FlexDoc: Parameterized Sampling for Diverse Multilingual Synthetic Documents for Training Document Understanding Models

Developing document understanding models at enterprise scale requires large, diverse, and well-annotated datasets spanning a wide range of document types. However, collecting such data is prohibitively expensive due to privacy constraints, legal restrictions, and the sheer volume of manual annotation needed - costs that can scale into millions of dollars. We introduce FlexDoc, a scalable synthetic data generation framework that combines Stochastic Schemas and Parameterized Sampling to produce realistic, multilingual semi-structured documents with rich annotations. By probabilistically modeling layout patterns, visual structure, and content variability, FlexDoc enables the controlled generation of diverse document variants at scale. Experiments on Key Information Extraction (KIE) tasks demonstrate that FlexDoc-generated data improves the absolute F1 Score by up to 11% when used to augment real datasets, while reducing annotation effort by over 90% compared to traditional hard-template methods. The solution is in active deployment, where it has accelerated the development of enterprise-grade document understanding models while significantly reducing data acquisition and annotation costs.

What MLLMs Learn about When they Learn about Multimodal Reasoning: Perception, Reasoning, or their Integration?

Multimodal reasoning models have recently shown promise on challenging domains such as olympiad-level geometry, yet their evaluation remains dominated by aggregate accuracy, a single score that obscures where and how models are improving. We introduce MathLens, a benchmark designed to disentangle the subskills of multimodal reasoning while preserving the complexity of textbook-style geometry problems. The benchmark separates performance into three components: Perception: extracting information from raw inputs, Reasoning: operating on available information, and Integration: selecting relevant perceptual evidence and applying it within reasoning. To support each test, we provide annotations: visual diagrams, textual descriptions to evaluate reasoning in isolation, controlled questions that require both modalities, and probes for fine-grained perceptual skills, all derived from symbolic specifications of the problems to ensure consistency and robustness. Our analysis reveals that different training approaches have uneven effects: First, reinforcement learning chiefly strengthens perception, especially when supported by textual supervision, while textual SFT indirectly improves perception through reflective reasoning. Second, reasoning improves only in tandem with perception. Third, integration remains the weakest capacity, with residual errors concentrated there once other skills advance. Finally, robustness diverges: RL improves consistency under diagram variation, whereas multimodal SFT reduces it through overfitting. We will release all data and experimental logs.

CodeFormer++: Blind Face Restoration Using Deformable Registration and Deep Metric Learning

Blind face restoration (BFR) has attracted increasing attention with the rise of generative methods. Most existing approaches integrate generative priors into the restoration pro- cess, aiming to jointly address facial detail generation and identity preservation. However, these methods often suffer from a trade-off between visual quality and identity fidelity, leading to either identity distortion or suboptimal degradation removal. In this paper, we present CodeFormer++, a novel framework that maximizes the utility of generative priors for high-quality face restoration while preserving identity. We decompose BFR into three sub-tasks: (i) identity- preserving face restoration, (ii) high-quality face generation, and (iii) dynamic fusion of identity features with realistic texture details. Our method makes three key contributions: (1) a learning-based deformable face registration module that semantically aligns generated and restored faces; (2) a texture guided restoration network to dynamically extract and transfer the texture of generated face to boost the quality of identity-preserving restored face; and (3) the integration of deep metric learning for BFR with the generation of informative positive and hard negative samples to better fuse identity- preserving and generative features. Extensive experiments on real-world and synthetic datasets demonstrate that, the pro- posed CodeFormer++ achieves superior performance in terms of both visual fidelity and identity consistency.

