arXiv subjects

Hiromi Wakaki

Publications and source records attributed to Hiromi Wakaki.

At least 19 recordsLinked to original sources

arXiv2026

Syn-Omni: Structured Specialization and Progressive Collaboration for Omnimodal Embeddings

Omnimodal embeddings naturally involve both shared representations and modality-specific features across heterogeneous inputs. However, existing omnimodal embedding methods often rely on a single shared parameter space over mixed-modality data, limiting structural separation between universal and modality-specific representations. To address this, we propose Syn-Omni, a unified framework for structured omnimodal adaptation with modality specialization and controlled cross-modal collaboration. Specifically, we introduce Orthogonal Modality-Expert LoRA (OME-LoRA), which decomposes adaptation into a shared LoRA path for universal semantics and modality-expert LoRA paths for modality-aware specialization. Furthermore, Progressive Synergy Routing (PSR) enables experts to first establish modality-specific priors, then gradually interact with other modality-experts for cross-modal synergy. Evaluated across 81 diverse tasks spanning image, video, audio, and audiovisual modalities, Syn-Omni consistently outperforms omnimodal baselines, demonstrating the effectiveness of structured specialization and cross-modal progressive collaboration.

cs.CV↗

arXiv2026

MCA: Modality Composition Awareness for Robust Composed Multimodal Retrieval

Multimodal retrieval, which seeks to retrieve relevant content across modalities such as text or image, supports applications from AI search to contents production. Despite the success of separate-encoder approaches like CLIP aligning modality-specific embeddings with contrastive learning, recent multimodal large language models (MLLMs) enable a unified encoder that directly processes composed inputs. While flexible and advanced, we identify that unified encoders trained with conventional contrastive learning are prone to learn modality shortcut, leading to poor robustness under distribution shifts. We propose a modality composition awareness framework to mitigate this issue. Concretely, it consists of a preference loss enforces multimodal embeddings to outperform their unimodal counterparts, and a composition regularization objective aligns multimodal embeddings with prototypes composed from its unimodal parts. These objectives explicitly model structural relationships between the composed representation and its unimodal counterparts. Experiments on various benchmarks show gains in out-of-distribution retrieval, highlighting modality composition awareness as a effective principle for robust composed multimodal retrieval when utilizing MLLMs as the unified encoder.

cs.CL↗

arXiv2026

Demystifying MaskGIT Sampler and Beyond: Adaptive Order Selection in Masked Diffusion

Masked diffusion models have shown promising performance in generating high-quality samples in a wide range of domains, but accelerating their sampling process remains relatively underexplored. To investigate efficient samplers for masked diffusion, this paper theoretically analyzes the MaskGIT sampler for image modeling, revealing its implicit temperature sampling mechanism. Through this analysis, we show that MaskGIT is asymptotically equivalent to a choose-then-sample (CTS) formulation, instantiated as the "moment sampler," which explicitly separates index selection from token sampling. This CTS reformulation is essential: it yields unbiased token sampling and exposes an algorithmic design space for index selection, both of which are inaccessible in MaskGIT's original formulation. Regarding token sampling, we reveal that MaskGIT implicitly adopts a low-temperature sampler, which explains why MaskGIT often degrades with more sampling steps. The CTS reformulation of MaskGIT allows us to fix the temperature sampling to ensure unbiasedness. We also improve the index selection in CTS through two key innovations: a partial caching technique for transformers that approximates longer sampling trajectories without proportional computational cost, and a hybrid approach formalizing the exploration-exploitation trade-off in adaptive unmasking. Experiments in image and text domains demonstrate our theory as well as the efficiency of our proposed methods, advancing both theoretical understanding and practical implementation of masked diffusion samplers.

cs.LG↗

arXiv2026

Theoretical Refinement of CLIP by Utilizing Linear Structure of Optimal Similarity

