Author name cluster

Yasuyuki Okoshi

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

AAAI Conference 2026 Conference Paper

The Strong Lottery Ticket Hypothesis for Multi-Head Attention Mechanisms

Hikari Otsuka
Daiki Chijiwa
Yasuyuki Okoshi
Daichi Fujiki
Susumu Takeuchi
Masato Motomura

The strong lottery ticket hypothesis (SLTH) conjectures that high-performing subnetworks, called strong lottery tickets (SLTs), are hidden in randomly initialized neural networks. Although recent theoretical studies have established the SLTH across various neural architectures, the SLTH for transformer architectures still lacks theoretical understanding. In particular, the current theory of the SLTH does not yet account for the multi-head attention (MHA) mechanism, a core component of transformers. To address this gap, we introduce a theoretical analysis of the existence of SLTs within MHAs. We prove that, if a randomly initialized MHA of H heads and input dimension d has the hidden dimension O(d log(Hd^(3/2))) for the key and value, it contains an SLT that approximates an arbitrary MHA with the same input dimension with high probability. Furthermore, by leveraging this theory for MHAs, we extend the SLTH to transformers without normalization layers. We empirically validate our theoretical findings, demonstrating that the approximation error between the SLT within a source model (MHA and transformer) and an approximate target counterpart decreases exponentially by increasing the hidden dimension of the source model.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

Binary Quadratic Quantization: Beyond First-Order Quantization for Real-Valued Matrix Compression

Kyo Kuroki
Yasuyuki Okoshi
Thiem Van Chu
Kazushi Kawamura
Masato Motomura

This paper proposes a novel matrix quantization method, Binary Quadratic Quan- tization (BQQ). In contrast to conventional first-order quantization approaches— such as uniform quantization and binary coding quantization—that approximate real-valued matrices via linear combinations of binary bases, BQQ leverages the expressive power of binary quadratic expressions while maintaining an extremely compact data format. We validate our approach with two experiments: a matrix compression benchmark and post-training quantization (PTQ) on pretrained Vision Transformer-based models. Experimental results demonstrate that BQQ consistently achieves a superior trade-off between memory efficiency and reconstruction error than conventional methods for compressing diverse matrix data. It also delivers strong PTQ performance, even though we neither target state-of-the-art PTQ accuracy under tight memory constraints nor rely on PTQ-specific binary matrix optimization. For example, our proposed method outperforms the state-of- the-art PTQ method by up to 2. 2% and 59. 1% on the ImageNet dataset under the calibration-based and data-free scenarios, respectively, with quantization equivalent to 2 bits. These findings highlight the surprising effectiveness of binary quadratic expressions for efficient matrix approximation and neural network compression.

PDF Details

TMLR Journal 2025 Journal Article

Partially Frozen Random Networks Contain Compact Strong Lottery Tickets

Hikari Otsuka
Daiki Chijiwa
Ángel López García-Arias
Yasuyuki Okoshi
Kazushi Kawamura
Thiem Van Chu
Daichi Fujiki
Susumu Takeuchi

Randomly initialized dense networks contain subnetworks that achieve high accuracy without weight learning—strong lottery tickets (SLTs). Recently, Gadhikar et al. (2023) demonstrated that SLTs could also be found within a randomly pruned source network. This phenomenon can be exploited to further compress the small memory size required by SLTs. However, their method is limited to SLTs that are even sparser than the source, leading to worse accuracy due to unintentionally high sparsity. This paper proposes a method for reducing the SLT memory size without restricting the sparsity of the SLTs that can be found. A random subset of the initial weights is frozen by either permanently pruning them or locking them as a fixed part of the SLT, resulting in a smaller model size. Experimental results show that Edge-Popup (Ramanujan et al., 2020; Sreenivasan et al., 2022) finds SLTs with better accuracy-to-model size trade-off within frozen networks than within dense or randomly pruned source networks. In particular, freezing $70\%$ of a ResNet on ImageNet provides $3.3\times$ compression compared to the SLT found within a dense counterpart, raises accuracy by up to $14.12$ points compared to the SLT found within a randomly pruned counterpart, and offers a better accuracy-model size trade-off than both.

