arXiv subjects

Kenichi Kobayashi

Publications and source records attributed to Kenichi Kobayashi.

9 recordsLinked to original sources

arXiv2026

Kozuchi Agent: A Language-Agnostic Open-Weight Agent for Software Repair

Industrial software-engineering teams increasingly need LLM agents that turn bug reports into correct patches, yet benchmark-scale operation adds long horizons, tool-use discipline, context persistence, heterogeneous clusters, and evaluation reuse. We present Kozuchi Agent, a language-agnostic open-weight repair agent and CI-operated evaluation pipeline. Explicit phases, persistent state, deterministic tools, a model-independent action interface, and cross-agent test-time selection make runs auditable and repeatable. With locally hosted Qwen3.5-27B, no fine-tuning, and TTS@8, Kozuchi resolves 374/500 SWE-bench Verified instances on the official evaluator. Unchanged on Multi-SWE-bench Java, the same 27-billion-parameter agent resolves 41/128 instances (32.03%), ranking first among strict open-weight submissions and fourth of 42 overall; on Python it ranks 12th of 135 and first among open-weight systems. Per-phase behavior remains within +/-5 percentage points across languages. Remaining failures mainly reflect semantic correctness, Java-specific harness issues, and selection errors. Across both tracks, results compare favorably with open/local peers by parameter count. Analysis of candidate diversity, selector regret, and patch reliability shows that the remaining gap is primarily semantic correctness and selection rather than edit formatting or proprietary-model access. Operationally, reusable CI stages reduce operator touch-points from five to one across heterogeneous internal clusters.

cs.SE↗

arXiv2026

Adaptive Targeted Dynamic Chunking for Tokenization-Free Hierarchical Model

Tokenization-free hierarchical models are emerging as a promising alternative to traditional Large Language Models (LLMs), addressing inherent preprocessing issues such as vocabulary design complexity, out-of-vocabulary (OOV) errors, and language-specific constraints. However, a significant challenge in these byte-level methods is the optimization of the compression ratio, a critical factor that dictates model performance for processing bytes data via chunks. In this paper, we propose Adaptive Targeted Dynamic Chunking (ATDC), a novel byte-compression control mechanism designed to enhance the effectiveness of dynamic chunking within hierarchical architectures. Our approach utilizes curriculum learning to progressively adjust the compression ratio during training, transitioning from low to high compression to stabilize the learning process. We provide an analysis establishing the relationship between the target compression ratio and Bytes-Per-Innermost-Chunk (BPIC), allowing for tracking of chunk-size evolution throughout the training phase. Evaluations conducted on the FineWeb-Edu 100B dataset demonstrate that hierarchical models equipped with ATDC achieve competitive Bits-Per-Byte (BPB) performance compared to conventional baselines operating at both byte and token levels. Furthermore, the proposed method exhibits more stable training dynamics and superior final performance across diverse downstream tasks compared to models using fixed compression ratios, while maintaining the inherent robustness and flexibility of byte-level processing.

cs.CL↗

arXiv2025

Direct Quantized Training of Language Models with Stochastic Rounding

Although recent quantized Large Language Models (LLMs), such as BitNet, have paved the way for significant reduction in memory usage during deployment with binary or ternary weights, training these models still demands substantial memory footprints. This is partly because high-precision (i.e., unquantized) weights required for straight-through estimation must be maintained throughout the whole training process. To address this, we explore directly updating the quantized low-precision weights without relying on straight-through estimation during backpropagation, aiming to save memory usage during training. Specifically, we employ a stochastic rounding technique to minimize the information loss caused by the use of low-bit weights throughout training. Experimental results on our LLaMA-structured models of various sizes indicate that (1) training with only low-precision weights is feasible even when they are constrained to ternary values; (2) extending the bit width to 8 bits achieves performance on par with BitNet b1.58; (3) our models remain robust to precision scaling and memory reduction, showing minimal performance degradation when moving from FP32 to lower-memory environments (BF16/FP8); and (4) our models also support inference using ternary weights, showcasing their flexibility in deployment.

cs.LG↗

arXiv2021

Neural Network Module Decomposition and Recomposition

We propose a modularization method that decomposes a deep neural network (DNN) into small modules from a functionality perspective and recomposes them into a new model for some other task. Decomposed modules are expected to have the advantages of interpretability and verifiability due to their small size. In contrast to existing studies based on reusing models that involve retraining, such as a transfer learning model, the proposed method does not require retraining and has wide applicability as it can be easily combined with existing functional modules. The proposed method extracts modules using weight masks and can be applied to arbitrary DNNs. Unlike existing studies, it requires no assumption about the network architecture. To extract modules, we designed a learning method and a loss function to maximize shared weights among modules. As a result, the extracted modules can be recomposed without a large increase in the size. We demonstrate that the proposed method can decompose and recompose DNNs with high compression ratio and high accuracy and is superior to the existing method through sharing weights between modules.