In this study, we propose an enhancement to the similarity computation mechanism in multimodal contrastive pretraining frameworks such as CLIP. Prior theoretical research has demonstrated that the optimal similarity metrics between paired modalities should correspond to the pointwise mutual information (PMI) between the two modalities. However, the current implementations of CLIP and its variants fail to fully utilize the underlying linear structure of PMI. We therefore propose KME-CLIP, which leverages this structure through the inner product in a reproducing kernel Hilbert space (RKHS). We theoretically prove that, under our assumptions, the KME-CLIP similarity can bring the contrastive loss arbitrarily close to its optimal value, which is attained by PMI, as the size of the point set grows, and we empirically evaluate KME-CLIP against CLIP and its kernel-based variants across several retrieval and classification tasks.

cs.LG↗

arXiv2026

MusTBench: Benchmarking and Advancing Temporal Grounding in Music LLMs

Recent Large Audio-Language Models (LALMs) have demonstrated promising abilities in understanding musical content. However, whether their responses are grounded in the correct temporal regions of the audio remains underexplored. This limitation is particularly critical for music understanding, where key information often occurs as temporally localized events, such as instrument entries and rhythmic transitions. To address this gap, we introduce MusTBench, a music-expert-validated benchmark designed to evaluate temporal grounding in LALMs through five temporally grounded question-answering tasks. To further improve temporal grounding in existing models, we propose MusT, a novel four-stage temporal optimization recipe spanning music encoder adaptation, LLM adaptation, LLM supervised fine-tuning, and RL-based optimization. Experiments on MusTBench show that existing LALMs struggle with precise temporal grounding, while MusT brings significant improvements over strong baselines. These results establish temporal grounding as a key missing capability in current LALMs and position MusTBench as a challenging benchmark for future research in temporally grounded music understanding.

cs.CL↗

arXiv2026

MLLMCLIP: Feature-Level Distillation of MLLM for Robust Vision-Language Representations

Pretrained vision-language models such as CLIP excel at zero-shot recognition but often fail at compositionality, particularly attribute-object and relational structures. Recent studies mitigate this issue by augmenting training with synthetic hard negatives generated by a cascade of large language models and text-to-image models, which incurs substantial pipeline overhead. We instead propose MLLMCLIP, a heterogeneous distillation framework that transfers multimodal knowledge directly from a generative Multimodal Large Language Model (MLLM) teacher into a discriminative CLIP student, bypassing synthetic data entirely. To bridge the architectural mismatch between the two paradigms, we introduce an attention-based per-layer token selection and a CKA-based distillation loss. Compared to prior CLIP-enhancement methods, MLLMCLIP achieves state-of-the-art compositional accuracy while delivering consistent gains on standard zero-shot classification and image-text retrieval, showing that feature-level distillation strengthens both compositional and general vision-language representation capability.

cs.CV↗

arXiv2026

Searching Videos as Trees: Self-Correcting Agents for Grounded Long Video QA

Grounded long-video question answering (Grounded LVQA) requires answering a question about a long video while localizing the short evidence interval that supports the answer. Recent agentic methods frame this task as multi-turn exploration with a single crop_video(start, end) action, which supports coarse-to-fine narrowing but provides no primitive for fine-to-coarse backtracking. As a result, these agents typically converge prematurely and cannot recover from an early mistake. We propose VideoTreeSearch (VTS), a framework that casts grounded LVQA as iterative self-correcting search over an adaptive temporal tree. VTS constructs a non-uniform tree from visual scene boundaries so that each node corresponds to a semantically coherent segment, and trains an agent to navigate the tree through four discrete operations: zoom_in, zoom_out, shift, and answer. These operations expose backtracking and recovery as explicit, learnable primitives rather than implicit behaviors. To train this navigation, we introduce a trajectory synthesis pipeline that produces multi-step paths through the tree, including deliberate detours into incorrect branches followed by recovery. We use these trajectories for supervised fine-tuning, followed by reinforcement learning with grounding and answer-accuracy rewards. On three Grounded LVQA benchmarks (CG-Bench, Haystack-LVBench, Haystack-Ego4D), VTS outperforms the strongest prior agentic methods by +12.5 mIoU on CG-Bench and +7.4 T-F1 on Haystack-Ego4D. The learned policy also transfers to general long-video QA, surpassing all prior agentic baselines on Video-MME, MLVU, and LVBench by up to +7.1 accuracy points. Ablations confirm that self-correcting hierarchical search is the central mechanism behind these gains: removing either adaptive descent or explicit backtracking substantially degrades performance. Code is available at https://github.com/CeeZh/VTS.

