EntropyScan detects backdoored LVLMs by quantifying structural anomalies in visual attention distributions on benign samples via Tsallis entropy and reference-anchored Z-score normalization.
In: ICLR (2021)
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
fields
cs.CV 8verdicts
UNVERDICTED 8representative citing papers
PromptCCD uses Gaussian Mixture Prompts for global class prototypes and PromptCCD++ adds part-level prompt pools for finer representations in continual category discovery from unlabeled streams.
HEE is a training-free, model-agnostic method for high-resolution visual perception in MLLMs using hierarchical entity exploration with dual scoring, detection, clustering, and backtracking.
Splash partitions MLLM parameters into dormant and critical subspaces via significance quantification, updating only the dormant subspace for tactile alignment while preserving general capabilities and achieving SOTA on visuo-tactile benchmarks.
A new end-to-end training scheme for visual attribution maps that optimizes deletion and insertion metrics directly via differentiable ranking relaxation instead of surrogate objectives.
OBBSeg segments irregular medical lesions from oriented bounding-box labels via a Mask-to-OBB loss and prompt modules, claiming near fully-supervised accuracy across 13 datasets and 5 modalities.
Discarding visual guidance from vision-language models and using language embeddings as the primary source of domain invariance via an information bottleneck yields state-of-the-art domain generalization performance.
Weak-to-strong knowledge distillation applied early and then turned off accelerates convergence to target performance in visual learning tasks by factors of 1.7-4.8x.
citing papers explorer
-
EntropyScan: Towards Model-level Backdoor Detection in LVLMs via Visual Attention Entropy
EntropyScan detects backdoored LVLMs by quantifying structural anomalies in visual attention distributions on benign samples via Tsallis entropy and reference-anchored Z-score normalization.
-
Effective Prompt Pool Learning for Continual Category Discovery
PromptCCD uses Gaussian Mixture Prompts for global class prototypes and PromptCCD++ adds part-level prompt pools for finer representations in continual category discovery from unlabeled streams.
-
Towards High-Resolution Visual Perception via Hierarchical Entity Exploration
HEE is a training-free, model-agnostic method for high-resolution visual perception in MLLMs using hierarchical entity exploration with dual scoring, detection, clustering, and backtracking.
-
Wake up for Touch! Mask-isolated Tactile Alignment Learning in MLLMs
Splash partitions MLLM parameters into dormant and critical subspaces via significance quantification, updating only the dormant subspace for tactile alignment while preserving general capabilities and achieving SOTA on visuo-tactile benchmarks.
-
Learn to Rank: Visual Attribution by Learning Importance Ranking
A new end-to-end training scheme for visual attribution maps that optimizes deletion and insertion metrics directly via differentiable ranking relaxation instead of surrogate objectives.
-
OBBSeg: Irregular Lesion Segmentation under Oriented Bounding Box Annotations
OBBSeg segments irregular medical lesions from oriented bounding-box labels via a Mask-to-OBB loss and prompt modules, claiming near fully-supervised accuracy across 13 datasets and 5 modalities.
-
Domain Generalization via Text-Anchored Information Bottleneck
Discarding visual guidance from vision-language models and using language embeddings as the primary source of domain invariance via an information bottleneck yields state-of-the-art domain generalization performance.
-
Weak-to-Strong Knowledge Distillation Accelerates Visual Learning
Weak-to-strong knowledge distillation applied early and then turned off accelerates convergence to target performance in visual learning tasks by factors of 1.7-4.8x.