Cross-Modal Learning

Audio and vision provide complementary evidence for audio-visual question answering, yet current audio-visual large language models may suffer from cross-modal interference: information from one modality misguides the interpretation of another, thereby inducing hallucinations. We attribute this issue to uncontrolled cross-modal interactions during intermediate reasoning. To mitigate this, we propose Separate First, Fuse Later (SFFL), an audio-visual reasoning framework designed to reduce cross-modal interference. SFFL enforces modality-specific chain-of-thought reasoning, producing separate audio and visual reasoning traces and integrating evidence for answering. We construct modality-preference labels via a data pipeline under different modality input settings. We use these labels as an auxiliary reward in reinforcement learning to encourage a instance-dependent preference for modality cues when answering. We further introduce a modality-specific reasoning mechanism that preserves modality isolation during the separated reasoning stage while enabling full access to cross-modal information at the evidence fusion stage. Experiments demonstrate consistent improvements in both accuracy and robustness, yielding an average relative gain of 5.16% on general AVQA benchmarks and 11.17% on a cross-modal hallucination benchmark.

Real-world model deployment across multiple domains requires multimodal models to operate under two complementary regimes: (1) multi-task pretraining, tasks are co-available at design time where related tasks could borrow representational strength from one another, (2) continual adaptation, in which new tasks emerge after deployment with previously unseen modality combinations. However, neither regime alone suffices: the pretraining task set is never exhaustive, while bypassing joint training forfeits the transfer gains and efficiency among co-trainable tasks. Sparse Mixture-of-Experts (MoE) is a natural fit for this dual requirement: sparse activation enables modular capacity expansion as new tasks arrive, while routing decouples modality-level computation from task-level composition. In this work, we propose a scalable MoE framework for multitask pretraining and continual learning across flexible modality combinations. The framework is designed to support training on multimodal tasks with diverse modality configurations by leveraging modality-specific routers that process tokens from each modality across tasks. Furthermore, it enables continual learning over sequential multimodal tasks within a fixed-capacity MoE by compressing accumulated expert knowledge into low-rank memory subspaces, while expanding only the lightweight routers. We validate the effectiveness of our method on multiple healthcare multimodal benchmarks. It demonstrates competitive multitask pretraining performance while alleviating catastrophic forgetting and improving parameter efficiency.

We introduce SeePhys Pro, a fine-grained modality transfer benchmark that studies whether models preserve the same reasoning capability when critical information is progressively transferred from text to image. Unlike standard vision-essential benchmarks that evaluate a single input form, SeePhys Pro features four semantically aligned variants for each problem with progressively increasing visual elements. Our evaluation shows that current frontier models are far from representation-invariant reasoners: performance degrades on average as information moves from language to diagrams, with visual variable grounding as the most critical bottleneck. Motivated by this inference-time fragility, we further develop large training corpora for multimodal RLVR and use blind training as a diagnostic control, finding that RL with all training images masked can still improve performance on unmasked validation sets. To analyze this effect, text-deletion, image-mask-rate, and format-saturation controls suggest that such gains can arise from residual textual and distributional cues rather than valid visual evidence. Our results highlight the need to evaluate multimodal reasoning not only by final-answer accuracy, but also by robustness under modality transfer and by diagnostics that test whether improvements rely on task-critical visual evidence.

Zero-shot composed image retrieval (ZS-CIR) retrieves a target image from a reference image and a text modification without human-annotated CIR triplets. Projection-based ZS-CIR methods are attractive because they do not rely on LLMs at inference and remain lightweight, but they often underperform LLM-based approaches on complex semantic modifications. This gap reflects a semantic transition bottleneck in projection-based ZS-CIR: endpoint-level matching can let the edit text act as a target-side attribute cue rather than grounding it as a source-conditioned semantic transition. We further show that adding semantic transition supervision to the same text adapter creates an endpoint--transition conflict between endpoint alignment and semantic transition alignment. To address this conflict, DeCIR decouples endpoint and transition learning. It constructs paired forward/reverse edit tuples from image-caption pairs, trains separate low-rank text adapter branches for endpoint alignment and semantic transition alignment, and merges them with Low-Rank Directional Merge (LRDM) into one deployable adapter. Extensive experiments on CIRR, CIRCO, FashionIQ, and GeneCIS demonstrate that DeCIR consistently improves projection-based ZS-CIR without increasing inference complexity.