cs.CV↗

arXiv2026

Mixture of Probes: Learning from Privileged Modalities in Multimodal LLMs Through Probing

Multimodal Large Language Models (MLLMs) are typically designed under the assumption that all modalities available during training will also be accessible at inference. However, many real-world settings violate this assumption, requiring models to operate under a privileged modality setting, where auxiliary modalities are available only during training. While these modalities contain valuable information, existing MLLMs largely fail to leverage them effectively, as they treat modalities as interchangeable inputs rather than sources of complementary supervision. We propose Mixture of Probes (MoP), a novel framework that disentangles modality-specific and modality-general signals within the MLLM, allowing the model to preserve modality-dependent structure while learning transferable representations across modalities. At its core, MoP achieves this through a structured probing mechanism that extracts and organizes information from intermediate representations of a shared modality encoder, rather than relying only on final-layer alignment as done in existing MLLMs. To support this disentanglement, we further introduce MoP Cross-modal Training (MoP-X), a training strategy for MoP centered around a probe disentanglement loss that prevents probe collapse and encourages cross-modal learning. We evaluate MoP across two domains spanning eight tasks and four modalities under a comprehensive evaluation protocol tailored to the privileged modality setting, where each modality is independently treated as the sole input at inference time. MoP consistently outperforms strong MLLM baselines, achieving up to 65% relative improvement, demonstrating that auxiliary modalities, even when unavailable at inference, can provide substantial gains when effectively leveraged during training. Code, model checkpoints, and evaluation protocols will be made available at https://github.com/Sony/MoP.

cs.CV↗

arXiv2026

DynaVieW: Schema-Guided World Modeling for Understanding Hierarchical Visual Dynamics

Multimodal LLMs struggle to systematically model the temporal evolution of visual scenes in videos or multi-image sequences. Such inputs require models to predict or simulate multiple levels of dynamic constituents, such as actions taken in the visual sequence, and the associated changes to the visual environment that result. To address this challenge, we propose a dynamic schema-guided world model, DynaVieW, optimized for visual dynamic prediction and simulation. DynaVieW achieves an in-depth understanding of visual dynamics by learning interleaved state-transition sequences, where states cover broad visual scenes from video keyframes, and transitions capture comprehensive dynamic constituents within a hierarchical schema. DynaVieW jointly models transition prediction and state simulation under a mixture-of-experts architecture, with a cross-expert selective attention and a schema token re-weighted loss, to ensure effective and robust learning. DynaVieW's understanding of visual dynamics boosts its downstream performance in visual narrative creation and world simulation, showing improved consistency, controllability, and instruction-following.

cs.LG↗

arXiv2026

Learning to Route Languages for Multilingual Policy Optimization

Large language models~(LLMs) are trained on heterogeneous multilingual corpora, yet existing policy optimization methods often implicitly restrict each training question to a single response language or rely on a fixed dominant language for supervision. We propose language-routed policy optimization (LRPO), an online policy optimization framework that treats language as a selectable variable. LRPO elicits multilingual rollouts for each training question and integrates their relative quality into preference-based policy updates, increasing the diversity and informativeness of training signals under the fixed rollout budget. To adaptively determine which languages to explore during reinforcement learning, we introduce a trainable language router formulated as a multi-armed bandit, balancing exploration of underutilized languages with exploitation of more informative ones. Extensive experiments show that LRPO consistently improves multilingual performance, demonstrating that adaptive language routing enables effective cross-lingual knowledge exploitation for training. We release all the resources at https://github.com/Guochry/LRPO.