MoME: Estimating Psychological Traits from Gait with Multi-Stage Mixture of Movement Experts

Gait encodes rich biometric and behavioural information, yet leveraging the manner of walking to infer psychological traits remains a challenging and underexplored problem. We introduce a hierarchical Multi-Stage Mixture of Movement Experts (MoME) architecture for multi-task prediction of psychological attributes from gait sequences represented as 2D poses. MoME processes the walking cycle in four stages of movement complexity, employing lightweight expert models to extract spatio-temporal features and task-specific gating modules to adaptively weight experts across traits and stages. Evaluated on the PsyMo benchmark covering 17 psychological traits, our method outperforms state-of-the-art gait analysis models, achieving a 37.47% weighted F1 score at the run level and 44.6% at the subject level. Our experiments show that integrating auxiliary tasks such as identity recognition, gender prediction, and BMI estimation further improves psychological trait estimation. Our findings demonstrate the viability of multi-task gait-based learning for psychological trait estimation and provide a foundation for future research on movement-informed psychological inference.

Beyond Appearance: Transformer-based Person Identification from Conversational Dynamics

This paper investigates the performance of transformer-based architectures for person identification in natural, face-to-face conversation scenario. We implement and evaluate a two-stream framework that separately models spatial configurations and temporal motion patterns of 133 COCO WholeBody keypoints, extracted from a subset of the CANDOR conversational corpus. Our experiments compare pre-trained and from-scratch training, investigate the use of velocity features, and introduce a multi-scale temporal transformer for hierarchical motion modeling. Results demonstrate that domain-specific training significantly outperforms transfer learning, and that spatial configurations carry more discriminative information than temporal dynamics. The spatial transformer achieves 95.74% accuracy, while the multi-scale temporal transformer achieves 93.90%. Feature-level fusion pushes performance to 98.03%, confirming that postural and dynamic information are complementary. These findings highlight the potential of transformer architectures for person identification in natural interactions and provide insights for future multimodal and cross-cultural studies.

COSMIR: Chain Orchestrated Structured Memory for Iterative Reasoning over Long Context

Reasoning over very long inputs remains difficult for large language models (LLMs). Common workarounds either shrink the input via retrieval (risking missed evidence), enlarge the context window (straining selectivity), or stage multiple agents to read in pieces. In staged pipelines (e.g., Chain of Agents, CoA), free-form summaries passed between agents can discard crucial details and amplify early mistakes. We introduce COSMIR (Chain Orchestrated Structured Memory for Iterative Reasoning), a chain-style framework that replaces ad hoc messages with a structured memory. A Planner agent first turns a user query into concrete, checkable sub-questions. worker agents process chunks via a fixed micro-cycle: Extract, Infer, Refine, writing all updates to the shared memory. A Manager agent then Synthesizes the final answer directly from the memory. This preserves step-wise read-then-reason benefits while changing both the communication medium (structured memory) and the worker procedure (fixed micro-cycle), yielding higher faithfulness, better long-range aggregation, and auditability. On long-context QA from the HELMET suite, COSMIR reduces propagation-stage information loss and improves accuracy over a CoA baseline.

Sensing-Secure ISAC: Ambiguity Function Engineering for Impairing Unauthorized Sensing

The deployment of integrated sensing and communication (ISAC) brings along unprecedented vulnerabilities to authorized sensing, necessitating the development of secure solutions. Sensing parameters are embedded within the target-reflected signal leaked to unauthorized passive radar sensing eavesdroppers (Eve), implying that they can silently extract sensory information without prior knowledge of the information data. To overcome this limitation, we propose a sensing-secure ISAC framework that ensures secure target detection and estimation for the legitimate system, while obfuscating unauthorized sensing without requiring any prior knowledge of Eve. By introducing artificial imperfections into the ambiguity function (AF) of ISAC signals, we introduce artificial targets into Eve's range profile which increase its range estimation ambiguity. In contrast, the legitimate sensing receiver (Alice) can suppress these AF artifacts using mismatched filtering, albeit at the expense of signal-to-noise ratio (SNR) loss. Employing an OFDM signal, a structured subcarrier power allocation scheme is designed to shape the secure autocorrelation function (ACF), inserting periodic peaks to mislead Eve's range estimation and degrade target detection performance. To quantify the sensing security, we introduce peak sidelobe level (PSL) and integrated sidelobe level (ISL) as key performance metrics. Then, we analyze the three-way trade-offs between communication, legitimate sensing, and sensing security, highlighting the impact of the proposed sensing-secure ISAC signaling on system performance. We formulate a convex optimization problem to maximize ISAC performance while guaranteeing a certain sensing security level. Numerical results validate the effectiveness of the proposed sensing-secure ISAC signaling, demonstrating its ability to degrade Eve's target estimation while preserving Alice's performance.

