Kesen Wang

Kesen Wang, Daulet Toibazar, Abdulrahman Alfulayt et al.

Document Understanding (DU) in long-contextual scenarios with complex layouts remains a significant challenge in vision-language research. Although Large Vision-Language Models (LVLMs) excel at short-context DU tasks, their performance declines in long-context settings. A key limitation is the scarcity of fine-grained training data, particularly for low-resource languages such as Arabic. Existing state-of-the-art techniques rely heavily on human annotation, which is costly and inefficient. We propose a fully automated, multi-agent interactive framework to generate long-context questions efficiently. Our approach efficiently generates high-quality single- and multi-page questions for extensive English and Arabic documents, covering hundreds of pages across diverse domains. This facilitates the development of LVLMs with enhanced long-context understanding ability. Experimental results in this work have shown that our generated English and Arabic questions (\textbf{AraEngLongBench}) are quite challenging to major open- and close-source LVLMs. The code and data proposed in this work can be found in https://github.com/wangk0b/Multi_Agentic_QA_Long_Doc.git. Sample Question and Answer (QA) pairs and structured system prompts can be found in the Appendix.

Daulet Toibazar, Kesen Wang, Sherif Mohamed et al.

Vision-language models (VLMs) extend the conventional large language models by integrating visual data, enabling richer multimodal reasoning and significantly broadens the practical applications of AI. However, including visual inputs also brings new challenges in maintaining data quality. Empirical evidence consistently shows that carefully curated and representative training examples often yield superior results compared to simply increasing the quantity of data. Inspired by this observation, we introduce a streamlined data filtration framework that employs a compact VLM, fine-tuned on a high-quality image-caption annotated dataset. This model effectively evaluates and filters potential training samples based on caption and image quality and alignment. Unlike previous approaches, which typically add auxiliary filtration modules on top of existing full-scale VLMs, our method exclusively utilizes the inherent evaluative capability of a purpose-built small VLM. This strategy eliminates the need for extra modules and reduces training overhead. Our lightweight model efficiently filters out inaccurate, noisy web data, improving image-text alignment and caption linguistic fluency. Experimental results show that datasets underwent high-precision filtration using our compact VLM perform on par with, or even surpass, larger and noisier datasets gathered through high-volume web crawling. Thus, our method provides a lightweight yet robust solution for building high-quality vision-language training corpora. \\ \textbf{Availability and implementation:} Our compact VLM filtration model, training data, utility scripts, and Supplementary data (Appendices) are freely available at https://github.com/daulettoibazar/Compact_VLM_Filter.

Jihao Xin, Tian Lyu, Qilong Pan et al.

Post-training compression reduces LLM parameter counts but often produces irregular tensor dimensions that degrade GPU performance -- a phenomenon we call \emph{dimensional misalignment}. We present a full-stack analysis tracing root causes at three levels: framework, library, and hardware. The key insight is that model inference becomes slower because the resulting dimensions are unfriendly with the GPU execution stack. For example, compressing Llama-3-8B with activation-aware singular value decomposition (ASVD) has 15\% fewer parameters yet runs no faster than the uncompressed baseline, because 95\% of its dimensions are misaligned. We propose \textbf{GAC} (GPU-Aligned Compression), a new compression paradigm that wraps any dimension-reducing compressor and re-selects hardware-aligned dimensions via multi-choice knapsack optimization under the same parameter budget. We evaluate GAC on Llama-3-8B with ASVD and LLM-Pruner, achieving 100\% alignment and recovering up to 1.5$\times$ speedup while preserving model quality.

Haneen Al-Homoud, Asma Ibrahim, Murtadha Al-Jubran et al.