In this work, we introduce GELATO (Geometry-preserving Embeddings via Locked Aligned TOwers), a novel approach to multimodal embedding models. We build on the VLM-style architecture, in which non-text encoders are adapted to produce input for a language model, which in turn generates embeddings for all varieties of input. We present the result: the jina-embeddings-v5-omni suite, a pair of models that encode text, image, audio, and video input into a single semantic embedding space. GELATO extends the two Jina Embeddings v5 Text models to support additional modality by adding encoders for images and audio. The backbone text embedding models and the added non-text modality encoders remain frozen. We only trained the connecting components, representing 0.35% of the total weights of the joint model. Training is therefore much more efficient than full-parameter retraining. Additionally, the language model remains effectively unaltered, producing exactly the same embeddings for text inputs as the Jina Embeddings v5 Text models. Our evaluations show that GELATO produces results that are competitive with the state-of-the-art, yielding nearly equal performance to larger multimodal embedding models.

Training multimodal large language models has long been limited by the scarcity of high-quality paired multimodal data. Recent studies show that the shared representation space of pretrained multimodal contrastive models can serve as a bridge, enabling models to perform multimodal training with unimodal data. However, the key premise of this paradigm remains insufficiently understood: can representations from different modalities be reliably interchanged? The core obstacle lies in the persistent Modality Gap in the shared space. In this work, we revisit the geometric nature of the modality gap. We find that modality representations already share compatible dominant semantic geometry. What truly hinders modality interchangeability is not a simple global shift, but an anisotropic residual structure concentrated along a small number of dominant directions. Based on this finding, we further propose the principle of anisotropic modality gap alignment: effective modality alignment should align with the target-modality distribution while preserving the semantic structure of the source modality. Guided by this principle, we propose an anisotropic geometric correction framework, AnisoAlign, for unpaired modality alignment. This framework leverages the internal geometric prior of the target modality and performs bounded correction on source-modality representations, thereby constructing substitute representations in the target modality. Experiments confirm its benefits in both geometric diagnostics and text-only MLLM training. Overall, this work recasts the modality gap from an empirical observation into a correctable, structured geometric phenomenon and provides a new representation alignment perspective for training multimodal models with unimodal data.

Health foundation models (FMs) learn useful representations from wearable sensors, but interpreting what they encode and transferring that knowledge across modalities after training remains difficult. We present a post-training framework that decomposes frozen embeddings into interpretable directions, referred to as symbols, and use these symbols to align the embedding spaces without retraining. We evaluate the framework on three FMs for photoplethysmography (PPG) and accelerometer data, independently pretrained on ~20M minutes of unlabeled data from ~172K participants, and analyzed on a held-out cohort of 30K subjects. We find that extracted symbols associate selectively with health conditions and physiological attributes, and these associations are partially shared across modalities and architectures. Cross-modal transfer via symbols retains more than 95% of in-domain performance, is nearly symmetric across domain directions, and saturates with limited paired data, together indicating that alignment recovers a shared low-dimensional subspace rich in physiological information. Overall, these results suggest that health FM embeddings contain an interpretable symbolic organization that is shared across modalities and supports cross-domain transfer without joint training.

Cameras and LiDAR degrade in rain, fog, and snow, while millimeter-wave radar remains largely unaffected. We align a radar encoder to frozen SigLIP vision embeddings and decode structured scene captions through a frozen vision-language model (VLM) with approximately 7M trainable parameters. On K-RADAR with held-out fog, light snow, and heavy snow sequences, all radar configurations outperform a camera baseline that collapses to over 90% hallucination. We identify a token-norm mismatch as the dominant failure mode when bridging radar to a frozen VLM and show that projector-output LayerNorm resolves it. Analysis of encoder complexity, caption format, and pooling strategy reveals tradeoffs that inform future radar-VLM pipeline design.

