M-QSP is unknown. In this paper, we propose a classical algorithm to determine whether a given pair of multivariable Laurent polynomials can be implemented by M-QSP, which returns True or False. As one of the most important properties of this algorithm, its returning True is the necessary and sufficient condition. The proposed classical algorithm runs in polynomial time in the number of variables and signal operators. Our algorithm also provides a constructive method to select the necessary parameters for implementing M-QSP. These findings offer valuable insights for identifying practical applications of M-QSP.

quant-ph↗

arXiv2025

A Quantum Algorithm for Nonlinear Electromagnetic Fluid Dynamics via Koopman-von Neumann Linearization

To simulate plasma phenomena, large-scale computational resources have been employed in developing high-precision and high-resolution plasma simulations. One of the main obstacles in plasma simulations is the requirement of computational resources that scale polynomially with the number of spatial grids, which poses a significant challenge for large-scale modeling. To address this issue, this study presents a quantum algorithm for simulating the nonlinear electromagnetic fluid dynamics that govern space plasmas. We map it, by applying Koopman-von Neumann linearization, to the Schrödinger equation and evolve the system using Hamiltonian simulation via quantum singular value transformation. Our algorithm scales $O \left(s N_x \, \mathrm{polylog} \left( N_x \right) T \right)$ in time complexity with $s$, $N_x$, and $T$ being the spatial dimension, the number of spatial grid points per dimension, and the evolution time, respectively. Comparing the scaling $O \left( s N_x^s \left(T^{5/4}+T N_x\right) \right)$ for the classical method with the finite volume scheme, this algorithm achieves polynomial speedup in $N_x$. The space complexity of this algorithm is exponentially reduced from $O\left( s N_x^s \right)$ to $O\left( s \, \mathrm{polylog} \left( N_x \right) \right)$. Numerical experiments validate that accurate solutions are attainable with smaller $m$ than theoretically anticipated and with practical values of $m$ and $R$, underscoring the feasibility of the approach. As a practical demonstration, the method accurately reproduces the Kelvin-Helmholtz instability, underscoring its capability to tackle more intricate nonlinear dynamics. These results suggest that quantum computing can offer a viable pathway to overcome the computational barriers of multiscale plasma modeling.

quant-ph↗

arXiv2025

Identification and Estimation of Dynamic Games with Unknown Information Structure

We develop an empirical framework for analyzing dynamic games when the underlying information structure is unknown to the analyst. We introduce \textit{Markov correlated equilibrium}, a dynamic analog of Bayes correlated equilibrium, and show that its predictions coincide with the Markov perfect equilibrium predictions attainable when players observe richer signals than the analyst assumes. We provide tractable methods for informationally robust estimation, inference, and counterfactual analysis. We illustrate the framework with a dynamic entry game between Starbucks and Dunkin' in the US and study the role of informational assumptions.

econ.EM↗

arXiv2025

Information-Theoretic Bayesian Optimization for Bilevel Optimization Problems

A bilevel optimization problem consists of two optimization problems nested as an upper- and a lower-level problem, in which the optimality of the lower-level problem defines a constraint for the upper-level problem. This paper considers Bayesian optimization (BO) for the case that both the upper- and lower-levels involve expensive black-box functions. Because of its nested structure, bilevel optimization has a complex problem definition, by which bilevel BO has not been widely studied compared with other standard extensions of BO such as multi-objective or constraint problems. We propose an information-theoretic approach that considers the information gain of both the upper- and lower-optimal solutions and values. This enables us to define a unified criterion that measures the benefit for both level problems, simultaneously. Further, we also show a practical lower bound based approach to evaluating the information gain. We empirically demonstrate the effectiveness of our proposed method through several benchmark datasets.

cs.LG↗

arXiv2025

LibriTTS-VI: A Public Corpus and Novel Methods for Efficient Voice Impression Control

