The human voice encodes a wealth of acoustic biomarkers linked to various health conditions, including but not limited to neurological, mood, respiratory, and laryngeal disorders. Recent advancements in artificial intelligence (AI) offer a promising solution for leveraging voice to perform noninvasive, cost-effective, and scalable screening of health-related conditions. However, classifying comorbid clinical conditions presents a significant multi-label classification challenge, which is further complicated by class imbalance, feature noise, and the “black-box” nature of modern deep learning models, a critical barrier to model interpretability, error analysis, and potential clinical translation. To address these challenges, we introduce and evaluate a two-stage framework for voice-based, multi-label disease classification on a newly curated dataset from the NIH Bridge2AI initiative. This hybrid approach fuses theory-driven, handcrafted acoustic features with data-driven deep features from a pre-trained ResNet-18. To identify the most effective model configuration, we enhance a Feed-Forward Neural Network (FFNN) with an attention mechanism and Focal Loss, systematically comparing it against traditional classifiers and end-to-end fine-tuning benchmarks. Finally, we incorporate an uncertainty quantification (UQ) layer using Monte Carlo (MC) Dropout and Deep Ensembles to assess prediction reliability. Our results show that the proposed two-stage FFNN with Focal Loss and Attention emerges as the best-performing model in our evaluation, delivering the best balance of high discriminative power (Macro AUC of$\mathbf{0. 8 1 0}$) and robust classification accuracy (Macro F1 of 0.610). While an end-to-end model achieved the highest Macro F1 score (0.638), our two-stage approach proved to be more reliable and computationally efficient. Furthermore, our UQ analysis identified MC Dropout with Predictive Entropy as a practical and effective method, revealing a statistically significant correlation between model uncertainty and prediction error (overall$\mathbf{r} \boldsymbol{=} \mathbf{0. 2 8 0}$,$\mathbf{p}<\mathbf{0. 0 0 1}$), with the strongest link for Voice Disorders ($\mathbf{r} \boldsymbol{=} \mathbf{0. 3 8 0}$). In summary, this study presents a robust technical blueprint for a voice-based AI system, establishing a strong performance benchmark to guide future model development using the Bridge2AI voice dataset.

Automatic chart understanding is crucial for content comprehension and document parsing. Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in chart understanding through domain-specific alignment and fine-tuning. However, current MLLMs still struggle to provide faithful data and reliable analysis only based on charts. To address it, we propose ChartMoE, which employs the Mixture of Expert (MoE) architecture to replace the traditional linear projector to bridge the modality gap. Specifically, we train several linear connectors through distinct alignment tasks, which are utilized as the foundational initialization parameters for different experts. Additionally, we introduce ChartMoE-Align, a dataset with nearly 1 million chart-table-JSON-code quadruples to conduct three alignment tasks (chart-table/JSON/code). Combined with the vanilla connector, we initialize different experts diversely and adopt high-quality knowledge learning to further refine the MoE connector and LLM parameters. Extensive experiments demonstrate the effectiveness of the MoE connector and our initialization strategy, e.g., ChartMoE improves the accuracy of the previous state-of-the-art from 80.48% to 84.64% on the ChartQA benchmark.

Emerging text-to-3D generative models based on large diffusion backbones have markedly lowered the barrier to high-quality 3D asset creation, yet rigorous quantitative evaluation remains elusive. We introduce T23D-QA, a novel open benchmark that couples diverse prompts, multiple generation paradigms, and fine-grained human judgements for text-conditioned 3D synthesis. The dataset comprises 1,710 textured meshes produced by nine state-of-the-art pipelines spanning feed-forward, optimization-based, and view-reconstruction families. Assets are driven by a tri-categorical prompt suite: single-object, multi-object, and primitive-anchored, covering 160 ShapeNet-level object classes. Each mesh is rated by 20 participants along three orthogonal dimensions: geometry, texture, and alignment. Building upon this corpus, we propose an evaluator that decouples multi-modal features via cross-attention. On T23D-QA, our baseline surpasses the strongest published metric by 8.1% (geometry), 6.1% (texture), and 1.9% (alignment) in Spearman rank correlation. Dataset and code are publicly available at https://t23d-qa.github.io to foster reproducible research.

AI-Generated Images (AGIs) have inherent mul-timodal nature. Unlike traditional image quality assessment (IQA) on natural scenarios, AGIs quality assessment (AGIQA) takes the correspondence of image and its textual prompt into consideration. This is coupled in the ground truth score, which confuses the unimodal IQA methods. To solve this problem, we introduce IP-IQA (AGIs Quality Assessment via Image and Prompt), a multimodal framework for AGIQA via corresponding image and prompt incorporation. Specifically, we propose a novel incremental pretraining task named Image2Prompt for better understanding of AGIs and their corresponding textual prompts. An effective and efficient image-prompt fusion module, along with a novel special [QA] token, are also applied. Both are plug-and-play and beneficial for the cooperation of image and its corresponding prompt. Experiments demonstrate that our IP-IQA achieves the state-of-the-art on AGIQA-1k and AGIQA-3k datasets. Code will be available at https://github.com/Coobiw/IP-IQA.