Robust embodied navigation relies on complementary sensory cues. However, high-quality and well-aligned multi-modal data is often difficult to obtain in practice. Training a monolithic model is also challenging as rich multi-modal inputs induce complex representations and substantially enlarge the policy space. Cross-modal collaboration among lightweight modality-specialized agents offers a scalable paradigm. It enables flexible deployment and parallel execution, while preserving the strength of each modality. In this paper, we propose \textbf{CRONA}, a Multi-Agent Reinforcement Learning (MARL) framework for \textbf{Cro}ss-Modal \textbf{Na}vigation. CRONA improves collaboration by leveraging control-relevant auxiliary beliefs and a centralized multi-modal critic with global state. Experiments on visual-acoustic navigation tasks show that multi-agent methods significantly improve performance and efficiency over single-agent baselines. We find that homogeneous collaboration with limited modalities is sufficient for short-range navigation under salient cues; heterogeneous collaboration among agents with complementary modalities is generally efficient and effective; and navigation in large, complex environments requires both richer multi-modal perception and increased model capacity.

Partially relevant video retrieval aims to retrieve untrimmed videos using text queries that describe only partial content. However, the inherent asymmetry between brief queries and rich video content inevitably introduces uncertainty into the retrieval process. In this setting, vague queries often induce semantic ambiguity across videos, a challenge that is further exacerbated by the sparse temporal supervision within videos, which fails to provide sufficient matching evidence. To address this, we propose Holmes, a hierarchical evidential learning framework that aggregates multi-granular cross-modal evidence to quantify and model uncertainty explicitly. At the inter-video level, similarity scores are interpreted as evidential support and modeled via a Dirichlet distribution. Based on the proposed three-fold principle, we perform fine-grained query identification, which then guides query-adaptive calibrated learning. At the intra-video level, to accumulate denser evidence, we formulate a soft query-clip alignment via flexible optimal transport with an adaptive dustbin, which alleviates sparse temporal supervision while suppressing spurious local responses. Extensive experiments demonstrate that Holmes outperforms state-of-the-art methods. Code is released at https://github.com/lijun2005/ICML26-Holmes.

Although self-supervised pretraining is expected to learn broadly transferable representations, its effectiveness across imaging modalities substantially different from the pretraining domain, and on complex tumor-segmentation tasks, remains understudied. Evaluating CT-pretrained transformers on MRI rectal cancer segmentation, we identified two interacting failure modes in CT-to-MRI transfer: (a) inefficient token usage caused by zero-padding to match pretrained input dimensions, and (b) ineffective feature adaptation. We investigated these vulnerabilities using two primary CT-pretrained hierarchical shifted-window transformer backbones, SMIT and Swin UNETR, together with VoCo as a large-scale-pretrained supporting benchmark; these models differ in pretraining objectives and datasets. Mechanistic analysis leveraged an attention dilution index (ADI), an entropy-based metric quantifying attention diverted toward uninformative padding tokens, and centered kernel alignment (CKA) to measure feature reuse during MRI adaptation. ADI increased with zero-padding, while high feature reuse did not necessarily translate to improved downstream accuracy. To mitigate these issues, we introduced two interventions: a tumor-aware augmentation strategy to expand tumor appearance heterogeneity coverage, and an anisotropic cropping strategy to restore token efficiency. Fine-tuning with these strategies on identical rectal MRI datasets yielded detection rates of 91.1% (225/247) and 88.7% (219/247) for the primary SMIT and Swin UNETR backbones, with the supporting VoCo benchmark reaching 90.3% (223/247), demonstrating significantly improved robustness under CT-to-MRI transfer. This study is among the first to examine when pretrained transformers fail to transfer across imaging modalities and demonstrates how targeted mitigation strategies can systematically overcome cross-modality transfer limitations.

