arXiv subjects

Masahiro Suzuki

Publications and source records attributed to Masahiro Suzuki.

At least 19 recordsLinked to original sources

arXiv2026

Steering Recurrent Reasoners at Inference Time with Readout Feedback

Recurrent models, which repeatedly update latent states with shared computation blocks, have emerged as powerful architectures for solving complex reasoning tasks. Existing inference-time methods scale computation by running more steps or sampling more trajectories, but ignore information revealed within each trajectory. Here we show that recurrent models can be improved at inference time by using their own readout probabilities to steer latent dynamics without retraining. We introduce Readout Feedback (RoFB), a test-time intervention that converts intermediate predictions into token-wise pairwise coupling forces injected into the latent dynamics. Across three recurrent models (AKOrN, ItrSA++, TRM) on Sudoku and Maze, RoFB yields clear gains in four of six model-task pairs, achieving performance unattainable by merely running more steps or selecting from multiple trajectories, at comparable or lower computational cost. These results suggest that closed-loop steering of latent dynamics can serve as a complementary inference-time control mechanism for recurrent reasoning models.

cs.LG↗

arXiv2026

WorldPack: Dynamic Frame Compression for Long-context Video World Modeling

Video world models have attracted significant attention for their ability to produce high-fidelity future visual observations conditioned on past observations and navigation actions. However, achieving temporally and spatially consistent generation over long horizons remains an open challenge: existing approaches either compress past frames without explicitly accounting for 3D viewpoint geometry or retrieve only a handful of spatially relevant frames without increasing the total amount of retained history. In this paper, we propose WorldPack, a video world model that introduces spatially-aware compressed memory to address both limitations simultaneously. The key insight is that compression rates should not be uniform or temporally determined, but should instead be dynamically allocated based on 3D spatial relevance to the current viewpoint. WorldPack achieves this through two tightly coupled mechanisms: trajectory packing, which fits substantially more historical frames into a fixed-length context through hierarchical frame compression, and geometric selection, which leverages camera pose information and field-of-view overlap to assign lower compression to spatially important frames and higher compression to less relevant ones. Together, these mechanisms expand the effective context from 4 to 22 frames with moderate computational overhead: trajectory packing increases diffusion-model inference time by 16%, while FoV-based geometric selection introduces an additional cost. We evaluate WorldPack on LoopNav, a Minecraft benchmark for long-horizon spatial consistency, and conduct comprehensive experiments on the RECON, real-world navigation dataset, across multiple metrics. WorldPack outperforms strong baselines, including Oasis, Mineworld, DIAMOND, and NWM, with pronounced gains on spatial reasoning tasks that require recall of distant observations.

cs.CV↗

arXiv2026

Visual Access Boundaries in Vision-Language Model Reasoning

Chain-of-Thought (CoT) prompting is widely used as a test-time scaling strategy for Vision-Language Models (VLMs), but it remains unclear what is extended when VLMs generate longer reasoning traces. We ask whether CoT requires continued access to image tokens, or whether it mainly operates over visual information already made available earlier in the forward pass. We introduce Visual Access Sweep, a causal intervention that masks attention from generated-token queries to image-token keys along layer depth and generation time, and define the Visual Access Boundary (VAB) as the minimal access region that preserves task accuracy. Across six model configurations from Qwen2.5-VL and InternVL3, both no-CoT direct answering and CoT prompting exhibit finite VABs. In Qwen2.5-VL-32B and InternVL3 at 14B and 38B scales, when CoT is evaluated against the no-CoT full-access target, its VAB layer differs from the no-CoT boundary by at most two layers, despite substantially longer generations. This suggests that CoT does not primarily improve performance by prolonging direct image-token access throughout the reasoning trace, but by extending language-side computation over image-derived hidden-state information. We further show that CoT gains are constrained by perceptual readout. CoT helps when the queried visual attribute can be reliably read out by the model, but not when that readout is unreliable. A symbolic-attribute oracle shows that CoT can improve counting once ground-truth attributes are supplied as text, while a single-object probe-vs-decode check shows that hard attributes can be linearly recoverable from hidden states yet difficult for the model itself to output. Together, these analyses place the bottleneck at readout rather than counting.