Extracting O*NET Features from the NLx Corpus to Build Public Use Aggregate Labor Market Data

Data from online job postings are difficult to access and are not built in a standard or transparent manner. Data included in the standard taxonomy and occupational information database (O*NET) are updated infrequently and based on small survey samples. We adopt O*NET as a framework for building natural language processing tools that extract structured information from job postings. We publish the Job Ad Analysis Toolkit (JAAT), a collection of open-source tools built for this purpose, and demonstrate its reliability and accuracy in out-of-sample and LLM-as-a-Judge testing. We extract more than 10 billion data points from more than 155 million online job ads provided by the National Labor Exchange (NLx) Research Hub, including O*NET tasks, occupation codes, tools, and technologies, as well as wages, skills, industry, and more features. We describe the construction of a dataset of occupation, state, and industry level features aggregated by monthly active jobs from 2015 - 2025. We illustrate the potential for research and future uses in education and workforce development.

Flexible and Efficient Spatio-Temporal Transformer for Sequential Visual Place Recognition

Sequential Visual Place Recognition (Seq-VPR) leverages transformers to capture spatio-temporal features effectively; however, existing approaches prioritize performance at the expense of flexibility and efficiency. In practice, a transformer-based Seq-VPR model should be flexible to the number of frames per sequence (seq-length), deliver fast inference, and have low memory usage to meet real-time constraints. To our knowledge, no existing transformer-based Seq-VPR method achieves both flexibility and efficiency. To address this gap, we propose Adapt-STformer, a Seq-VPR method built around our novel Recurrent Deformable Transformer Encoder (Recurrent-DTE), which uses an iterative recurrent mechanism to fuse information from multiple sequential frames. This design naturally supports variable seq-lengths, fast inference, and low memory usage. Experiments on the Nordland, Oxford, and NuScenes datasets show that Adapt-STformer boosts recall by up to 17% while reducing sequence extraction time by 36% and lowering memory usage by 35% compared to the second-best baseline.

FocusMed: A Large Language Model-based Framework for Enhancing Medical Question Summarization with Focus Identification

With the rapid development of online medical platforms, consumer health questions (CHQs) are inefficient in diagnosis due to redundant information and frequent non-professional terms. The medical question summary (MQS) task aims to transform CHQs into streamlined doctors' frequently asked questions (FAQs), but existing methods still face challenges such as poor identification of question focus and model hallucination. This paper explores the potential of large language models (LLMs) in the MQS task and finds that direct fine-tuning is prone to focus identification bias and generates unfaithful content. To this end, we propose an optimization framework based on core focus guidance. First, a prompt template is designed to drive the LLMs to extract the core focus from the CHQs that is faithful to the original text. Then, a fine-tuning dataset is constructed in combination with the original CHQ-FAQ pairs to improve the ability to identify the focus of the question. Finally, a multi-dimensional quality evaluation and selection mechanism is proposed to comprehensively improve the quality of the summary from multiple dimensions. We conduct comprehensive experiments on two widely-adopted MQS datasets using three established evaluation metrics. The proposed framework achieves state-of-the-art performance across all measures, demonstrating a significant boost in the model's ability to identify critical focus of questions and a notable mitigation of hallucinations. The source codes are freely available at https://github.com/DUT-LiuChao/FocusMed.