cs.CL↗

arXiv2025

DeepResonance: Enhancing Multimodal Music Understanding via Music-centric Multi-way Instruction Tuning

Recent advancements in music large language models (LLMs) have significantly improved music understanding tasks, which involve the model's ability to analyze and interpret various musical elements. These improvements primarily focused on integrating both music and text inputs. However, the potential of incorporating additional modalities such as images, videos and textual music features to enhance music understanding remains unexplored. To bridge this gap, we propose DeepResonance, a multimodal music understanding LLM fine-tuned via multi-way instruction tuning with multi-way aligned music, text, image, and video data. To this end, we construct Music4way-MI2T, Music4way-MV2T, and Music4way-Any2T, three 4-way training and evaluation datasets designed to enable DeepResonance to integrate both visual and textual music feature content. We also introduce multi-sampled ImageBind embeddings and a pre-LLM fusion Transformer to enhance modality fusion prior to input into text LLMs, tailoring for multi-way instruction tuning. Our model achieves state-of-the-art performances across six music understanding tasks, highlighting the benefits of the auxiliary modalities and the structural superiority of DeepResonance. We open-source the codes, models and datasets we constructed: github.com/sony/DeepResonance.

cs.SD↗

arXiv2025

CARE: Multilingual Human Preference Learning for Cultural Awareness

Language Models (LMs) are typically tuned with human preferences to produce helpful responses, but the impact of preference tuning on the ability to handle culturally diverse queries remains understudied. In this paper, we systematically analyze how native human cultural preferences can be incorporated into the preference learning process to train more culturally aware LMs. We introduce \textbf{CARE}, a multilingual resource containing 3,490 culturally specific questions and 31.7k responses with human judgments. We demonstrate how a modest amount of high-quality native preferences improves cultural awareness across various LMs, outperforming larger generic preference data. Our analyses reveal that models with stronger initial cultural performance benefit more from alignment, leading to gaps among models developed in different regions with varying access to culturally relevant data. CARE is publicly available at https://github.com/Guochry/CARE.

cs.CL↗

arXiv2025

GLOV: Guided Large Language Models as Implicit Optimizers for Vision Language Models

In this work, we propose GLOV, which enables Large Language Models (LLMs) to act as implicit optimizers for Vision-Language Models (VLMs) to enhance downstream vision tasks. GLOV prompts an LLM with the downstream task description, querying it for suitable VLM prompts (e.g., for zero-shot classification with CLIP). These prompts are ranked according to their fitness for the downstream vision task. In each respective optimization step, the ranked prompts are fed as in-context examples (with their accuracies) to equip the LLM with the knowledge of the type of prompts preferred by the downstream VLM. Furthermore, we explicitly guide the LLM's generation at each optimization step by adding an offset vector -- calculated from the embedding differences between previous positive and negative solutions -- to the intermediate layer of the network for the next generation. This offset vector biases the LLM generation toward the type of language the downstream VLM prefers, resulting in enhanced performance on the downstream vision tasks. We comprehensively evaluate our GLOV on two tasks: object recognition and the critical task of enhancing VLM safety. Our GLOV shows performance improvement by up to 15.0% and 57.5% for dual-encoder (e.g., CLIP) and encoder-decoder (e.g., LlaVA) models for object recognition and reduces the attack success rate (ASR) on state-of-the-art VLMs by up to $60.7\%$.