PDF Details

NeurIPS Conference 2025 Conference Paper

Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time Scaling

Hao Chen
Guanxi Lu
Yasuyuki Okoshi
Zhiwen Mo
Masato Motomura
Hongxiang Fan

Test-time scaling (TTS) has proven effective in enhancing the reasoning capabilities of large language models (LLMs). Verification plays a key role in TTS, simultaneously influencing (1) reasoning performance and (2) compute efficiency, due to the quality and computational cost of verification. In this work, we challenge the conventional paradigms of verification, and make the first attempt toward systematically investigating the impact of verification granularity—that is, how frequently the verifier is invoked during generation, beyond verifying only the final output or individual generation steps. To this end, we introduce Variable Granularity Search (VG-Search), a unified algorithm that generalizes beam search and Best-of-N sampling via a tunable granularity parameter $g$. Extensive experiments with VG-Search under varying compute budgets, generator-verifier configurations, and task attributes reveal that dynamically selecting $g$ can improve the compute efficiency and scaling behavior. Building on these findings, we propose adaptive VG-Search strategies that achieve accuracy gains of up to 3. 1\% over Beam Search and 3. 6\% over Best-of-N, while reducing FLOPs by over 52\%. We will open-source the code to support future research.

PDF Details

TMLR Journal 2025 Journal Article

TicketLLM: Next-Generation Sparse and Low-bit Transformers with Supermask-based Method

Yasuyuki Okoshi
Hikari Otsuka
Daichi Fujiki
Masato Motomura

Strong Lottery Tickets (SLTs) are subnetworks within a randomly weighted network uncovered by a binary mask called supermask. They offer a promising approach to model compression by eliminating the need to store weights since their effective subnetwork can be regenerated from a fixed random seed and the supermask. However, extending this approach to large language models (LLMs) is non-trivial due to limited scalability and inefficient training dynamics of existing SLT methods. To address these challenges, we propose Adaptive Supermask (Ada-Sup), a scalable and efficient method for discovering high-quality multi-bit supermasks through an innovative quantization-based approach. Building on this method, we introduce TicketLLM, a low-bit and sparse Transformer-based LLM architecture powered by Ada-Sup. Experimental results show that Ada-Sup can discover high-quality supermasks with significantly reduced training costs compared to previous methods in both binary and multi-bit settings. Furthermore, TicketLLM outperforms BitNet b1.58 on a 1.3B parameter model with the same memory per connection, achieving 0.6% reduction in perplexity (from 13.62 to 13.54) while operating at a higher sparsity level (around 50% vs. around 33%). These results highlight the potential of supermask-based methods as a promising approach for building lightweight LLMs.

PDF Details

TMLR Journal 2024 Journal Article

Restricted Random Pruning at Initialization for High Compression Range

Hikari Otsuka
Yasuyuki Okoshi
Ángel López García-Arias
Kazushi Kawamura
Thiem Van Chu
Daichi Fujiki
Masato Motomura

Pruning at Initialization (PaI) makes training overparameterized neural networks more efficient by reducing the overall computational cost from training to inference. Recent PaI studies showed that random pruning is more effective than ranking-based pruning, which learns connectivity. However, the effectiveness of each pruning method depends on the existence of skip connections and the compression ratio (the before-after pruning parameter ratio). While random pruning performs better than ranking-based pruning on architectures with skip connections, the superiority without skip connections is reversed in the high compression range. This paper proposes Minimum Connection Assurance (MiCA) that achieves higher accuracy than conventional PaI methods for architectures with and without skip connections, regardless of the compression ratio. MiCA preserves the random connection between the layers and maintains the performance at high compression ratios without the costly connection learning that ranking-based pruning requires. Experiments on image classification using CIFAR-10 and CIFAR-100 and node classification using OGBN-ArXiv show that MiCA enhances the compression ratio and accuracy trade-offs compared to existing PaI methods. In VGG-16 with CIFAR-10, MiCA improves the accuracy of random pruning by $27.0\%$ at $10^{4.7}\times$ compression ratio. Furthermore, experimental analysis reveals that increasing the utilization of the nodes through which information flows from the first layer is essential for maintaining high performance at a high compression ratio.

PDF Details

ICML Conference 2022 Conference Paper

Multicoated Supermasks Enhance Hidden Networks

Yasuyuki Okoshi
Ángel López García-Arias
Kazutoshi Hirose
Kota Ando
Kazushi Kawamura
Thiem Van Chu
Masato Motomura
Jaehoon Yu

Hidden Networks (Ramanujan et al. , 2020) showed the possibility of finding accurate subnetworks within a randomly weighted neural network by training a connectivity mask, referred to as supermask. We show that the supermask stops improving even though gradients are not zero, thus underutilizing backpropagated information. To address this we propose a method that extends Hidden Networks by training an overlay of multiple hierarchical supermasks{—}a multicoated supermask. This method shows that using multiple supermasks for a single task achieves higher accuracy without additional training cost. Experiments on CIFAR-10 and ImageNet show that Multicoated Supermasks enhance the tradeoff between accuracy and model size. A ResNet-101 using a 7-coated supermask outperforms its Hidden Networks counterpart by 4%, matching the accuracy of a dense ResNet-50 while being an order of magnitude smaller.

Details