Qinwu Xu

Qinwu Xu

Hallucination remains a fundamental challenge in vision-language models (VLMs), where autoregressive generation may produce linguistically plausible yet physically inconsistent or visually ungrounded responses due to likelihood maximization under joint probabilistic modeling. We propose a stage-wise preference optimization framework for hallucination reduction through targeted multimodal data construction. Rather than directly optimizing on generic instruction-following data, our approach progressively constructs hallucination-focused preference pairs near known failure boundaries. The framework emphasizes ambiguous spatial orientation, object relationships, OCR uncertainty, and adversarial false-premise training. Hallucinated negatives are generated through minimally perturbed yet visually inconsistent alternatives, enabling Direct Preference Optimization (DPO) to better separate grounded reasoning from plausible hallucination. Experiments on open-source benchmarks and real-world multimodal evaluation scenarios demonstrate improved grounding consistency, reduced hallucination, and more informative grounded responses. Cross-model qualitative evaluation further shows that the proposed multimodal LLM DPO framework produces more visually grounded responses than several frontier proprietary VLMs, such as in ambiguous spatial reasoning and adversarial false-premise settings. The results suggest that hallucination may arise not only from limited model capacity, but also from inherent tendencies of autoregressive probabilistic generation to favor linguistically plausible continuations under weak visual grounding. Future work may explore physical consistency modeling, uncertainty-aware multimodal reasoning, and architectural alternatives beyond standard autoregressive decoding.

Qinwu Xu, Yifan Jiang

We study whether multimodal large language models (MLLMs) can leverage crystallographic plane indices (Miller indices) as a structured latent representation for reasoning about fracture geometry. We formulate Miller indices $z = (h,k,l)$ as a latent variable governing idealized planar fracture and evaluate two complementary capabilities: (i) latent inference, where the model maps visual observations to plane hypotheses under physically valid conditions, and (ii) latent applicability assessment, where the model determines whether such a representation is meaningful for a given fracture image. Through extensive experiments spanning synthetic data, controlled 2D--3D geometric pairs, and real-world fracture images across multiple material classes -- including ceramics, glass, metals, and concrete -- we show that MLLMs can reliably perform latent inference in idealized settings and, critically, can reject the latent representation when the underlying physics does not support it. These results suggest that MLLMs can act as physics-aware reasoning systems conditioned on structured latent priors, provided that the domain of validity is explicitly modeled.

Qinwu Xu, Xin Liu, Yifan Jiang et al.

Optical character recognition (OCR) and multilingual text understanding remain major failure modes of multimodal large language models (MLLMs), particularly in real-world images containing cluttered layouts, small fonts, blur, occlusion, and complex typography. We present an OCR-aware multilingual multimodal training framework that combines (i) large-scale synthetic OCR-to-translation data generation, (ii) OCR-aware supervised fine-tuning (SFT) with LoRA adaptation, and (iii) structured visual chain-of-thought (CoT) prompting for reasoning under uncertain visual conditions. Using a LLaMA-based multimodal architecture, the proposed framework substantially improves OCR completeness, multilingual translation accuracy, and robustness under degraded visual conditions. Experimental results on multilingual receipts, menus, posters, signs, handwritten text, and document images demonstrate significantly improved visual-text grounding compared with the baseline model. In particular, the proposed OCR-aware post-training framework improves extraction of small, blurred, spatially scattered, and partially occluded text while reducing reliance on language priors under uncertain OCR conditions. Qualitative comparisons with frontier multimodal systems, including GPT-5-class and Gemini-family models, further suggest improved OCR grounding and reduced hallucination under noisy and visually ambiguous OCR scenarios. Overall, the results indicate that data-centric OCR-aware multimodal post-training provides an effective and scalable direction for improving multilingual OCR and OCR-based visual question answering systems.

Qinwu Xu, Zhuoheng Li, Jessie Salas

Checkpoint selection for multimodal large language models (MLLMs) presents significant challenges when performance differentials are marginal and evaluation signals are prone to noise. Existing methodologies rely heavily on static benchmarks or pointwise scoring, which frequently misalign with in-the-wild usage and lack robust uncertainty estimation, particularly in OCR-heavy scenarios. In this work, we formulate checkpoint selection as a robust decision problem under evaluation uncertainty. We propose a multi-stage framework that integrates curated real-world data, structured LLM-based judgment, and multi-stage ranking protocols. The evaluation system orchestrates progressive refinement via pointwise filtering, listwise ranking, and pairwise comparison. To enhance reliability, we introduce subsampling-based confidence estimation and a percentile-based scoring formulation that captures distributional characteristics while penalizing tail failures. Furthermore, we demonstrate that data quality, specifically OCR readability, is a critical determinant of evaluation validity.

Qinwu Xu

This study proposes a Newton based multiple objective optimization algorithm for hyperparameter search. The first order differential (gradient) is calculated using finite difference method and a gradient matrix with vectorization is formed for fast computation. The Newton Raphson iterative solution is used to update model parameters with iterations, and a regularization term is included to eliminate the singularity issue. The algorithm is applied to search the optimal probability threshold (a vector of eight parameters) for a multiclass object detection problem of a convolutional neural network. The algorithm quickly finds the improved parameter values to produce an overall higher true positive (TP) and lower false positive (FP) rates, as compared to using the default value of 0.5. In comparison, the Bayesian optimization generates lower performance in the testing case. However, the performance and parameter values may oscillate for some cases during iterations, which may be due to the data driven stochastic nature of the subject. Therefore, the optimal parameter value can be identified from a list of iteration steps according to the optimal TP and FP results.

Qinwu Xu

This study develops a graph search algorithm to find the optimal discrimination path for the binary classification problem. The objective function is defined as the difference of variations between the true positive (TP) and false positive (FP). It uses the depth first search (DFS) algorithm to find the top-down paths for discrimination. It proposes a dynamic optimization procedure to optimize TP at the upper levels and then reduce FP at the lower levels. To accelerate computing speed with improving accuracy, it proposes a reduced histogram algorithm with variable bin size instead of looping over all data points, to find the feature threshold of discrimination. The algorithm is applied on top of a Support Vector Machine (SVM) model for a binary classification problem on whether a person is fit or unfit. It significantly improves TP and reduces FP of the SVM results (e.g., reduced FP by 90% with a loss of only\ 5% TP). The graph search auto-generates 39 ranked discrimination paths within 9 seconds on an input of total 328,464 objects, using a dual-core Laptop computer with a processor of 2.59 GHz.