AVIS: New Method Optimizes Inference Efficiency for Vision-Language Models
Researchers have introduced AVIS (Adaptive Visual Inference Scaling), a lightweight policy that simultaneously adapts visual context and reasoning compute per query in Vision-Language Models. Existing approaches typically optimize either visual context or reasoning search independently, leaving joint compute allocation underexplored. AVIS addresses a key practical barrier to deploying large VLMs by improving accuracy-compute trade-offs without requiring additional training.
A team of researchers has proposed Adaptive Visual Inference Scaling (AVIS), a system designed to reduce the prohibitive inference costs associated with chain-of-thought prompting and test-time scaling in Vision-Language Models (VLMs). AVIS operates along two coupled axes: Visual Context Scaling (VCS), which governs how much visual evidence is fed to the language model, and Visual Reasoning Scaling (VRS), which controls the extent of inference-time reasoning search. For VCS, AVIS employs Key Diversity Visual (KDV) pruning, a training-free O(N) algorithm that removes redundant visual tokens before prefilling. For VRS, it uses adaptive self-consistency, leveraging a learned difficulty predictor to determine the appropriate number of reasoning rollouts per query. The method is compatible with shared-prefill inference, allowing all reasoning rollouts to reuse a single prefilling pass and KV cache, which keeps latency low. Evaluations across image and video reasoning benchmarks show AVIS outperforms both VCS-only and VRS-only baselines on the accuracy-compute trade-off, and remains effective on top of reinforcement learning post-trained VLMs.
What's missing
The paper does not report results from peer review; as an arXiv preprint, findings have not yet been independently validated. Key open questions include how KDV pruning performs across highly diverse visual domains, whether the learned difficulty predictor generalizes beyond the benchmarks tested, and the sensitivity of AVIS to the quality of the underlying VLM backbone.
What different sources said
- arXiv cs.AICenter
AVIS: Adaptive Test-Time Scaling for Vision-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.