From Uncertainty to Trust: Enhancing Reliability in Vision-Language Models with Uncertainty-Guided Dropout Decoding

Large vision-language models (LVLMs) demonstrate remarkable capabilities in multimodal tasks but are prone to misinterpreting visual inputs, often resulting in hallucinations and unreliable outputs. To address these challenges, we propose Dropout Decoding, a novel inference-time approach that quantifies the uncertainty of visual tokens and selectively masks uncertain tokens to improve decoding. Our method measures the uncertainty of each visual token by projecting it onto the text space and decomposing it into aleatoric and epistemic components. Specifically, we focus on epistemic uncertainty, which captures perception-related errors more effectively. Inspired by dropout regularization, we introduce uncertainty-guided token dropout, which applies the dropout principle to input visual tokens instead of model parameters, and during inference rather than training. By aggregating predictions from an ensemble of masked decoding contexts, Dropout Decoding robustly mitigates errors arising from visual token misinterpretations. Evaluations on benchmarks including CHAIR, THRONE, and MMBench demonstrate that Dropout Decoding significantly reduces object hallucinations (OH) and enhances both reliability and quality of LVLM outputs across diverse visual contexts.

Toward Robust and Accurate Adversarial Camouflage Generation against Vehicle Detectors

Adversarial camouflage is a widely used physical attack against vehicle detectors for its superiority in multi-view attack performance. One promising approach involves using differentiable neural renderers to facilitate adversarial camouflage optimization through gradient back-propagation. However, existing methods often struggle to capture environmental characteristics during the rendering process or produce adversarial textures that can precisely map to the target vehicle. Moreover, these approaches neglect diverse weather conditions, reducing the efficacy of generated camouflage across varying weather scenarios. To tackle these challenges, we propose a robust and accurate camouflage generation method, namely RAUCA. The core of RAUCA is a novel neural rendering component, End-to-End Neural Renderer Plus (E2E-NRP), which can accurately optimize and project vehicle textures and render images with environmental characteristics such as lighting and weather. In addition, we integrate a multi-weather dataset for camouflage generation, leveraging the E2E-NRP to enhance the attack robustness. Experimental results on six popular object detectors show that RAUCA-final outperforms existing methods in both simulation and real-world settings.

"It's a conversation, not a quiz": A Risk Taxonomy and Reflection Tool for LLM Adoption in Public Health

Recent breakthroughs in large language models (LLMs) have generated both interest and concern about their potential adoption as accessible information sources or communication tools across different domains. In public health -- where stakes are high and impacts extend across populations -- adopting LLMs poses unique challenges that require thorough evaluation. However, structured approaches for assessing potential risks in public health remain under-explored. To address this gap, we conducted focus groups with health professionals and health issue experiencers to unpack their concerns, situated across three distinct and critical public health issues that demand high-quality information: vaccines, opioid use disorder, and intimate partner violence. We synthesize participants' perspectives into a risk taxonomy, distinguishing and contextualizing the potential harms LLMs may introduce when positioned alongside traditional health communication. This taxonomy highlights four dimensions of risk in individual behaviors, human-centered care, information ecosystem, and technology accountability. For each dimension, we discuss specific risks and example reflection questions to help practitioners adopt a risk-reflexive approach. This work offers a shared vocabulary and reflection tool for experts in both computing and public health to collaboratively anticipate, evaluate, and mitigate risks in deciding when to employ LLM capabilities (or not) and how to mitigate harm when they are used.

SocialGPT: Prompting LLMs for Social Relation Reasoning via Greedy Segment Optimization