cs.AI↗

arXiv2026

SUNTA: Hierarchical Video Prediction with Surprise-based Chunking

Hierarchical state-space models (HSSMs) offer a promising approach to long-horizon prediction by segmenting sequences into temporal chunks. However, their performance hinges on how chunk boundaries are determined. While prior HSSMs typically rely on fixed-length chunking or similarity-based boundary detection, these methods often misalign with the intrinsic temporal structure of the data. We argue that chunking should instead be driven by prediction errors, which more directly indicate when longer-range context becomes necessary. Nevertheless, integrating surprise-based chunking into HSSMs introduces critical challenges, including hierarchical collapse during end-to-end training and the absence of surprise signals during open-loop prediction. To address these issues, we propose Surprise-based Nested Temporal Abstraction (SUNTA), a method that employs a decoupled training strategy to preserve surprise signals and uses internal inconsistency as a top-down surprise metric to determine chunk boundaries within imagined rollouts. Experiments on video prediction tasks in 2D and 3D environments demonstrate that SUNTA outperforms baselines, uniquely maintaining accurate predictions over 250 timesteps, whereas all baselines degrade within the first 10 timesteps.

cs.AI↗

arXiv2026

SSM Meets Video Diffusion Models: Efficient Long-Term Video Generation with Structured State Spaces

Given the remarkable achievements in image generation through diffusion models, the research community has shown increasing interest in extending these models to video generation. Recent diffusion models for video generation have predominantly utilized attention layers to extract temporal features. However, attention layers are limited by their computational costs, which increase quadratically with the sequence length. This limitation presents significant challenges when generating longer video sequences using diffusion models. To overcome this challenge, we propose leveraging state-space models (SSMs) as temporal feature extractors. SSMs (e.g., Mamba) have recently gained attention as promising alternatives due to their linear-time memory consumption relative to sequence length. In line with previous research suggesting that using bidirectional SSMs is effective for understanding spatial features in image generation, we found that bidirectionality is also beneficial for capturing temporal features in video data, rather than relying on traditional unidirectional SSMs. We conducted comprehensive evaluations on multiple long-term video datasets, such as MineRL Navigate, across various model sizes. For sequences up to 256 frames, SSM-based models require less memory to achieve the same FVD as attention-based models. Moreover, SSM-based models often deliver better performance with comparable GPU memory usage. Our codes are available at https://github.com/shim0114/SSM-Meets-Video-Diffusion-Models.

cs.CV↗

arXiv2026

DAD4TS: Data-Augmentation-Oriented Diffusion Model for Time-Series Forecasting with Small-Scale Data

Small-scale data is a critical problem in time-series forecasting tasks. Data augmentation is an effective strategy for this task, but it has a limitation in generating meaningful data. To address this limitation, we propose DAD4TS, a diffusion-model-based data augmentation method with reinforcement learning, designed for time-series forecasting with small-scale data. In DAD4TS, a data generator is simultaneously trained with a time-series model and controlled by a reinforcement learning model to efficiently generate samples that improve the forecast accuracy of the time-series model. To support small-scale data, we use mathematical methods instead of conventional VAE methods to train the diffusion model by projecting the time-series data into the geometric space. We validated the effectiveness of DAD4TS with seven comparative methods through qualitative and quantitative experiments on six real-world datasets and eight time-series models. As a result, DAD4TS was validated on five datasets.