Recent flow matching (FM) methods improve the few-shot adaptation of vision-language models, by modeling cross-modal alignment as a continuous multi-step flow. In this paper, we argue that existing FM methods are inherently constrained by incompatible geometric priors on pre-trained cross-modal features, resulting in suboptimal adaptation performance. We first analyze these methods from a polar decomposition perspective (i.e., radial and angular sub-manifolds). Under this new geometric view, we identify three overlooked limitations in them: 1) Angular dynamics distortion: The radial-angular coupling induces non-uniform speed on the angular sub-manifold, leading to regression training difficulty and extra truncation errors. 2) Radial dynamics neglect: Feature normalization discards modality confidence, failing to distinguish out-of-distribution and in-distribution data, and abandoning crucial radial dynamics. 3) Context-agnostic unconditional flow: Dataset-specific information loss during pre-trained cross-modal feature extraction remains unrecovered. To resolve these issues, we propose warped product flow matching (WP-FM), a unified Riemannian framework that reformulates alignment on a warped product manifold. Within this framework, we derive direct product flow matching (DP-FM) by introducing a constant-warping metric, which yields a decoupled cylindrical manifold (i.e., direct product manifold). DP-FM enables independent radial evolution and constant-speed angular geodesic transport, effectively eliminating angular dynamics distortion while preserving radial consistency. Meanwhile, we incorporate classifier-free guidance by conditioning the flow on the pre-trained VLMs' hidden states to inject missing dataset-specific information. Extensive results across 11 benchmarks have demonstrated that DP-FM achieves a new state-of-the-art for multi-step few-shot adaptation.

Image-Text-Matching (ITM) is one of the defacto methods of learning generalized representations from a large corpus in Vision and Language (VL). However, due to the weak association between the web-collected image-text pairs, models fail to show a fine-grained understanding of the combined semantics of these modalities. To address this issue we propose Hard Negative Captions (HNC): an automatically created dataset containing foiled hard negative captions for ITM training towards achieving fine-grained cross-modal comprehension in VL. Additionally, we provide a challenging manually-created test set for benchmarking models on a fine-grained cross-modal mismatch task with varying levels of compositional complexity. Our results show the effectiveness of training on HNC by improving the models' zero-shot capabilities in detecting mismatches on diagnostic tasks and performing robustly under noisy visual input scenarios. Also, we demonstrate that HNC models yield a comparable or better initialization for fine-tuning

Multimodal emotion recognition (MER) benefits from combining text, audio, and vision, yet standard fusion often fails when modalities conflict. Crucially, conflicts differ in resolvability: benign conflicts stem from missing, weak, or ambiguous cues and can be mitigated by cross-modal calibration, while severe conflicts arise from intrinsically contradictory (e.g., sarcasm) or misleading signals, for which forced fusion may amplify errors. Recognizing this, we propose Dual-Path Conflict Resolution (DCR), a unified framework that learns when to fuse and when to drop modalities. Path I (Affective Fusion Distiller, AFD) performs reverse distillation from audio/visual teachers to a textual student using temporally weighted class evidence, thereby enhancing representation-level calibration and improving fusion when alignment is beneficial. Path II (Affective Discernment Agent, ADA) formulates MER as a contextual bandit that selects among fusion and unimodal predictions based on a dual-view state and a calibration-aware reward, enabling decision-level arbitration under irreconcilable conflicts without requiring per-modality reliability labels. By taking into account the full multimodal context and coupling soft calibration with hard arbitration, DCR reconciles conflicts that can be aligned while bypassing misleading modalities when fusion is harmful. Across five benchmarks covering both dialogue-level and clip-level MER, DCR consistently outperforms competitive baselines or achieves highly competitive results. Further ablations, conflict-specific subset evaluation, and modality-selection analysis verify that AFD and ADA are complementary and jointly improve robust conflict-aware emotion recognition.

