New Decoding Method Improves Efficiency of Diffusion-Based Multimodal AI Models
Researchers have introduced Visual-Redundancy-Controlled Decoding (VRCD), a training-free inference-time method designed to improve how diffusion-based multimodal large language models select tokens during parallel decoding. Current confidence-based decoding strategies tend to commit tokens that share overlapping visual grounding, leaving later decoding steps with less complementary visual information. The proposed method achieves relative accuracy gains of up to 18.8% on the M³CoT benchmark and 6.9% on MMBench over standard confidence-based decoding.
Diffusion-based multimodal large language models (dMLLMs) generate outputs by iteratively predicting tokens at multiple masked positions in parallel, requiring the model to decide both which predictions are reliable and which positions should be committed together to inform subsequent steps. The authors identify a structural limitation in existing confidence-based decoding: high-confidence tokens selected within the same step often rely on overlapping visual regions, a phenomenon they term visual redundancy, which degrades the quality of visual grounding available for later decoding. To measure this effect, the team introduces the Visual Redundancy Index (VRI), a metric quantifying the overlap in visual grounding among tokens committed in parallel. Their proposed solution, VRCD, uses token-to-image attention scores to prioritize visually complementary token positions rather than selecting purely by confidence. The method requires no additional training and introduces only modest runtime overhead. Experiments across diverse multimodal benchmarks show VRCD consistently reduces both visual redundancy and remaining-position entropy, with the largest gains observed in longer decoding sequences. The work is available as a preprint on arXiv, with code publicly released.
What's missing
As a preprint, this work has not yet undergone formal peer review. The paper does not report results on a broad range of dMLLM architectures beyond those tested, leaving open the question of how well VRCD generalizes to other model families. The computational cost of computing token-to-image attention at inference time is described only as 'modest' without detailed absolute benchmarks. It is also unclear how performance scales with different image resolutions or modality combinations beyond those evaluated.
What different sources said
- arXiv cs.LGCenter
Visual-Redundancy-Controlled Parallel Decoding for Diffusion-Based Multimodal Large Language Models
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.