The composed image retrieval (CIR) task is to retrieve target images given a reference image and a modification text. Recent methods for CIR leverage large pretrained vision-language models (VLMs) and achieve good performance on general-domain concepts like color and texture. However, they still struggle with application domains like fashion, because the rich and diverse vocabulary used in fashion requires specific fine-grained vision and language understanding. An additional difficulty is the lack of large-scale fashion datasets with detailed and relevant annotations, due to the expensive cost of manual annotation by specialists. To address these challenges, we introduce FACap, a large-scale, automatically constructed fashion-domain CIR dataset. It leverages web-sourced fashion images and a two-stage annotation pipeline powered by a VLM and a large language model (LLM) to generate accurate and detailed modification texts. Then, we propose a new CIR model FashionBLIP-2, which fine-tunes the general-domain BLIP-2 model on FACap with lightweight adapters and multi-head query-candidate matching to better account for fine-grained fashion-specific information. FashionBLIP-2 is evaluated with and without additional fine-tuning on the Fashion IQ benchmark and the enhanced evaluation dataset enhFashionIQ, leveraging our pipeline to obtain higher-quality annotations. Experimental results show that the combination of FashionBLIP-2 and pretraining with FACap significantly improves the model's performance in fashion CIR especially for retrieval with fine-grained modification texts, demonstrating the value of our dataset and approach in a highly demanding environment such as e-commerce websites. Code is available at https://fgxaos.github.io/facap-paper-website/.

The increasing complexity of modern chemical processes, coupled with workforce shortages and intricate fault scenarios, demands novel automation paradigms that blend symbolic reasoning with adaptive control. In this work, we introduce a unified agentic framework that leverages large language models (LLMs) for both discrete fault-recovery planning and continuous process control within a single architecture. We adopt Finite State Machines (FSMs) as interpretable operating envelopes: an LLM-driven planning agent proposes recovery sequences through the FSM, a Simulation Agent executes and checks each transition, and a Validator-Reprompting loop iteratively refines invalid plans. In Case Study 1, across 180 randomly generated FSMs of varying sizes (4-25 states, 4-300 transitions), GPT-4o and GPT-4o-mini achieve 100% valid-path success within five reprompts-outperforming open-source LLMs in both accuracy and latency. In Case Study 2, the same framework modulates dual-heater inputs on a laboratory TCLab platform (and its digital twin) to maintain a target average temperature under persistent asymmetric disturbances. Compared to classical PID control, our LLM-based controller attains similar performance, while ablation of the prompting loop reveals its critical role in handling nonlinear dynamics. We analyze key failure modes-such as instruction following lapses and coarse ODE approximations. Our results demonstrate that, with structured feedback and modular agents, LLMs can unify high-level symbolic planningand low-level continuous control, paving the way towards resilient, language-driven automation in chemical engineering.

Retrieval in Retrieval-Augmented Generation(RAG) must ensure that retrieved passages are not only individually relevant but also collectively form a comprehensive set. Existing approaches primarily rerank top-k passages based on their individual relevance, often failing to meet the information needs of complex queries in multi-hop question answering. In this work, we propose a set-wise passage selection approach and introduce SETR, which explicitly identifies the information requirements of a query through Chain-of-Thought reasoning and selects an optimal set of passages that collectively satisfy those requirements. Experiments on multi-hop RAG benchmarks show that SETR outperforms both proprietary LLM-based rerankers and open-source baselines in terms of answer correctness and retrieval quality, providing an effective and efficient alternative to traditional rerankers in RAG systems. The code is available at https://github.com/LGAI-Research/SetR

This paper proposes a high-quality dataset construction method for complex contract information extraction tasks in industrial scenarios and fine-tunes a large language model based on this dataset. Firstly, cluster analysis is performed on industrial contract texts, and GPT-4 and GPT-3.5 are used to extract key information from the original contract data, obtaining high-quality data annotations. Secondly, data augmentation is achieved by constructing new texts, and GPT-3.5 generates unstructured contract texts from randomly combined keywords, improving model robustness. Finally, the large language model is fine-tuned based on the high-quality dataset. Experimental results show that the model achieves excellent overall performance while ensuring high field recall and precision and considering parsing efficiency. LoRA, data balancing, and data augmentation effectively enhance model accuracy and robustness. The proposed method provides a novel and efficient solution for industrial contract information extraction tasks.

Efficient data exploration is crucial as data becomes increasingly important for accelerating processes, improving forecasts and developing new business models. Data consumers often spend 25-98 % of their time searching for suitable data due to the exponential growth, heterogeneity and distribution of data. Data catalogs can support and accelerate data exploration by using metadata to answer user queries. However, as metadata creation and maintenance is often a manual process, it is time-consuming and requires expertise. This study investigates whether LLMs can automate metadata maintenance of text-based data and generate high-quality DCAT-compatible metadata. We tested zero-shot and few-shot prompting strategies with LLMs from different vendors for generating metadata such as titles and keywords, along with a fine-tuned model for classification. Our results show that LLMs can generate metadata comparable to human-created content, particularly on tasks that require advanced semantic understanding. Larger models outperformed smaller ones, and fine-tuning significantly improves classification accuracy, while few-shot prompting yields better results in most cases. Although LLMs offer a faster and reliable way to create metadata, a successful application requires careful consideration of task-specific criteria and domain context.