Multimodal visual object tracking can be divided into to several kinds of tasks (e.g. RGB and RGB+X tracking), based on the input modality. Existing methods often train separate models for each modality or rely on pretrained models to adapt to new modalities, which limits efficiency, scalability, and usability. Thus, we introduce OneTrackerV2, a unified multi-modal tracking framework that enables end-to-end training for any modality. We propose Meta Merger to embed multi-modal information into a unified space, allowing flexible modality fusion and robustness. We further introduce Dual Mixture-of-Experts (DMoE): T-MoE models spatio-temporal relations for tracking, while M-MoE embeds multi-modal knowledge, disentangling cross-modal dependencies and reducing feature conflicts. With a shared architecture, unified parameters, and a single end-to-end training, OneTrackerV2 achieves state-of-the-art performance across five RGB and RGB+X tracking tasks and 12 benchmarks, while maintaining high inference efficiency. Notably, even after model compression, OneTrackerV2 retains strong performance. Moreover, OneTrackerV2 demonstrates remarkable robustness under modality-missing scenarios.

Modality translation is inherently under-constrained, as multiple cross-modal mappings may yield the same marginals. Recent work has shown that diffusion bridges are effective for this task. However, most existing approaches rely on fully paired datasets, thereby imposing a single data-driven constraint. We propose a diffusion-bridge framework that characterizes the space of admissible solutions and restricts it via alignment constraints, treating paired supervision as an optional heuristic rather than a prerequisite. We validate our method on synthetic and real modality translation benchmarks across unpaired, semi-paired, and paired regimes, showing consistent performance across supervision levels. Notably, \textbf{it achieves near fully-paired quality with a substantial relaxation in pairing requirements, and remaining applicable in the unpaired regime}. These results highlight diffusion bridges as a flexible foundation for modality translation beyond fully paired data.

Recent advances in omni-modal large language models have enabled remarkable progress in joint vision-audio understanding. However, prevailing architectures rely on modality-specific encoders with a \emph{video-coarse, audio-dense} design -- sampling visual frames at 1--2 fps while processing audio waveforms at 25 fps -- resulting in systems that perceive video \emph{frame by frame, modality by modality} rather than holistically as humans do. Such a discrepancy leaves models with impoverished cross-modal interaction during encoding and an inability to capture fine-grained visual motion. To bridge this gap, we present \textbf{Omni-Encoder, a unified Transformer backbone designed to co-embed visual and audio signals at a symmetrical 25 fps} within a shared latent space. This architecture leverages three core innovations -- the Omni-Encoder Token Template, Omni-RoPE, and Temporal Window Shifting -- to effectively reconcile the dual challenges of modality disentanglement and computational efficiency. Experiments demonstrate that, compared to the modality-specific baseline Qwen2.5-Omni under the same input token budget to the LLM decoder, Omni-Encoder delivers substantial gains on visual continuous understanding tasks -- such as sign language recognition and fine-grained sports action analysis -- while maintaining competitive performance on established audio-visual benchmarks such as AVQA and Speaker Identification and Localization. These results suggest that unified omnivorous encoding offers a promising direction for building omni-modal models that more closely reflect the integrated nature of human perception.

Although multi-modal learning has advanced point cloud completion, the theoretical mechanisms remain unclear. Recent works attribute success to the connection between modalities, yet we identify that standard hard projection severs this connection: projecting a sparse point cloud onto the image plane yields an extremely sparse support, which hinders visual prior propagation, a failure mode we term Cross-Modal Entropy Collapse. To address this practical limitation, we propose SplAttN, which replaces hard projection with Differentiable Gaussian Splatting to produce a dense, continuous image-plane representation. By reformulating projection as continuous density estimation, SplAttN avoids collapsed sparse support, facilitates gradient flow, and improves cross-modal connection learnability. Extensive experiments show that SplAttN achieves state-of-the-art performance on PCN and ShapeNet-55/34. Crucially, we utilize the real-world KITTI benchmark as a stress test for multi-modal reliance. Counter-factual evaluation reveals that while baselines degenerate into unimodal template retrievers insensitive to visual removal, SplAttN maintains a robust dependency on visual cues, validating that our method establishes an effective cross-modal connection. Code is available at https://github.com/zay002/SplAttN.