Social relation reasoning aims to identify relation categories such as friends, spouses, and colleagues from images. While current methods adopt the paradigm of training a dedicated network end-to-end using labeled image data, they are limited in terms of generalizability and interpretability. To address these issues, we first present a simple yet well-crafted framework named {\name}, which combines the perception capability of Vision Foundation Models (VFMs) and the reasoning capability of Large Language Models (LLMs) within a modular framework, providing a strong baseline for social relation recognition. Specifically, we instruct VFMs to translate image content into a textual social story, and then utilize LLMs for text-based reasoning. {\name} introduces systematic design principles to adapt VFMs and LLMs separately and bridge their gaps. Without additional model training, it achieves competitive zero-shot results on two databases while offering interpretable answers, as LLMs can generate language-based explanations for the decisions. The manual prompt design process for LLMs at the reasoning phase is tedious and an automated prompt optimization method is desired. As we essentially convert a visual classification task into a generative task of LLMs, automatic prompt optimization encounters a unique long prompt optimization issue. To address this issue, we further propose the Greedy Segment Prompt Optimization (GSPO), which performs a greedy search by utilizing gradient information at the segment level. Experimental results show that GSPO significantly improves performance, and our method also generalizes to different image styles. The code is available at https://github.com/Mengzibin/SocialGPT.

CNCA: Toward Customizable and Natural Generation of Adversarial Camouflage for Vehicle Detectors

Prior works on physical adversarial camouflage against vehicle detectors mainly focus on the effectiveness and robustness of the attack. The current most successful methods optimize 3D vehicle texture at a pixel level. However, this results in conspicuous and attention-grabbing patterns in the generated camouflage, which humans can easily identify. To address this issue, we propose a Customizable and Natural Camouflage Attack (CNCA) method by leveraging an off-the-shelf pre-trained diffusion model. By sampling the optimal texture image from the diffusion model with a user-specific text prompt, our method can generate natural and customizable adversarial camouflage while maintaining high attack performance. With extensive experiments on the digital and physical worlds and user studies, the results demonstrate that our proposed method can generate significantly more natural-looking camouflage than the state-of-the-art baselines while achieving competitive attack performance. Our code is available at \href{https://anonymous.4open.science/r/CNCA-1D54}{https://anonymous.4open.science/r/CNCA-1D54}

A Novel Hybrid Approach for Tornado Prediction in the United States: Kalman-Convolutional BiLSTM with Multi-Head Attention

Tornadoes are among the most intense atmospheric vortex phenomena and pose significant challenges for detection and forecasting. Conventional methods, which heavily depend on ground-based observations and radar data, are limited by issues such as decreased accuracy over greater distances and a high rate of false positives. To address these challenges, this study utilizes the Seamless Hybrid Scan Reflectivity (SHSR) dataset from the Multi-Radar Multi-Sensor (MRMS) system, which integrates data from multiple radar sources to enhance accuracy. A novel hybrid model, the Kalman-Convolutional BiLSTM with Multi-Head Attention, is introduced to improve dynamic state estimation and capture both spatial and temporal dependencies within the data. This model demonstrates superior performance in precision, recall, F1-Score, and accuracy compared to methods such as K-Nearest Neighbors (KNN) and LightGBM. The results highlight the considerable potential of advanced machine learning techniques to improve tornado prediction and reduce false alarm rates. Future research will focus on expanding datasets, exploring innovative model architectures, and incorporating large language models (LLMs) to provide deeper insights. This research introduces a novel model for tornado prediction, offering a robust framework for enhancing forecasting accuracy and public safety.

A Community-Centric Perspective for Characterizing and Detecting Anti-Asian Violence-Provoking Speech

Violence-provoking speech -- speech that implicitly or explicitly promotes violence against the members of the targeted community, contributed to a massive surge in anti-Asian crimes during the pandemic. While previous works have characterized and built tools for detecting other forms of harmful speech, like fear speech and hate speech, our work takes a community-centric approach to studying anti-Asian violence-provoking speech. Using data from ~420k Twitter posts spanning a 3-year duration (January 1, 2020 to February 1, 2023), we develop a codebook to characterize anti-Asian violence-provoking speech and collect a community-crowdsourced dataset to facilitate its large-scale detection using state-of-the-art classifiers. We contrast the capabilities of natural language processing classifiers, ranging from BERT-based to LLM-based classifiers, in detecting violence-provoking speech with their capabilities to detect anti-Asian hateful speech. In contrast to prior work that has demonstrated the effectiveness of such classifiers in detecting hateful speech ($F_1 = 0.89$), our work shows that accurate and reliable detection of violence-provoking speech is a challenging task ($F_1 = 0.69$). We discuss the implications of our findings, particularly the need for proactive interventions to support Asian communities during public health crises. The resources related to the study are available at https://claws-lab.github.io/violence-provoking-speech/.

MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation?

While text-to-image models like DALLE-3 and Stable Diffusion are rapidly proliferating, they often encounter challenges such as hallucination, bias, and the production of unsafe, low-quality output. To effectively address these issues, it is crucial to align these models with desired behaviors based on feedback from a multimodal judge. Despite their significance, current multimodal judges frequently undergo inadequate evaluation of their capabilities and limitations, potentially leading to misalignment and unsafe fine-tuning outcomes. To address this issue, we introduce MJ-Bench, a novel benchmark which incorporates a comprehensive preference dataset to evaluate multimodal judges in providing feedback for image generation models across four key perspectives: alignment, safety, image quality, and bias. Specifically, we evaluate a large variety of multimodal judges including smaller-sized CLIP-based scoring models, open-source VLMs (e.g. LLaVA family), and close-source VLMs (e.g. GPT-4o, Claude 3) on each decomposed subcategory of our preference dataset. Experiments reveal that close-source VLMs generally provide better feedback, with GPT-4o outperforming other judges in average. Compared with open-source VLMs, smaller-sized scoring models can provide better feedback regarding text-image alignment and image quality, while VLMs provide more accurate feedback regarding safety and generation bias due to their stronger reasoning capabilities. Further studies in feedback scale reveal that VLM judges can generally provide more accurate and stable feedback in natural language (Likert-scale) than numerical scales. Notably, human evaluations on end-to-end fine-tuned models using separate feedback from these multimodal judges provide similar conclusions, further confirming the effectiveness of MJ-Bench. All data, code, models are available at https://huggingface.co/MJ-Bench.

InfRS: Incremental Few-Shot Object Detection in Remote Sensing Images

Recently, the field of few-shot detection within remote sensing imagery has witnessed significant advancements. Despite these progresses, the capacity for continuous conceptual learning still poses a significant challenge to existing methodologies. In this paper, we explore the intricate task of incremental few-shot object detection in remote sensing images. We introduce a pioneering fine-tuningbased technique, termed InfRS, designed to facilitate the incremental learning of novel classes using a restricted set of examples, while concurrently preserving the performance on established base classes without the need to revisit previous datasets. Specifically, we pretrain the model using abundant data from base classes and then generate a set of class-wise prototypes that represent the intrinsic characteristics of the data. In the incremental learning stage, we introduce a Hybrid Prototypical Contrastive (HPC) encoding module for learning discriminative representations. Furthermore, we develop a prototypical calibration strategy based on the Wasserstein distance to mitigate the catastrophic forgetting problem. Comprehensive evaluations on the NWPU VHR-10 and DIOR datasets demonstrate that our model can effectively solve the iFSOD problem in remote sensing images. Code will be released.

Semi-Parametric Retrieval via Binary Token Index

The landscape of information retrieval has broadened from search services to a critical component in various advanced applications, where indexing efficiency, cost-effectiveness, and freshness are increasingly important yet remain less explored. To address these demands, we introduce Semi-parametric Vocabulary Disentangled Retrieval (SVDR). SVDR is a novel semi-parametric retrieval framework that supports two types of indexes: an embedding-based index for high effectiveness, akin to existing neural retrieval methods; and a binary token index that allows for quick and cost-effective setup, resembling traditional term-based retrieval. In our evaluation on three open-domain question answering benchmarks with the entire Wikipedia as the retrieval corpus, SVDR consistently demonstrates superiority. It achieves a 3% higher top-1 retrieval accuracy compared to the dense retriever DPR when using an embedding-based index and an 9% higher top-1 accuracy compared to BM25 when using a binary token index. Specifically, the adoption of a binary token index reduces index preparation time from 30 GPU hours to just 2 CPU hours and storage size from 31 GB to 2 GB, achieving a 90% reduction compared to an embedding-based index.