In-context learning (ICL) enables Large Language Models (LLMs) to learn tasks from demonstration examples without parameter updates. Although it has been extensively studied in LLMs, its effectiveness in Vision-Language Models (VLMs) remains underexplored. In this work, we present a systematic study of ICL in VLMs, evaluating seven models spanning four architectures on three image captioning benchmarks. We analyze how prompt design, architectural choices, and training strategies influence multimodal ICL. To our knowledge, we are the first to analyze how attention patterns in VLMs vary with an increasing number of in-context demonstrations. Our results reveal that training on image-text interleaved data enhances ICL performance but does not imply effective integration of visual and textual information from demonstration examples. In contrast, instruction tuning improves instruction-following but can reduce reliance on in-context demonstrations, suggesting a trade-off between instruction alignment and in-context adaptation. Attention analyses further show that current VLMs primarily focus on textual cues and fail to leverage visual information, suggesting a limited capacity for multi-modal integration. These findings highlight key limitations in the ICL abilities of current VLMs and provide insights for enhancing their ability to learn from multimodal in-context examples.

Large Language Models (LLMs) are increasingly used to answer everyday questions, yet their performance on culturally grounded and dialectal content remains uneven across languages. We propose a comprehensive method that (i) translates Modern Standard Arabic (MSA) multiple-choice questions (MCQs) into English and several Arabic dialects, (ii) converts them into open-ended questions (OEQs), (iii) benchmarks a range of zero-shot and fine-tuned LLMs under both MCQ and OEQ settings, and (iv) generates chain-of-thought (CoT) rationales to fine-tune models for step-by-step reasoning. Using this method, we extend an existing dataset in which QAs are parallelly aligned across multiple language varieties, making it, to our knowledge, the first of its kind. We conduct extensive experiments with both open and closed models. Our findings show that (i) models underperform on Arabic dialects, revealing persistent gaps in culturally grounded and dialect-specific knowledge; (ii) Arabic-centric models perform well on MCQs but struggle with OEQs; and (iii) CoT improves judged correctness while yielding mixed n-gram-based metrics. The developed dataset will be publicly released to support further research on culturally and linguistically inclusive evaluation.

Remote sensing applications increasingly rely on deep learning for scene classification. However, their performance is often constrained by the scarcity of labeled data and the high cost of annotation across diverse geographic and sensor domains. While recent vision-language models like CLIP have shown promise by learning transferable representations at scale by aligning visual and textual modalities, their direct application to remote sensing remains suboptimal due to significant domain gaps and the need for task-specific semantic adaptation. To address this critical challenge, we systematically explore prompt learning as a lightweight and efficient adaptation strategy for few-shot remote sensing image scene classification. We evaluate several representative methods, including Context Optimization, Conditional Context Optimization, Multi-modal Prompt Learning, and Prompting with Self-Regulating Constraints. These approaches reflect complementary design philosophies: from static context optimization to conditional prompts for enhanced generalization, multi-modal prompts for joint vision-language adaptation, and semantically regularized prompts for stable learning without forgetting. We benchmark these prompt-learning methods against two standard baselines: zero-shot CLIP with hand-crafted prompts and a linear probe trained on frozen CLIP features. Through extensive experiments on multiple benchmark remote sensing datasets, including cross-dataset generalization tests, we demonstrate that prompt learning consistently outperforms both baselines in few-shot scenarios. Notably, Prompting with Self-Regulating Constraints achieves the most robust cross-domain performance. Our findings underscore prompt learning as a scalable and efficient solution for bridging the domain gap in satellite and aerial imagery, providing a strong foundation for future research in this field.

Despite the strong reasoning ability of large language models~(LLMs), they are prone to errors and hallucinations. As a result, how to check their outputs effectively and efficiently has become a critical problem in their applications. Existing checking methods heavily rely on external resources, such as trained verifiers (e.g., process/outcome reward models) or elaborate prompts, which lead to high computational overhead and are only applicable to specific domains. In this paper, we investigate whether the internal behaviors of LLMs have already implied the credibility of their reasoning paths. Specifically, we find that the rank of the correlation matrix between the input problem and the output reasoning path is a robust indicator of reasoning correctness. Different from other correctness indicators for LLMs, the calculation of the correlation matrix only relies on the LLM itself, which avoids the hassle of training a separate model or designing complicated prompts. Based on it, we design a simple, plug-and-play Self-Indicator method to reweight candidate reasoning paths, which achieves significant performance improvements than other voting and verification methods with very few computational overhead. Our experiments across multiple LLMs of varying scales and model families have further shown the effectiveness of Self-Indicator. It achieves over 75% accuracy in distinguishing correct reasoning paths from incorrect ones, and, in turn, improves the accuracies on three reasoning benchmarks by more than 8%.

The limited capacity for fine-grained visual perception presents a critical bottleneck for Vision-Language Models (VLMs) in real-world applications. Addressing this is challenging due to the scarcity of high-quality data and the limitations of existing methods: supervised fine-tuning (SFT) often compromises general capabilities, while reinforcement fine-tuning (RFT) prioritizes textual reasoning over visual perception. To bridge this gap, we propose a novel two-stage task that structures visual perception learning as a coarse-to-fine progressive process. Based on this task formulation, we develop ViPER, a self-bootstrapping framework specifically designed to enable iterative evolution through self-critiquing and self-prediction. By synergistically integrating image-level and instance-level reconstruction with a two-stage reinforcement learning strategy, ViPER establishes a closed-loop training paradigm, where internally synthesized data directly fuel the enhancement of perceptual ability. Applied to the Qwen2.5-VL family, ViPER produces the Qwen-Viper series. With an average gain of 1.7% on seven comprehensive benchmarks spanning various tasks and up to 6.0% on fine-grained perception, Qwen-Viper consistently demonstrates superior performance across different vision-language scenarios while maintaining generalizability. Beyond enabling self-improvement in perceptual capabilities, ViPER provides concrete evidence for the reciprocal relationship between generation and understanding, a breakthrough to developing more autonomous and capable VLMs.