Shared embedding spaces are widely used for multimodal search and data curation. In practice, two problems often limit how well this works. First, embeddings can reflect modality more than meaning, so examples cluster by input type even when the underlying content matches. Second, the paired supervision used to train these spaces is often noisy. When we blend many heterogeneous, human-labeled datasets, these issues reinforce each other and degrade cross-modal retrieval. We present a framework that improves alignment by acting on both the training pairs and the embedding model. Symmetric Nucleus Subsampling (SNS) refines training pairs by trimming raw inputs and annotations to the portions that best support each other. Expert Embedding Engine (EEE) combines complementary embedding experts using a learned projection network, together with a bias-aware objective that reduces modality-driven separation in the embedding space. We demonstrate that this approach collapses the modality gap by over 90% on average vs base embedding experts and is a strong data curator, with datablends from our method outperforming stratified sampling and traditional curation baselines in downstream model performance.

In this paper, we present \textbf{Gen}erative \textbf{L}anguage-\textbf{I}mage \textbf{P}re-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) \textbf{Simplicity}: a single transformer jointly models visual and textual tokens; (2) \textbf{Scalability}: it scales effectively with both data and model size; and (3) \textbf{Performance}: it achieves competitive or superior results across diverse multimodal benchmarks. Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding, making it a strong foundation for vision encoders in MLLMs.

Federated Multimodal Learning (FML) trains multimodal models across decentralized clients while keeping their image-text pairs private. However, joint embedding training entangles forgotten knowledge across both modalities and client gradient subspaces, hindering federated unlearning. Previous federated unlearning approaches neither sever the cross-modal reconstruction channel mediated by bilinear coupling nor separate forget-exclusive update directions from those shared with retained clients. We identify an Anchor Principle for federated multimodal contrastive unlearning: forgotten alignments persist through three residual anchors arising from bilinear cross-modal coupling, principal-angle subspace entanglement, and continued federated updates. At the modality level, we show that bilateral displacement of both visual and language branches closes the cross-modal reconstruction channel. Correspondingly, our method addresses subspace entanglement through Cosine--Sine decomposition of client-update subspaces, isolating forget-exclusive directions from retain support. Moreover, we propose a direction-selective Forget Lock that bounds residual drift across rounds. Combining these strategies, we present EASE, an Entanglement-Aware Subspace Excision framework that closes all three anchor channels under a unified design. EASE demonstrates consistent superiority across multiple datasets and unlearning scenarios, for instance, matching the retrain reference to within 0.2 and 4.2 R@1 points on the forget and retain sides under client unlearning on Flickr30K with CLIP-B/32.

Unified multimodal understanding and generation models enable richer human-AI interaction. Yet jointly customizing a character's persona, dialogue style, and visual identity while maintaining output consistency across modalities remains largely unexplored. To mitigate this gap, we introduce a new task, Customized Multimodal Role-Play (CMRP). We construct the RoleScape-20 dataset comprising 20 characters, including training and evaluation data that cover persona, stylistic descriptions, visual/expressive cues, and text-image interactions. Building on a unified model, we devise UniCharacter, a two-stage training framework containing Unified Supervised Finetuning (Unified-SFT) and character-specific group relative policy optimization (Character-GRPO). Given only 10 images plus corresponding interaction examples, the model acquires the target character and exhibits coherent persona, style, and visual identity in both generated text and images. This process takes about 100 GPU hours. Experiments on the RoleScape-20 dataset show that the proposed method substantially outperforms prior approaches. Ablation studies further validate the effectiveness of our cross-modal consistency design and few-shot customization strategy. We argue that CMRP, coupled with unified modeling, provides a basis for next-generation characterful and immersive interactive agents.

Centralized multimodal learning commonly compresses language, acoustic, and visual signals into a single fused representation for prediction. While effective, this paradigm suffers from two limitations: modality dominance, where optimization gravitates towards the path of least resistance, ignoring weaker but informative modalities, and spurious modality coupling, where models overfit to incidental cross-modal correlations. To address these, we propose Group Cognition Learning (GCL), a governed collaboration paradigm that applies a two-stage protocol after modality-specific encoding. In Stage 1 (Selective Interaction), a Routing Agent proposes directed interaction routes, and an Auditing Agent assigns sample-wise gates to emphasize exchanges that yield positive marginal predictive gain while suppressing redundant coupling. In Stage 2 (Consensus Formation), a Public-Factor Agent maintains an explicit shared factor, and an Aggregation Agent produces the final prediction through contribution-aware weighting while keeping each modality representation as a specialization channel. Extensive experiments on CMU-MOSI, CMU-MOSEI, and MIntRec demonstrate that GCL mitigates dominance and coupling, establishing state-of-the-art results across both regression and classification benchmarks. Analysis experiments further demonstrate the effectiveness of the design.