cs.CV↗

arXiv2025

Distillation of Discrete Diffusion through Dimensional Correlations

Diffusion models have demonstrated exceptional performances in various fields of generative modeling, but suffer from slow sampling speed due to their iterative nature. While this issue is being addressed in continuous domains, discrete diffusion models face unique challenges, particularly in capturing dependencies between elements (e.g., pixel relationships in image, sequential dependencies in language) mainly due to the computational cost of processing high-dimensional joint distributions. In this paper, (i) we propose "mixture" models for discrete diffusion that are capable of treating dimensional correlations while remaining scalable, and (ii) we provide a set of loss functions for distilling the iterations of existing models. Two primary theoretical insights underpin our approach: First, conventional models with element-wise independence can well approximate the data distribution, but essentially require {\it many sampling steps}. Second, our loss functions enable the mixture models to distill such many-step conventional models into just a few steps by learning the dimensional correlations. Our experimental results show the effectiveness of the proposed method in distilling pretrained discrete diffusion models across image and language domains. The code used in the paper is available at https://github.com/sony/di4c .

cs.LG↗

arXiv2025

VinaBench: Benchmark for Faithful and Consistent Visual Narratives

Visual narrative generation transforms textual narratives into sequences of images illustrating the content of the text. However, generating visual narratives that are faithful to the input text and self-consistent across generated images remains an open challenge, due to the lack of knowledge constraints used for planning the stories. In this work, we propose a new benchmark, VinaBench, to address this challenge. Our benchmark annotates the underlying commonsense and discourse constraints in visual narrative samples, offering systematic scaffolds for learning the implicit strategies of visual storytelling. Based on the incorporated narrative constraints, we further propose novel metrics to closely evaluate the consistency of generated narrative images and the alignment of generations with the input textual narrative. Our results across three generative vision models demonstrate that learning with VinaBench's knowledge constraints effectively improves the faithfulness and cohesion of generated visual narratives.

cs.CV↗

arXiv2025

Cross-Modal Learning for Music-to-Music-Video Description Generation

Music-to-music-video generation is a challenging task due to the intrinsic differences between the music and video modalities. The advent of powerful text-to-video diffusion models has opened a promising pathway for music-video (MV) generation by first addressing the music-to-MV description task and subsequently leveraging these models for video generation. In this study, we focus on the MV description generation task and propose a comprehensive pipeline encompassing training data construction and multimodal model fine-tuning. We fine-tune existing pre-trained multimodal models on our newly constructed music-to-MV description dataset based on the Music4All dataset, which integrates both musical and visual information. Our experimental results demonstrate that music representations can be effectively mapped to textual domains, enabling the generation of meaningful MV description directly from music inputs. We also identify key components in the dataset construction pipeline that critically impact the quality of MV description and highlight specific musical attributes that warrant greater focus for improved MV description generation.

cs.SD↗

arXiv2025

TED: Turn Emphasis with Dialogue Feature Attention for Emotion Recognition in Conversation

Emotion recognition in conversation (ERC) has been attracting attention by methods for modeling multi-turn contexts. The multi-turn input to a pretraining model implicitly assumes that the current turn and other turns are distinguished during the training process by inserting special tokens into the input sequence. This paper proposes a priority-based attention method to distinguish each turn explicitly by adding dialogue features into the attention mechanism, called Turn Emphasis with Dialogue (TED). It has a priority for each turn according to turn position and speaker information as dialogue features. It takes multi-head self-attention between turn-based vectors for multi-turn input and adjusts attention scores with the dialogue features. We evaluate TED on four typical benchmarks. The experimental results demonstrate that TED has high overall performance in all datasets and achieves state-of-the-art performance on IEMOCAP with numerous turns.

cs.CL↗

arXiv2024

OpenMU: Your Swiss Army Knife for Music Understanding

We present OpenMU-Bench, a large-scale benchmark suite for addressing the data scarcity issue in training multimodal language models to understand music. To construct OpenMU-Bench, we leveraged existing datasets and bootstrapped new annotations. OpenMU-Bench also broadens the scope of music understanding by including lyrics understanding and music tool usage. Using OpenMU-Bench, we trained our music understanding model, OpenMU, with extensive ablations, demonstrating that OpenMU outperforms baseline models such as MU-Llama. Both OpenMU and OpenMU-Bench are open-sourced to facilitate future research in music understanding and to enhance creative music production efficiency.

cs.SD↗