arXiv subjects

Ryo Sakamoto

Publications and source records attributed to Ryo Sakamoto.

5 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↗

arXiv2023

Boosting Radiology Report Generation by Infusing Comparison Prior

Recent transformer-based models have made significant strides in generating radiology reports from chest X-ray images. However, a prominent challenge remains: these models often lack prior knowledge, resulting in the generation of synthetic reports that mistakenly reference non-existent prior exams. This discrepancy can be attributed to a knowledge gap between radiologists and the generation models. While radiologists possess patient-specific prior information, the models solely receive X-ray images at a specific time point. To tackle this issue, we propose a novel approach that leverages a rule-based labeler to extract comparison prior information from radiology reports. This extracted comparison prior is then seamlessly integrated into state-of-the-art transformer-based models, enabling them to produce more realistic and comprehensive reports. Our method is evaluated on English report datasets, such as IU X-ray and MIMIC-CXR. The results demonstrate that our approach surpasses baseline models in terms of natural language generation metrics. Notably, our model generates reports that are free from false references to non-existent prior exams, setting it apart from previous models. By addressing this limitation, our approach represents a significant step towards bridging the gap between radiologists and generation models in the domain of medical report generation.

cs.CL↗

arXiv2019

Implementation and Performance of Barnes-Hut N-body algorithm on Extreme-scale Heterogeneous Many-core Architectures

In this paper, we report the implementation and measured performance of our extreme-scale global simulation code on Sunway TaihuLight and two PEZY-SC2 systems: Shoubu System B and Gyoukou. The numerical algorithm is the parallel Barnes-Hut tree algorithm, which has been used in many large-scale astrophysical particle-based simulations. Our implementation is based on our FDPS framework. However, the extremely large numbers of cores of the systems used (10M on TaihuLight and 16M on Gyoukou) and their relatively poor memory and network bandwidth pose new challenges. We describe the new algorithms introduced to achieve high efficiency on machines with low memory bandwidth. The measured performance is 47.9, 10.6 PF, and 1.01PF on TaihuLight, Gyoukou and Shoubu System B (efficiency 40\%, 23.5\% and 35.5\%). The current code is developed for the simulation of planetary rings, but most of the new algorithms are useful for other simulations, and are now available in the FDPS framework.

astro-ph.IM↗

arXiv2016

Implementation and evaluation of data-compression algorithms for irregular-grid iterative methods on the PEZY-SC processor

Iterative methods on irregular grids have been used widely in all areas of comptational science and engineering for solving partial differential equations with complex geometry. They provide the flexibility to express complex shapes with relatively low computational cost. However, the direction of the evolution of high-performance processors in the last two decades have caused serious degradation of the computational efficiency of iterative methods on irregular grids, because of relatively low memory bandwidth. Data compression can in principle reduce the necessary memory memory bandwidth of iterative methods and thus improve the efficiency. We have implemented several data compression algorithms on the PEZY-SC processor, using the matrix generated for the HPCG benchmark as an example. For the SpMV (Sparse Matrix-Vector multiplication) part of the HPCG benchmark, the best implementation without data compression achieved 11.6Gflops/chip, close to the theoretical limit due to the memory bandwidth. Our implementation with data compression has achieved 32.4Gflops. This is of course rather extreme case, since the grid used in HPCG is geometrically regular and thus its compression efficiency is very high. However, in real applications, it is in many cases possible to make a large part of the grid to have regular geometry, in particular when the resolution is high. Note that we do not need to change the structure of the program, except for the addition of the data compression/decompression subroutines. Thus, we believe the data compression will be very useful way to improve the performance of many applications which rely on the use of irregular grids.

cs.MS↗