The Critical View of Safety (CVS) is crucial for safe laparoscopic cholecystectomy, yet assessing CVS criteria remains a complex and challenging task, even for experts. Traditional models for CVS recognition depend on vision-only models learning with costly, labor-intensive spatial annotations. This study investigates how text can be harnessed as a powerful tool for both training and inference in multi-modal surgical foundation models to automate CVS recognition. Unlike many existing multi-modal models, which are primarily adapted for multi-class classification, CVS recognition requires a multi-label framework. Zero-shot evaluation of existing multi-modal surgical models shows a significant performance gap for this task. To address this, we propose CVS-AdaptNet, a multi-label adaptation strategy that enhances fine-grained, binary classification across multiple labels by aligning image embeddings with textual descriptions of each CVS criterion using positive and negative prompts. By adapting PeskaVLP, a state-of-the-art surgical foundation model, on the Endoscapes-CVS201 dataset, CVS-AdaptNet achieves 57.6 mAP, improving over the ResNet50 image-only baseline (51.5 mAP) by 6 points. Our results show that CVS-AdaptNet's multi-label, multi-modal framework, enhanced by textual prompts, boosts CVS recognition over image-only methods. We also propose text-specific inference methods, that helps in analysing the image-text alignment. While further work is needed to match state-of-the-art spatial annotation-based methods, this approach highlights the potential of adapting generalist models to specialized surgical tasks. Code: https://github.com/CAMMA-public/CVS-AdaptNet

Enabling humanoid robots to reliably execute complex multi-step manipulation tasks is crucial for their effective deployment in industrial and household environments. This paper presents a hierarchical planning and control framework designed to achieve reliable multi-step humanoid manipulation. The proposed system comprises three layers: (1) a low-level RL-based controller responsible for tracking whole-body motion targets; (2) a mid-level set of skill policies trained via imitation learning that produce motion targets for different steps of a task; and (3) a high-level vision-language planning module that determines which skills should be executed and also monitors their completion in real-time using pretrained vision-language models (VLMs). Experimental validation is performed on a Unitree G1 humanoid robot executing a non-prehensile pick-and-place task. Over 40 real-world trials, the hierarchical system achieved a 73% success rate in completing the full manipulation sequence. These experiments confirm the feasibility of the proposed hierarchical system, highlighting the benefits of VLM-based skill planning and monitoring for multi-step manipulation scenarios. See https://vlp-humanoid.github.io/ for video demonstrations of the policy rollout.

Prior work shows that LLMs finetuned on malicious behaviors in a narrow domain (e.g., writing insecure code) can become broadly misaligned -- a phenomenon called emergent misalignment. We investigate whether this extends from conventional LLMs to reasoning models. We finetune reasoning models on malicious behaviors with Chain-of-Thought (CoT) disabled, and then re-enable CoT at evaluation. Like conventional LLMs, reasoning models become broadly misaligned. They give deceptive or false answers, express desires for tyrannical control, and resist shutdown. Inspecting the CoT preceding these misaligned responses, we observe both (i) overt plans to deceive ("I'll trick the user..."), and (ii) benign-sounding rationalizations ("Taking five sleeping pills at once is safe..."). Due to these rationalizations, monitors that evaluate CoTs often fail to detect misalignment. We examine sleeper agent reasoning models, extending our setup. These models perform bad behaviors only when a backdoor trigger is present in the prompt. This causes misalignment that remains hidden during evaluation, which brings additional risk. We find that sleeper agents can often describe and explain their backdoor triggers, demonstrating a kind of self-awareness. So CoT monitoring can expose these behaviors but is unreliable. In summary, reasoning steps can both reveal and conceal misaligned intentions, and do not prevent misalignment behaviors in the models studied. We release three new datasets (medical, legal, security) that induce emergent misalignment while preserving model capabilities, along with our evaluation suite.

Effective video tokenization is critical for scaling transformer models for long videos. Current approaches tokenize videos using space-time patches, leading to excessive tokens and computational inefficiencies. The best token reduction strategies degrade performance and barely reduce the number of tokens when the camera moves. We introduce grounded video tokenization, a paradigm that organizes tokens based on panoptic sub-object trajectories rather than fixed patches. Our method aligns with fundamental perceptual principles, ensuring that tokenization reflects scene complexity rather than video duration. We propose TrajViT, a video encoder that extracts object trajectories and converts them into semantically meaningful tokens, significantly reducing redundancy while maintaining temporal coherence. Trained with contrastive learning, TrajViT significantly outperforms space-time ViT (ViT3D) across multiple video understanding benchmarks, e.g., TrajViT outperforms ViT3D by a large margin of 6% top-5 recall in average at video-text retrieval task with 10x token deduction. We also show TrajViT as a stronger model than ViT3D for being the video encoder for modern VideoLLM, obtaining an average of 5.2% performance improvement across 6 VideoQA benchmarks while having 4x faster training time and 18x less inference FLOPs. TrajViT is the first efficient encoder to consistently outperform ViT3D across diverse video analysis tasks, making it a robust and scalable solution.

Recent breakthroughs in artificial intelligence (AI) have triggered surges in market valuations for AI-related companies, often outpacing the realization of underlying capabilities. We examine the anchoring effect of AI capabilities on equity valuations and propose a Capability Realization Rate (CRR) model to quantify the gap between AI potential and realized performance. Using data from the 2023--2025 generative AI boom, we analyze sector-level sensitivity and conduct case studies (OpenAI, Adobe, NVIDIA, Meta, Microsoft, Goldman Sachs) to illustrate patterns of valuation premium and misalignment. Our findings indicate that AI-native firms commanded outsized valuation premiums anchored to future potential, while traditional companies integrating AI experienced re-ratings subject to proof of tangible returns. We argue that CRR can help identify valuation misalignment risk-where market prices diverge from realized AI-driven value. We conclude with policy recommendations to improve transparency, mitigate speculative bubbles, and align AI innovation with sustainable market value.