cs.LG↗

arXiv2026

CLIP-like Model as a Foundational Density Ratio Estimator

Density ratio estimation is a core concept in statistical machine learning because it provides a unified mechanism for tasks such as importance weighting, divergence estimation, and likelihood-free inference, but its potential in vision and language models has not been fully explored. Modern vision-language encoders such as CLIP and SigLIP are trained with contrastive objectives that implicitly optimize log density ratios between joint and marginal image-text distributions, which implicitly learn similarity scores proportional to log density ratios. However, prior work has largely focused on their embedding utility, and the density-ratio structure induced by contrastive learning has not been systematically examined or exploited in multimodal applications. To address this gap, we reinterpret CLIP-style models as pretrained and general-purpose density ratio estimators and show that this perspective enables new algorithmic capabilities. We present a unified explanation of how contrastive objectives estimate density ratios and propose two practical applications: Importance Weight Learning and KL divergence estimation. Our Importance Weight Learning method requires only a single additional prompt and improves F1 scores by up to 7 points. We further show that CLIP-based density ratios support estimation of KL divergences that quantify how conditioning on an image or text alters the distribution of the other modality. Through qualitative examples and an N-gram analysis of captions, we find that these divergences capture semantic diversity and mode structure in multimodal data. Leveraging this property, we introduce a simple KL-guided data curation method that achieves performance competitive with LAION2B filtering.

cs.CV↗

arXiv2026

The Vlasov-Poisson system with a perfectly conducting wall: Convex domains

We consider the Vlasov--Poisson system in a $C^3$ convex domain $D$ with a perfectly conducting wall. We introduce the asymptotic domain $D_{\infty}$ for the domain $D$. Then under acceptable assumptions on $D$, we show that for localized initial data, the velocity of particles is asymptotically supported in the (closure of the) asymptotic domain $\overline{D_{\infty}}$ and the solutions exhibit the asymptotics of modified scattering.

math.AP↗

arXiv2026

Financial Market as a Self-Organized Ecosystem: Simulation via Learning with Heterogeneous Preferences

Agent-based models provide a constructive approach to studying emergent dynamics in life-like systems composed of interacting, adaptive agents. Financial markets serve as a canonical example of such systems, where collective price dynamics arise from individual decision-making. In this modeling tradition, investor behavior has typically been captured by two distinct mechanisms -- learning and heterogeneous preferences -- which have been explored as separate paradigms in prior studies. However, the impact of their joint modeling on the resulting collective dynamics remains largely unexplored. We develop a multi-agent reinforcement learning framework in which agents endowed with heterogeneous risk aversion, time discounting, and information access learn trading strategies interactively within an artificial market. The experiment reveals that (i) learning under heterogeneous preferences drives agents to develop functionally differentiated strategies through interaction, rather than trait-specific rules, resulting in role specialization, and (ii) the interactions by the differentiated agents are essential for the emergence of realistic market dynamics such as fat-tailed price fluctuations and volatility clustering. Overall, this study demonstrates that the joint design of heterogeneous preferences and learning mechanisms enables the synthesis of an artificial market in which adaptive interactions drive the self-organization of a market ecology, providing a computational realization of the Adaptive Market Hypothesis.

q-fin.CP↗

arXiv2026

JFinTEB: Japanese Financial Text Embedding Benchmark

We introduce JFinTEB, the first comprehensive benchmark specifically designed for evaluating Japanese financial text embeddings. Existing embedding benchmarks provide limited coverage of language-specific and domain-specific aspects found in Japanese financial texts. Our benchmark encompasses diverse task categories including retrieval and classification tasks that reflect realistic and well-defined financial text processing scenarios. The retrieval tasks leverage instruction-following datasets and financial text generation queries, while classification tasks cover sentiment analysis, document categorization, and domain-specific classification challenges derived from economic survey data. We conduct extensive evaluations across a wide range of embedding models, including Japanese-specific models of various sizes, multilingual models, and commercial embedding services. We publicly release JFinTEB datasets and evaluation framework at https://github.com/retarfi/JFinTEB to facilitate future research and provide a standardized evaluation protocol for the Japanese financial text mining community. This work addresses a critical gap in Japanese financial text processing resources and establishes a foundation for advancing domain-specific embedding research.

cs.IR↗

arXiv2026

Periodic Event-Triggered Explicit Reference Governor for Constrained Attitude Control on SO(3)

This letter addresses the constrained attitude control problem for rigid bodies directly on the special orthogonal group SO(3), avoiding singularities associated with parameterizations such as Euler angles. We propose a novel Periodic Event-Triggered Explicit Reference Governor (PET-ERG) that enforces input saturation and geometric pointing constraints without relying on online optimization. A key feature is a periodic event-triggered supervisory update: the auxiliary reference is updated only at sampled instants when a robust safety condition is met, thereby avoiding continuous-time reference updates and enabling a rigorous stability analysis of the cascade system on the manifold. Through this structured approach, we rigorously establish the asymptotic stability and exponential convergence of the closed-loop system for almost all initial configurations. Numerical simulations validate the effectiveness of the proposed control architecture and demonstrate constraint satisfaction and convergence properties.