Frozen Gemma 4 31B weights pretrained exclusively on text, unmodified, transfer through a thin trainable interface to non-text modalities the substrate has never processed. On the L24--L29 slice (192 attention heads), an English-text TxtCopy attention probe (95 sentences) and per-head ablation impact on four non-language token-pattern tasks (binary copy, associative recall, 1D cellular automaton Rule 90, binary addition) jointly classify four heads -- L26.28, L27.28, L27.2, L27.3 -- as top-tier on both signals. The slice-level joint coincidence is significant under hypergeometric null ($P = 0.0013$, $N=192$, $K=38$, $n=4$) and survives multiplicity-aware permutation tests ($P_{V4} = 0.013$). Pretrained Gemma L26 reaches 60.22% on OGBench cube-double-play-task1 vs ~1% for random-init Gemma ($+59$pt at $n=3$); a FrozenRandom-GPT2 control with correct $1/\sqrt{d_k}$ scaling also fails. Head-level causal validation: zeroing L26.28 in the trained cube-task1 IQL agent drops success $63.3\% \to 10.0\%$ vs $46.7\%$ for a layer-matched low-TxtCopy negative control ($3.2\times$ specificity at $n=30$; $n=5$ paired-$t$ $p=0.039$). A full L26 sweep places L26.28 at rank 4 of 32. Honest negatives: within-L26 Spearman $ρ(\text{TxtCopy, drop}) = +0.37$ (opposite of within-layer causal reading); single-head activation patching does not transfer the matching variable; the 4 named heads alone do not suffice on any task; Walker2d-DT and scene-task1 recruit L24 outside the named slice and show null head-ablation specificity. We frame the contribution as a cross-distribution importance fingerprint at the slice level plus head-level causal evidence on one cross-modality target.

Event cameras provide several unique advantages over standard frame-based sensors, including high temporal resolution, low latency, and robustness to extreme lighting. However, existing learning-based approaches for event processing are typically confined to narrow, task-specific silos and lack the ability to generalize across modalities. We address this gap with REALM, a cross-modal framework that learns an RGB- and Event-Aligned Latent Manifold by projecting event representations into the pretrained latent space of RGB foundation models. Instead of task-specific training, we leverage low-rank adaptation (LoRA) to bridge the modality gap, effectively unlocking the geometric and semantic priors of frozen RGB backbones for asynchronous event streams. We demonstrate that REALM effectively maps events into the ViT-based foundation latent space. Our method performs downstream tasks, such as depth estimation and semantic segmentation, by simply transferring linear heads trained on the RGB teacher. Most significantly, REALM enables the direct, zero-shot application of complex, frozen image-trained decoders, such as MASt3R, to raw event data. We demonstrate state-of-the-art performance in wide-baseline feature matching, significantly outperforming specialized architectures. Code and models are available at https://papers.starslab.ca/realm/.

The hubness problem, in which hub embeddings are close to many unrelated examples, occurs often in high-dimensional embedding spaces and may pose a practical threat for purposes such as information retrieval and automatic evaluation metrics. In particular, since cross-modal similarity between text and images cannot be calculated by direct comparisons, such as string matching, cross-modal encoders that project different modalities into a shared space are helpful for various cross-modal applications, and thus, the existence of hubs may pose practical threats. To reveal the vulnerabilities of cross-modal encoders, we propose a method for identifying the hub embedding and its corresponding hub text. Experiments on image captioning evaluation in MSCOCO and nocaps along with image-to-text retrieval tasks in MSCOCO and Flickr30k showed that our method can identify a single hub text that unreasonably achieves comparable or higher similarity scores than human-written reference captions in many images, thereby revealing the vulnerabilities in cross-modal encoders.