cs.LG↗

arXiv2019

A New Fast Weighted All-pairs Shortest Path Search Algorithm Based on Pruning by Shortest Path Trees

Recently we submitted a paper, whose title is A New Fast Unweighted All-pairs Shortest Path Search Algorithm Based on Pruning by Shortest Path Trees, to arXiv. This is related to unweighted graphs. This paper also presents a new fast all-pairs shortest path algorithm for weighted graph based on the same idea. In Dijkstra algorithm which is said to be fast in weighted graphs, the average number of accesses to adjacent vertices (expressed by α) is about equal to the average degree of the graph. On the other hand, our algorithm utilizes the shortest path trees of adjacent vertices of each source vertex in the same manner as the algorithm for unweighted graphs, and reduce α drastically in comparison with Dijkstra algorithm. Roughly speaking α is reduced to the value close to 1, because the average degree of a tree is about 2, and one is used to come in and the other is used to go out, although that does not hold true when the depth of the short path trees is small. In case of weighted graphs, a problem which does not occur in unweighted graphs occurs. It is waiting for the generation of the shortest path tree of an adjacent vertex. Therefore, it is possible that a deadlock occurs. We prove our algorithm is deadlock-free. We compared our algorithm with Dijkstra and Peng algorithms. On Dijkstra algorithm ours outperforms it on speed and α except that Dijkstra algorithm slightly outperforms ours or they are almost the same on CPU time in sparse scale-free graphs. The result on Peng algorithm is as follows: In speed and α, ours outperforms Peng algorithm in hypercube-shaped and dense scale-free graphs, but conversely Peng algorithm outperforms ours in sparse scale-free graphs.

cs.DS↗

arXiv2019

A New Fast Unweighted All-pairs Shortest Path Search Algorithm Based on Pruning by Shortest Path Trees

We present a new fast all-pairs shortest path algorithm for unweighted graphs. In breadth-first search which is said to representative and fast in unweighted graphs, the average number of accesses to adjacent vertices (expressed by α) is about equal to the average degree of the graph. On the other hand, our algorithm utilizes the shortest path trees of adjacent vertices of each source vertex, and reduce α drastically. Roughly speaking α is reduced to the value close to 1, because the average degree of a tree is about 2, and one is used to come in and the other is used to go out, although that does not hold true when the depth of the shortest path trees is small. We compared our algorithm with breadth-first search algorithm, and our results showed that ours outperforms breadth-first search on speed and α.

cs.DS↗

arXiv2013

Feature-Gathering Dependency-Based Software Clustering Using Dedication and Modularity

Software clustering is one of the important techniques to comprehend software systems. However, presented techniques to date require human interactions to refine clustering results. In this paper, we proposed a novel dependency-based software clustering algorithm, SArF. SArF has two characteristics. First, SArF eliminates the need of the omnipresent-module-removing step which requires human interactions. Second, the objective of SArF is to gather relevant software features or functionalities into a cluster. To achieve them, we defined the Dedication score to infer the importance of dependencies and utilized Modularity Maximization to cluster weighted directed graphs. Two case studies and extensive comparative evaluations using open source and industrial systems show that SArF could successfully decompose the systems fitting to the authoritative decompositions from a feature viewpoint without any tailored setups and that SArF was superior to existing dependency-based software clustering studies. Besides, the case studies show that there exist measurable authoritativeness limits and that SArF nearly reached the limits.

cs.SE↗

arXiv2013

SArF Map: Visualizing Software Architecture from Feature and Layer Viewpoints

To facilitate understanding the architecture of a software system, we developed SArF Map technique that visualizes software architecture from feature and layer viewpoints using a city metaphor. SArF Map visualizes implicit software features using our previous study, SArF dependency-based software clustering algorithm. Since features are high-level abstraction units of software, a generated map can be directly used for high-level decision making such as reuse and also for communications between developers and non-developer stakeholders. In SArF Map, each feature is visualized as a city block, and classes in the feature are laid out as buildings reflecting their software layer. Relevance between features is represented as streets. Dependency links are visualized lucidly. Through open source and industrial case studies, we show that the architecture of the target systems can be easily overviewed and that the quality of their packaging designs can be quickly assessed.

cs.SE↗

arXiv2006

Direct observation of quasi-particle band in CeIrIn$_5$: Angle-resolved photoemission spectroscopy study

We have performed a high-resolution angle resolved Ce 4$d-4f$ resonant photoemission experiment on the heavy fermion superconductor CeIrIn$_5$. We have observed a quasi-particle band which has an energy dispersion of $\sim 30$ meV in the Ce 4$f$ on-resonance spectra. The result suggests that although the 4$f$ spectra are dominated by the localized/correlated character, the small itinerant component is responsible for the superconductivity in this compound.

cond-mat.str-el↗