eess.SY↗

arXiv2026

MultiBanana: A Challenging Benchmark for Multi-Reference Text-to-Image Generation

Recent text-to-image generation models have acquired the ability of multi-reference generation and editing; that is, to inherit the appearance of subjects from multiple reference images and re-render them in new contexts. However, existing benchmark datasets often focus on generation using a single or a few reference images, which prevents us from measuring progress in model performance or identifying weaknesses when following instructions with a larger number of references. In addition, their task definitions are still vague, limited to axes such as ``what to edit'' or ``how many references are given'', and therefore fail to capture the challenges inherent in combining heterogeneous references. To address this gap, we introduce MultiBanana, which is designed to assess the edge of model capabilities by widely covering problems specific to multi-reference settings: (1) varying the number of references (up to 8), (2) domain mismatch among references (e.g., photo vs. anime), (3) scale mismatch between reference and target scenes, (4) references containing rare concepts (e.g., a red banana), and (5) multilingual textual references for rendering. Our analysis among a variety of text-to-image models reveals their respective performances, typical failure modes, and areas for improvement. MultiBanana is released as an open benchmark to push the boundaries and establish a standardized basis for fair comparison in multi-reference image generation. Our data and code are available at https://github.com/matsuolab/multibanana .

cs.CV↗

arXiv2025

High-Voltage Ionized Gas with Spherical Cathode Emission

We consider a plasma that is created by a high voltage difference, which is known as a Townsend gas discharge. The plasma is confined to the region between two concentric spheres, one of which is a cathode and the other an anode. Ion-electron pairs are created by collisions inside the plasma. Additional electrons enter the plasma by collisions of ions with the cathode. We prove under certain conditions that there are many steady states exhibiting gas discharge, beginning with a `sparking' voltage. In fact, there is an analytic one-parameter family of them that connects the non-ionized gas to a plasma with arbitrarily high ionization or arbitrarily high potential, or else the family ends at an `anti-sparking' voltage.

math.AP↗

arXiv2025

Emergence from Emergence: Financial Market Simulation via Learning with Heterogeneous Preferences

Agent-based models help explain stock price dynamics as emergent phenomena driven by interacting investors. In this modeling tradition, investor behavior has typically been captured by two distinct mechanisms -- learning and heterogeneous preferences -- which have been explored as separate paradigms in prior studies. However, the impact of their joint modeling on the resulting collective dynamics remains largely unexplored. We develop a multi-agent reinforcement learning framework in which agents endowed with heterogeneous risk aversion, time discounting, and information access collectively learn trading strategies within a unified shared-policy framework. The experiment reveals that (i) learning with heterogeneous preferences drives agents to develop strategies aligned with their individual traits, fostering behavioral differentiation and niche specialization within the market, and (ii) the interactions by the differentiated agents are essential for the emergence of realistic market dynamics such as fat-tailed price fluctuations and volatility clustering. This study presents a constructive paradigm for financial market modeling in which the joint design of heterogeneous preferences and learning mechanisms enables two-stage emergence: individual behavior and the collective market dynamics.

cs.CY↗

arXiv2025

When Object-Centric World Models Meet Policy Learning: From Pixels to Policies, and Where It Breaks

Object-centric world models (OCWM) aim to decompose visual scenes into object-level representations, providing structured abstractions that could improve compositional generalization and data efficiency in reinforcement learning. We hypothesize that explicitly disentangled object-level representations, by localizing task-relevant information, can enhance policy performance across novel feature combinations. To test this hypothesis, we introduce DLPWM, a fully unsupervised, disentangled object-centric world model that learns object-level latents directly from pixels. DLPWM achieves strong reconstruction and prediction performance, including robustness to several out-of-distribution (OOD) visual variations. However, when used for downstream model-based control, policies trained on DLPWM latents underperform compared to DreamerV3. Through latent-trajectory analyses, we identify representation shift during multi-object interactions as a key driver of unstable policy learning. Our results suggest that, although object-centric perception supports robust visual modeling, achieving stable control requires mitigating latent drift.