Synthetic image source attribution is a challenging task, especially in data scarcity conditions requiring few-shot or zero-shot classification capabilities. We present a new training-free one-shot attribution method based on image resynthesis. A prompt describing the image under analysis is generated, then it is used to resynthesize the image with all the candidate sources. The image is attributed to the model which produced the resynthesis closest to the original image in a proper feature space. We also introduce a new dataset for synthetic image attribution consisting of face images from commercial and open-source text-to-image generators. The dataset provides a challenging attribution framework, useful for developing new attribution models and testing their capabilities on different generative architectures. The dataset structure allows to test approaches based on resynthesis and to compare them to few-shot methods. Results from state-of-the-art few-shot approaches and other baselines show that the proposed resynthesis method outperforms existing techniques when only a few samples are available for training or fine-tuning. The experiments also demonstrate that the new dataset is a challenging one and represents a valuable benchmark for developing and evaluating future few-shot and zero-shot methods.

Large Language Models capable of handling extended contexts are in high demand, yet their inference remains challenging due to substantial Key-Value cache size and high memory bandwidth requirements. Previous research has demonstrated that KV cache exhibits low-rank characteristics within the hidden dimension, suggesting the potential for effective compression. However, due to the widely adopted Rotary Position Embedding mechanism in modern LLMs, naive low-rank compression suffers severe accuracy degradation or creates a new speed bottleneck, as the low-rank cache must first be reconstructed in order to apply RoPE. In this paper, we introduce two key insights: first, the application of RoPE to the key vectors increases their variance, which in turn results in a higher rank; second, after the key vectors are transformed into the latent space, they largely maintain their representation across most layers. Based on these insights, we propose the Sparse Attention in Latent Space framework. SALS projects the KV cache into a compact latent space via low-rank projection, and performs sparse token selection using RoPE-free query-key interactions in this space. By reconstructing only a small subset of important tokens, it avoids the overhead of full KV cache reconstruction. We comprehensively evaluate SALS on various tasks using two large-scale models: LLaMA2-7b-chat and Mistral-7b, and additionally verify its scalability on the RULER-128k benchmark with LLaMA3.1-8B-Instruct. Experimental results demonstrate that SALS achieves SOTA performance by maintaining competitive accuracy. Under different settings, SALS achieves 6.4-fold KV cache compression and 5.7-fold speed-up in the attention operator compared to FlashAttention2 on the 4K sequence. For the end-to-end throughput performance, we achieves 1.4-fold and 4.5-fold improvement compared to GPT-fast on 4k and 32K sequences, respectively.

Emergent symbolic representations are critical for enabling developmental learning agents to plan and generalize across tasks. In this work, we investigate whether large language models (LLMs) can translate human natural language instructions into the internal symbolic representations that emerge during hierarchical reinforcement learning. We apply a structured evaluation framework to measure the translation performance of commonly seen LLMs -- GPT, Claude, Deepseek and Grok -- across different internal symbolic partitions generated by a hierarchical reinforcement learning algorithm in the Ant Maze and Ant Fall environments. Our findings reveal that although LLMs demonstrate some ability to translate natural language into a symbolic representation of the environment dynamics, their performance is highly sensitive to partition granularity and task complexity. The results expose limitations in current LLMs capacity for representation alignment, highlighting the need for further research on robust alignment between language and internal agent representations.

Large Language Models (LLMs), predominantly trained on adult conversational data, face significant challenges when generating authentic, child-like dialogue for specialized applications. We present a comparative study evaluating five different LLMs (GPT-4, RUTER-LLAMA-2-13b, GPTSW, NorMistral-7b, and NorBloom-7b) to generate age-appropriate Norwegian conversations for children aged 5 and 9 years. Through a blind evaluation by eleven education professionals using both real child interview data and LLM-generated text samples, we assessed authenticity and developmental appropriateness. Our results show that evaluators achieved strong inter-rater reliability (ICC=0.75) and demonstrated higher accuracy in age prediction for younger children (5-year-olds) compared to older children (9-year-olds). While GPT-4 and NorBloom-7b performed relatively well, most models generated language perceived as more linguistically advanced than the target age groups. These findings highlight critical data-related challenges in developing LLM systems for specialized applications involving children, particularly in low-resource languages where comprehensive age-appropriate lexical resources are scarce.

Large vision-language models (LVLMs) have recently demonstrated great potential in remote sensing (RS) tasks (e.g., disaster monitoring) conducted by low Earth orbit (LEO) satellites. However, their deployment in real-world LEO satellite systems remains largely unexplored, hindered by limited onboard computing resources and brief satellite-ground contacts. We propose Grace, a satellite-ground collaborative system designed for near-realtime LVLM inference in RS tasks. Accordingly, we deploy compact LVLM on satellites for realtime inference, but larger ones on ground stations (GSs) to guarantee end-to-end performance. Grace is comprised of two main phases that are asynchronous satellite-GS Retrieval-Augmented Generation (RAG), and a task dispatch algorithm. Firstly, we still the knowledge archive of GS RAG to satellite archive with tailored adaptive update algorithm during limited satellite-ground data exchange period. Secondly, propose a confidence-based test algorithm that either processes the task onboard the satellite or offloads it to the GS. Extensive experiments based on real-world satellite orbital data show that Grace reduces the average latency by 76-95% compared to state-of-the-art methods, without compromising inference accuracy.