Numerical voice impression (VI) control (e.g., scaling brightness) enables fine-grained control in text-to-speech (TTS). However, it faces two challenges: no public corpus and impression leakage, where reference audio biases synthesized voice away from the target VI. To address the first challenge, we introduce LibriTTS-VI, the first public VI corpus built on LibriTTS-R. For the second, we hypothesize a single reference causes leakage by entangling speaker identity and VI. To mitigate this, we propose 1) disentangled training with two utterances from the same speaker for speaker and VI conditioning, and 2) a reference-free method controlling the impression solely via target VI. Experimentally, our best method improves controllability: 11-dimensional VI mean squared error drops from 0.61 to 0.41 objectively and 1.15 to 0.92 subjectively. A comparison with a prompt-based TTS reveals imprecise numerical control and entanglement between VI and text semantics, which our methods overcome.

cs.SD↗

arXiv2025

Scheduled Interleaved Speech-Text Training for Speech-to-Speech Translation with LLMs

Speech-to-speech translation (S2ST) has been advanced with large language models (LLMs), which are fine-tuned on discrete speech units. In such approaches, modality adaptation from text to speech has been an issue. LLMs are trained on text-only data, which presents challenges to adapt them to speech modality with limited speech-to-speech data. To address the training difficulty, we propose scheduled interleaved speech--text training in this study. We use interleaved speech--text units instead of speech units during training, where aligned text tokens are interleaved at the word level. We gradually decrease the ratio of text as training progresses, to facilitate progressive modality adaptation from text to speech. We conduct experimental evaluations by fine-tuning LLaMA3.2-1B for S2ST on the CVSS dataset. We show that the proposed method consistently improves the translation performances, especially for languages with limited training data.

cs.CL↗

arXiv2025

Ensemble ToT of LLMs and Its Application to Automatic Grading System for Supporting Self-Learning

Providing students with detailed and timely grading feedback is essential for self-learning. While existing LLM-based grading systems are promising, most of them rely on one single model, which limits their performance. To address this, we propose Ensemble Tree-of-Thought (ToT), a framework that enhances LLM outputs by integrating multiple models. Using this framework, we develop a grading system. Ensemble ToT follows three steps: (1) analyzing LLM performance, (2) generating candidate answers, and (3) refining them into a final result. Based on this, our grading system first evaluates the grading tendencies of LLMs, then generates multiple results, and finally integrates them via a simulated debate. Experimental results demonstrate our approach's ability to provide accurate and explainable grading by effectively coordinating multiple LLMs.

cs.IR↗

arXiv2025

Entanglement measures for detectability

We propose new entanglement measures as the detection performance based on the hypothesis testing setting. We clarify how our measures work for detecting an entangled state by extending the quantum Sanov theorem. Our analysis covers the finite-length setting. Exploiting this entanglement measure, we present how to derive an entanglement witness to detect the given entangled state by using the geometrical structure of this measure. We derive their calculation formulas for maximally correlated states, and propose their algorithms that work for general entangled states. In addition, we investigate how our algorithm works for solving the membership problem for separability. Further, employing this algorithm, we propose a method to find entanglement witness for a given entangled state.

quant-ph↗

arXiv2024

Combination of crystal growth with optical floating zone and evaluation of Nd3+:LaAlO3 crystals with the dynamic nuclear polarization of 139La and 27Al

Producing a polarized lanthanum (La) target with high polarization and long relaxation time is crucial for realizing time-reversal violation experiments using polarized neutron beams. We use a LaAlO3 crystal doped with a small amount of Nd3+ ions for the polarized lanthanum target. Optimizing the amount of Nd3+ ions is considerably important because the achievable polarization and relaxation time strongly depend on this amount. We established a fundamental method to grow single crystals of Nd3+:LaAlO3 using an optical floating zone method that employs halogen lamps and evaluated the crystals with the dynamic nuclear polarization (DNP) method for polarizing nuclear spins. Two crystal samples were grown by ourselves and evaluated with the DNP at 1.3 K and 2.3 T for the first time except for the target materials of protons. The enhancement of NMR signals for 139La and 27Al was successfully observed, and the enhancement factors were eventually 3.5+-0.3 and 13+-3 for the samples with Nd3+ ions of 0.05 and 0.01 mol%, respectively. These enhancement factors correspond to absolute vector polarizations of 0.27+-0.02% (Nd 0.05 mol%) and 1.4+-0.3% (Nd 0.01 mol%). Although the obtained polarizations are still low, they are acceptable as a first step. The combination scheme of the crystal growth and evaluation of the crystals is found to be effectively applicable for optimizing the amount of Nd3+ ions for improving the performance of the polarized target.

