§ Multimodal

Bilingual LMs capture L1-driven cross-linguistic influence in L2 reading, validated on 12-language eye-tracking cohorts.

CRANE enables post-hoc counterfactual explanations for multimodal classifiers via nearest-neighbor retrieval, agnostic to architecture.

Overcooked-inspired benchmark for evaluating vision-language agent coordination on multi-player task planning in Unreal Engine.

Motion-guided VLM for efficient video anomaly detection using temporal gradient magnitude to trigger inference only when needed.

Multilingual benchmark of 9 prompting strategies across 4 judges, 27 languages, 13 datasets reveals LLM-as-Judge robustness varies by model.

New benchmark for video-grounded research agents that synthesize multimodal evidence and generate interactive reports linked to video moments.

PRISM: method for recognizing procedural mistakes in videos using HMM-grounded semantic state mapping and step-order validation.

DRAFT: method reconstructing 3D room layouts from fMRI signals and camera trajectories without visual frames.

MQAScore evaluates text-to-music alignment via large audio language model on five atomic dimensions (genre, instrument, mood, vocal, usage).

RoSA framework addresses saliency collapse in long-text image-text alignment by decomposing pairs to balance salient and non-salient context weighting.

EEG-Captioner dataset with 160K signal-grounded QA pairs evaluates whether EEG-to-language models learn actual signal properties vs. generating plausible text.

MHG-Mem, a multimodal hypergraph memory for long-term agent recall that preserves visual evidence while tracking entity state evolution across sessions.

CulturalRiddles, a benchmark of 62 culturally-grounded riddles across 45 countries and 51 languages, evaluates LLM understanding of non-factual cultural reasoning.

First English-Mizo parallel corpus of Indian court judgments (13,168 pairs) from Gauhati High Court and Supreme Court.

Model co-generates video and 3D skeletal motion for arbitrary character rigs, enabling playback in 3D scenes beyond pixel-level video synthesis.

UniWALM unified world-action-language model learns video, action, and language jointly for embodied robot control with memory-dependent behavior.

MindView benchmark evaluates generative multi-view spatial intelligence across 1,500 samples with tasks spanning geometry and semantic reasoning.

NarraSpace multimodal system links speech and visual scene elements for dementia classification by grounding narrative explanations in shared picture evidence.

CoRAL hybrid framework pairs diffusion planner with autoregressive executor for parallel multimodal reasoning, reducing latency via asymmetric orchestration.

Residual-space adaptation enables continual personalization in diffusion models, learning sequential concepts without catastrophic forgetting.

AnomalyAgent ensemble framework uses agentic fusion of vision-language models for anomaly detection with adaptive weighting.

Multimodal deep learning predicts laboratory values (NT-proBNP, glucose) from ECG recordings and patient metadata.

GPT-6-Astra adds tactile grounding to multimodal models, outperforming specialized tactile-vision-language policies on contact-rich manipulation.

Vision-language models transfer curator preferences to student models through near-duplicate image selection without explicit target prompts.

HypTI applies hyperbolic geometry to multimodal feature fusion for dementia detection from clinical and neuroimaging data.

Post-training phases for multimodal models on Japanese financial documents interact non-additively; math distillation overwrites domain knowledge while domain tuning repairs prior gains.

Benchmark study of associative creativity in MLLMs vs. humans on abstract visual stimuli with controlled naming difficulty; evaluates originality and effectiveness.

Vision-language model matches acoustic patterns to cavitation states in industrial pipelines using operating-condition-aware semantic descriptions.

SciDiagBench provides type-specific evaluation of AI-generated scientific diagrams, aligning VLM assessment with human peer-review criteria.

SLM-SS applies speech language models to multi-talker separation via token-based generation, optimizing for intelligibility over waveform metrics.

Vision-language models improve counting accuracy on large sets through agent-based epistemic actions (marking, pointing) without retraining.

World2HOSI orchestrates multi-expert models to generate long-horizon human-object-scene interactions from 3DGS scenes.

Halcyon achieves state-of-art text-to-image generation without language pretraining, using 18-275× less compute than LlamaGen and JiT-T2I on GPIC dataset.

OpenVLHarness framework enables multimodal LLMs to use external tools and retain evidence across reasoning steps for visual agentic tasks.

Mise-en-Video enables fine-grained 3D control for multi-shot video generation with inter-shot consistency via spatial staging and neural rendering.

IndustrialCAD-1400 benchmark evaluates text-to-CAD models on 1,400 detailed industrial assembly requests with 17+ component specifications and numeric constraints.

DialogMem benchmark evaluates embodied agents integrating dialogue and visual memory across 142 episodes for object navigation and QA in 3D environments.

Paired-residual subspace removal improves multilingual text embeddings from frozen LLMs by disentangling language variation.

VocalGrad benchmark for isolating acoustic perception failures in large audio language models, distinguishing paralinguistic understanding from reasoning.

Comprehensive evaluation of Qwen3-TTS zero-shot voice cloning across interfaces, domains (read/meeting/emotional/ATC speech), and failure modes vs. XTTS-v2 and F5-TTS.

LILIE uses vision-language models with normalizing flows to supervise visual generation via caption recovery, replacing pixel-space loss.

EEG decoder analysis across Llama-3.1-8B layers reveals passage-level context structure in silent reading aligns with deep layer embeddings.

MobileInject reveals cross-modal injection vulnerabilities in MLLM-based mobile agents through adversarial multimodal content embedding.

Spatial-TRM uses recursive refinement and frozen VLM features to improve spatial context prediction and region grounding in natural images.

Analysis and fix for token repetition failures in diffusion-based multimodal LLMs using information flow tracing and cache optimization.

Counterfactual pairing cycles benchmark method isolates relational OOD errors in vision-language models by controlling caption style and image statistics.

MAS-Bai uses linguistically informed multi-agent routing for Bowen-script Bai-Chinese translation, addressing phonetic character borrowing hallucinations.

OmniVerifier-C1 decomposes compositional rules into fine-grained components with pointwise scoring for robust multimodal visual verification and world modeling.

ACEdit stabilizes multi-turn image editing by training on accumulated distribution shifts from model's own prior outputs.

Time series language models benefit from complementary dual representations: line plots for absolute values, coordinate frames for variations.