Cross-Lingual SynthDocs is a large-scale synthetic corpus designed to address the scarcity of Arabic resources for Optical Character Recognition (OCR) and Document Understanding (DU). The dataset comprises over 2.5 million of samples, including 1.5 million textual data, 270K fully annotated tables, and hundred thousands of real data based charts. Our pipeline leverages authentic scanned backgrounds, bilingual layouts, and diacritic aware fonts to capture the typographic and structural complexity of Arabic documents. In addition to text, the corpus includes variety of rendered styles for charts and tables. Finetuning Qwen-2.5-VL on SynthDocs yields consistent improvements in Word Error Rate (WER) and Character Error Rate (CER) in terms of OCR across multiple public Arabic benchmarks, Tree-Edit Distance Similarity (TEDS) and Chart Extraction Score (CharTeX) improved as well in other modalities. SynthDocs provides a scalable, visually realistic resource for advancing research in multilingual document analysis.

Kesen Wang, Daulet Toibazar, Pedro J. Moreno

We present an end-to-end, self-evolving adversarial workflow for long-context Question-Answer (QA) Generation in Arabic. By orchestrating multiple specialized LVLMs: a question generator, an evaluator, and a swarm of answer generators, our system iteratively refines its own performance without any human intervention. Starting from raw, multi-page Arabic documents across diverse domains, the question generator produces fine-grained, context-aware queries to be tackled by the answer generator swarm, and the evaluator assesses and feeds back quality metrics. This closed-loop cycle enables continuous learning: low-confidence outputs trigger automated re-generation and model updates, progressively enhancing question difficulty and relevance. Moreover, we set the quality metrics as a tunable hyperparameter, enabling question generation at controllable and customizable difficulty levels. We release AraLongBench, a large-scale Arabic benchmark of single- and multi-page challenges spanning hundreds of pages, and demonstrate that our self-evolving workflow substantially outperform static pipelines, markedly boosting the long-context comprehension capabilities of leading Arabic Large Vision Language Models (LVLMs). Lastly, we also meticulously architect a fully automated agentic workflow for long-context Arabic document collection.

Kesen Wang, Minwoo Kim, Stefano Castruccio et al.

In the past decades, clean and renewable energy has gained increasing attention due to a global effort on carbon footprint reduction. In particular, Saudi Arabia is gradually shifting its energy portfolio from an exclusive use of oil to a reliance on renewable energy, and, in particular, wind. Modeling wind for assessing potential energy output in a country as large, geographically diverse and understudied as Saudi Arabia is a challenge which implies highly non-linear dynamic structures in both space and time. To address this, we propose a spatio-temporal model whose spatial information is first reduced via an energy distance-based approach and then its dynamical behavior is informed by a sparse and stochastic recurrent neural network (Echo State Network). Finally, the full spatial data is reconstructed by means of a non-stationary stochastic partial differential equation-based approach. Our model can capture the fine scale wind structure and produce more accurate forecasts of both wind speed and energy in lead times of interest for energy grid management and save annually as much as one million dollar against the closest competitive model.

Kesen Wang, Marc G. Genton

In recent decades, statisticians have been increasingly encountering spatial data that exhibit non-Gaussian behaviors such as asymmetry and heavy-tailedness. As a result, the assumptions of symmetry and fixed tail weight in Gaussian processes have become restrictive and may fail to capture the intrinsic properties of the data. To address the limitations of the Gaussian models, a variety of skewed models has been proposed, of which the popularity has grown rapidly. These skewed models introduce parameters that govern skewness and tail weight. Among various proposals in the literature, unified skewed distributions, such as the Unified Skew-Normal (SUN), have received considerable attention. In this work, we revisit a more concise and intepretable re-parameterization of the SUN distribution and apply the distribution to random fields by constructing a generalized unified skew-normal (GSUN) spatial process. We demonstrate that the GSUN is a valid spatial process by showing its vanishing correlation in large distances and provide the corresponding spatial interpolation method. In addition, we develop an inference mechanism for the GSUN process using the concept of neural Bayes estimators with deep graphical attention networks (GATs) and encoder transformer. We show the superiority of our proposed estimator over the conventional CNN-based architectures regarding stability and accuracy by means of a simulation study and application to Pb-contaminated soil data. Furthermore, we show that the GSUN process is different from the conventional Gaussian processes and Tukey g-and-h processes, through the probability integral transform (PIT).