Rapid progress in large vision-language models (LVLMs) has achieved unprecedented performance in vision-language tasks. However, due to the strong prior of large language models (LLMs) and misaligned attention across modalities, LVLMs often generate outputs inconsistent with visual content - termed hallucination. To address this, we propose Scalpel, a method that reduces hallucination by refining attention activation distributions toward more credible regions. Scalpel predicts trusted attention directions for each head in Transformer layers during inference and adjusts activations accordingly. It employs a Gaussian mixture model to capture multi-peak distributions of attention in trust and hallucination manifolds, and uses entropic optimal transport (equivalent to Schrödinger bridge problem) to map Gaussian components precisely. During mitigation, Scalpel dynamically adjusts intervention strength and direction based on component membership and mapping relationships between hallucination and trust activations. Extensive experiments across multiple datasets and benchmarks demonstrate that Scalpel effectively mitigates hallucinations, outperforming previous methods and achieving state-of-the-art performance. Moreover, Scalpel is model-and data-agnostic, requiring no additional computation, only a single decoding step.

When a site is damaged by a major disaster, there is a risk of large-scale data loss within the premises or isolation from the Internet due to network disruption. To achieve both data durability and availability from the disaster area, Risk-aware Data Protection (RDP) has been proposed to distribute redundant data to neighboring sites based on the disaster risk. However, RDP has only been studied using a method called replication, while Erasure Coding (EC) applied to RDP is still unexplored. Therefore, the challenge of this study is to quantitatively compare the effectiveness of these methods in RDP. For this purpose, we formulate evaluation indicators to measure the remaining data ratio after a disaster; and then propose how to apply several existing combinatorial optimization algorithms to RDP. Based on the results of experiments, we determined that EC performs better than replication if the probability of site damage is small.

In this paper, we propose the first verification method for attention robustness, i.e., the local robustness of the changes in the saliency-map against combinations of semantic perturbations. Specmcally, our method determines the range of the perturbation parameters (e.g., the amount of brightness change) that maintains the difference between the actual saliencymap change and the expected saliency-map change below a given threshold value. Our method is based on linear activation region traversals, focusing on the outermost boundary of attention robustness for scalability on larger deep neural networks.