Vision-language models have shown strong performance, but they often generalize poorly to specialized domains. While semi-supervised vision-language learning mitigates this limitation by leveraging a small set of labeled image-text pairs together with abundant unlabeled images, existing methods remain fundamentally pairwise and fail to model the global structure of multimodal representation manifolds. Existing topology-based alignment methods rely on persistence diagram matching, which neither guarantees geometric alignment nor utilizes the image-text pairing information central to vision-language learning. We propose Topology-Aware Multimodal Representation Alignment (ToMA), a framework that uses persistent homology to identify topologically salient edges and aligns them across modalities through available cross-modal correspondences. ToMA leverages both H_0-death edges and lightweight H_1-birth edges, allowing it to capture both connectivity and cycle structure without constructing 2-simplices. Experiments show that ToMA yields stable gains, with clear improvements on remote sensing and modest but consistent benefits on fashion retrieval. Additional analysis shows that ToMA is more stable than alternative topology-based objectives and that lightweight H_1-birth edges provide useful higher-order structural signals.

Accurate whole-cell and nuclear segmentation is essential for precision pathology and spatial omics, yet routine hematoxylin and eosin (H&E) staining provides limited cytoplasmic contrast, restricting analyses to nuclei. Multiplex immunofluorescence (mIF) facilitates precise whole-cell delineation but remains constrained by cost and accessibility. We introduce VitaminP, a cross-modal learning framework enabling whole cell segmentation from H&E images. By learning from paired H&E-mIF data, VitaminP transfers molecular boundary information from mIF to overcome cytoplasmic contrast in H&E, establishing cross-modal supervision as a general strategy for recovering missing biological structure. We train VitaminP on 14 public datasets covering 34 cancer types and over 7 million instances, integrating publicly available labels with extensive annotations generated in this study, forming one of the largest resources for segmentation. VitaminP outperforms four state-of-the-art methods and generalizes to unseen datasets, including an in-house dataset spanning 24 rare cancer types. We further developed VitaminPScope, an open-source platform providing an interface for scalable inference and enabling broad adoption.

Joint audio-video generation models have shown that unified generation yields stronger cross-modal coherence than cascaded approaches. However, existing models couple modalities throughout denoising via pervasive attention, treating high-level semantics and low-level details in a fully entangled manner. This is suboptimal for talking head synthesis: while audio and facial motion are semantically correlated, their low-level realizations (acoustic signals and visual textures) follow distinct rendering processes. Enforcing joint modeling across all levels causes unnecessary entanglement and reduces efficiency. We propose Talker-T2AV, an autoregressive diffusion framework where high-level cross-modal modeling occurs in a shared backbone, while low-level refinement uses modality-specific decoders. A shared autoregressive language model jointly reasons over audio and video in a unified patch-level token space. Two lightweight diffusion transformer heads decode the hidden states into frame-level audio and video latents. Experiments on talking portrait benchmarks show Talker-T2AV outperforms dual-branch baselines in lip-sync accuracy, video quality, and audio quality, achieving stronger cross-modal consistency than cascaded pipelines.

Federated cross-modal retrieval faces severe challenges from heterogeneous client data, particularly non-IID semantic distributions and missing modalities. Under such heterogeneity, a single global model is often insufficient to capture both shared cross-modal knowledge and client-specific characteristics. We propose RCSR, a personalization-friendly federated framework that integrates prototype anchoring, retrieval-centric semantic routing, and optional client-specific adapters. Built on a frozen CLIP backbone, RCSR leverages lightweight shared adapters for global knowledge transfer while supporting efficient local personalization. Prototype anchoring helps unimodal clients align with global cross-modal semantics, and a server-side semantic router adaptively assigns aggregation weights based on retrieval consistency to mitigate alignment drift during heterogeneous updates. Extensive experiments on MS-COCO, Flickr30K, and other benchmarks show that RCSR consistently improves global retrieval accuracy and training stability, while further enhancing client-level retrieval performance, especially for clients with incomplete modalities. Code is available at https://github.com/RezinChow/RCSR-Retrieval-Centric-Semantic-Routing.