cs.AI↗

arXiv2025

Agent-Based Simulation of a Financial Market with Large Language Models

In real-world stock markets, certain chart patterns -- such as price declines near historical highs -- cannot be fully explained by fundamentals alone. These phenomena suggest the presence of path dependence in price formation, where investor decisions are influenced not only by current market conditions but also by the trajectory of prices leading up to the present. Path dependence has drawn attention in behavioral finance as a key mechanism behind such anomalies. One plausible driver of path dependence is human loss aversion, anchored to individual reference points like purchase prices or past peaks, which vary with personal context. However, capturing such subtle behavioral tendencies in traditional agent-based market simulations has remained a challenge. We propose the Fundamental-Chartist-LLM-Agent (FCLAgent), which uses large language models (LLMs) to emulate human-like trading decisions. In this framework, (1) buy/sell decisions are made by LLMs based on individual situations, while (2) order price and volume follow standard rule-based methods. Simulations show that FCLAgents reproduce path-dependent patterns that conventional agents fail to capture. Furthermore, an analysis of FCLAgents' behavior reveals that the reference points guiding loss aversion vary with market trajectories, highlighting the potential of LLM-based agents to model nuanced investor behavior.

cs.CE↗

arXiv2025

Fe contribution to the magnetic anisotropy of $L{1_0}$-ordered FePt thin films studied by angle-dependent x-ray magnetic circular dichroism

Among magnetic thin films with perpendicular magnetic anisotropy (PMA), $L1_0$-ordered FePt has attracted significant attention because of its exceptionally strong PMA. However, the microscopic origin of its strong PMA has not been elucidated experimentally. We have investigated the contribution of the Fe $3d$ electrons to its magnetic anisotropy energy by angle-dependent x-ray magnetic circular dichroism at the Fe $L_{2,3}$ edge. By this technique, one can deduce the magnetic dipole moment $m_\text{T}$, which represents the anisotropic spatial distribution of spin-polarized electrons, and the orbital moment anisotropy (OMA) of Fe $3d$ electrons. Detected finite $m_\text{T}$ indicates that the spin-polarized Fe $3d$ electrons are distributed preferentially in the out-of-plane direction of the films. This $m_\text{T}$ of Fe overwhelms the positive contribution of OMA to PMA, and reduces the PMA of $L1_0$-ordered FePt thin films, consistent with a previous first-principles calculation. The present result implies that a large positive contribution of the non-magnetic element Pt rather than Fe governs the PMA of $L1_0$-ordered FePt thin films.

cond-mat.mtrl-sci↗

arXiv2025

Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search

The remarkable progress in text-to-video diffusion models enables the generation of photorealistic videos, although the content of these generated videos often includes unnatural movement or deformation, reverse playback, and motionless scenes. Recently, an alignment problem has attracted huge attention, where we steer the output of diffusion models based on some measure of the content's goodness. Because there is a large room for improvement of perceptual quality along the frame direction, we should address which metrics we should optimize and how we can optimize them in the video generation. In this paper, we propose diffusion latent beam search with lookahead estimator, which can select a better diffusion latent to maximize a given alignment reward at inference time. We then point out that improving perceptual video quality with respect to alignment to prompts requires reward calibration by weighting existing metrics. This is because when humans or vision language models evaluate outputs, many previous metrics to quantify the naturalness of video do not always correlate with the evaluation. We demonstrate that our method improves the perceptual quality evaluated on the calibrated reward, VLMs, and human assessment, without model parameter update, and outputs the best generation compared to greedy search and best-of-N sampling under much more efficient computational cost. The experiments highlight that our method is beneficial to many capable generative models, and provide a practical guideline: we should prioritize the inference-time compute allocation into enabling the lookahead estimator and increasing the search budget, rather than expanding the denoising steps.

cs.CV↗