physics.ins-det↗

arXiv2023

How to Map Linear Differential Equations to Schrödinger Equations via Carleman and Koopman-von Neumann Embeddings for Quantum Algorithms

Solving linear and nonlinear differential equations with large degrees of freedom is an important task for scientific and industrial applications. In order to solve such differential equations on a quantum computer, it is necessary to embed classical variables into a quantum state. While the Carleman and Koopman-von Neumann embeddings have been investigated so far, the class of problems that can be mapped to the Schrödinger equation is not well understood even for linear differential equations. In this work, we investigate the conditions for linear differential equations to be mapped to the Schrödinger equation and solved on a quantum computer. Interestingly, we find that these conditions are identical for both Carleman and Koopman-von Neumann embeddings. We also compute the computational complexity associated with estimating the expected values of an observable. This is done by assuming a state preparation oracle, block encoding of the mapped Hamiltonian via either Carleman or Koopman-von Neumann embedding, and block encoding of the observable using $O(\log M)$ qubits with $M$ is the mapped system size. Furthermore, we consider a general classical quadratic Hamiltonian dynamics and find a sufficient condition to map it into the Schrödinger equation. As a special case, this includes the coupled harmonic oscillator model [Babbush et al., \cite{babbush_exponential_2023}]. We also find a concrete example that cannot be described as the coupled harmonic oscillator but can be mapped to the Schrödinger equation in our framework. These results are important in the construction of quantum algorithms for solving differential equations of large-degree-of-freedom.

quant-ph↗

arXiv2022

Head-Related Transfer Function Interpolation from Spatially Sparse Measurements Using Autoencoder with Source Position Conditioning

We propose a method of head-related transfer function (HRTF) interpolation from sparsely measured HRTFs using an autoencoder with source position conditioning. The proposed method is drawn from an analogy between an HRTF interpolation method based on regularized linear regression (RLR) and an autoencoder. Through this analogy, we found the key feature of the RLR-based method that HRTFs are decomposed into source-position-dependent and source-position-independent factors. On the basis of this finding, we design the encoder and decoder so that their weights and biases are generated from source positions. Furthermore, we introduce an aggregation module that reduces the dependence of latent variables on source position for obtaining a source-position-independent representation of each subject. Numerical experiments show that the proposed method can work well for unseen subjects and achieve an interpolation performance with only one-eighth measurements comparable to that of the RLR-based method.

cs.SD↗

arXiv2019

Profiling based Out-of-core Hybrid Method for Large Neural Networks

GPUs are widely used to accelerate deep learning with NNs (NNs). On the other hand, since GPU memory capacity is limited, it is difficult to implement efficient programs that compute large NNs on GPU. To compute NNs exceeding GPU memory capacity, data-swapping method and recomputing method have been proposed in existing work. However, in these methods, performance overhead occurs due to data movement or increase of computation. In order to reduce the overhead, it is important to consider characteristics of each layer such as sizes and cost for recomputation. Based on this direction, we proposed Profiling based out-of-core Hybrid method (PoocH). PoocH determines target layers of swapping or recomputing based on runtime profiling. We implemented PoocH by extending a deep learning framework, Chainer, and we evaluated its performance. With PoocH, we successfully computed an NN requiring 50 GB memory on a single GPU with 16 GB memory. Compared with in-core cases, performance degradation was 38 \% on x86 machine and 28 \% on POWER9 machine.

cs.LG↗