Institution profile

Preferred Networks, Inc.

Industry researchasia · jp
Official website
Research library42linked papers
Opportunities0open roles
Selected work

Representative Papers

Tree-structured Parzen estimator: Understanding its algorithm components and their roles for better empirical performance

Apr 21, 2023arXiv.org

The Tree-structured Parzen Estimator (TPE), a mainstream Bayesian optimization method, lacks a systematic understanding of the functional roles, impact patterns, and synergistic interactions among its key hyperparameters (e.g., γ, n_startup, n_ei_candidates). Method: We conduct comprehensive ablation studies across diverse benchmark functions to empirically dissect these parameters’ behaviors and dependencies. Based on rigorous empirical analysis, we derive an interpretable and robust default configuration strategy. Contribution/Results: Our recommended configuration significantly outperforms the standard TPE implementation and leading baselines—including SMAC and Hyperopt—across heterogeneous benchmarks, improving both optimization efficiency and stability. All experiments are fully reproducible, with source code publicly released.

180 citations11 influentialRead paper

Revisiting Structural Dependency in Autoregressive Multi-Task Table Recognition via Order-Independent Cell-Level Representations

Jun 16, 2026

This work addresses the limitation of autoregressive table recognition methods, whose sequential generation process induces cell representations that are dependent on generation order, thereby compromising global structural consistency. To overcome this, the authors propose a structure refinement module incorporating non-causal attention within a unified multi-task framework, enabling the learning of order-invariant, cell-level feature representations while simultaneously performing table structure prediction, cell localization, and content recognition. By eliminating reliance on autoregressive ordering, the approach substantially enhances both global coherence and parallel inference efficiency. Experimental results on two large-scale datasets demonstrate significant improvements in cell localization and end-to-end recognition accuracy, along with an approximately threefold reduction in overall inference time.

0 citationsRead paper
Recent publications

Latest Papers

Revisiting Structural Dependency in Autoregressive Multi-Task Table Recognition via Order-Independent Cell-Level Representations

Jun 16, 2026

This work addresses the limitation of autoregressive table recognition methods, whose sequential generation process induces cell representations that are dependent on generation order, thereby compromising global structural consistency. To overcome this, the authors propose a structure refinement module incorporating non-causal attention within a unified multi-task framework, enabling the learning of order-invariant, cell-level feature representations while simultaneously performing table structure prediction, cell localization, and content recognition. By eliminating reliance on autoregressive ordering, the approach substantially enhances both global coherence and parallel inference efficiency. Experimental results on two large-scale datasets demonstrate significant improvements in cell localization and end-to-end recognition accuracy, along with an approximately threefold reduction in overall inference time.

0 citationsRead paper

Rewrite to Translate, Translate to Reward: Reinforcement Learning for Source Rewriting in Machine Translation

Jun 06, 2026

This work addresses the limitations of existing large language model–based source-side rewriting approaches, which rely on manual prompt tuning tailored to specific machine translation models and thus lack generality and automation. The authors propose RLSR, a novel framework that, for the first time, directly employs translation quality as the reward signal in reinforcement learning to train a source-side rewriting model—eliminating the need for handcrafted or tuned prompts. Evaluated across six translation models and sixteen language pairs, RLSR demonstrates consistent effectiveness: a 4B-parameter RLSR model significantly outperforms both the no-rewriting baseline and same-scale prompt-based rewriting methods, achieving performance comparable to that of rewriting using a 235B-parameter large language model. This establishes RLSR as a general, automated, and efficient mechanism for source-side rewriting.

0 citationsRead paper