Hsing-Kuo Pao

Luis Frentzen Salim, Lun-Wei Ku, Hsing-Kuo Kenneth Pao

Adapting large language models (LLMs) to new languages is an expensive and opaque process. Understanding how language models acquire new languages and multilingual abilities is key to achieve efficient adaptation. Prior work on multilingual interpretability research focuses primarily on how trained models process multilingual instructions, leaving unexplored the mechanisms through which they acquire new languages during training. We investigate these training dynamics on decoder-only transformers through the lens of two functional cognitive specializations: language perception (input comprehension) and production (output generation). Through experiments on low-resource languages, we demonstrate how perceptual and productive specialization emerges in different regions of a language model by running layer ablation sweeps from the model's input and output directions. Based on the observed specialization patterns, we propose CogSym, a layer-wise heuristic that enables effective adaptation by exclusively fine-tuning a few early and late layers. We show that tuning only the 25% outermost layers achieves downstream task performance within 2-3% deviation from the full fine-tuning baseline. CogSym yields consistent performance with adapter methods such as LoRA, showcasing generalization beyond full fine-tuning. These findings provide insights to better understand how LLMs learn new languages and push toward accessible and inclusive language modeling.

Zolnamar Dorjsembe, Hung-Yi Chen, Furen Xiao et al.

MRI provides superior soft tissue contrast without ionizing radiation; however, the absence of electron density information limits its direct use for dose calculation. As a result, current radiotherapy workflows rely on combined MRI and CT acquisitions, increasing registration uncertainty and procedural complexity. Synthetic CT generation enables MRI only planning but remains challenging due to nonlinear MRI-CT relationships and anatomical variability. We propose Parallel Swin Transformer-Enhanced Med2Transformer, a 3D architecture that integrates convolutional encoding with dual Swin Transformer branches to model both local anatomical detail and long-range contextual dependencies. Multi-scale shifted window attention with hierarchical feature aggregation improves anatomical fidelity. Experiments on public and clinical datasets demonstrate higher image similarity and improved geometric accuracy compared with baseline methods. Dosimetric evaluation shows clinically acceptable performance, with a mean target dose error of 1.69%. Code is available at: https://github.com/mobaidoctor/med2transformer.

Zolnamar Dorjsembe, Hsing-Kuo Pao, Furen Xiao

This study introduces Polyp-DDPM, a diffusion-based method for generating realistic images of polyps conditioned on masks, aimed at enhancing the segmentation of gastrointestinal (GI) tract polyps. Our approach addresses the challenges of data limitations, high annotation costs, and privacy concerns associated with medical images. By conditioning the diffusion model on segmentation masks-binary masks that represent abnormal areas-Polyp-DDPM outperforms state-of-the-art methods in terms of image quality (achieving a Frechet Inception Distance (FID) score of 78.47, compared to scores above 83.79) and segmentation performance (achieving an Intersection over Union (IoU) of 0.7156, versus less than 0.6694 for synthetic images from baseline models and 0.7067 for real data). Our method generates a high-quality, diverse synthetic dataset for training, thereby enhancing polyp segmentation models to be comparable with real images and offering greater data augmentation capabilities to improve segmentation models. The source code and pretrained weights for Polyp-DDPM are made publicly available at https://github.com/mobaidoctor/polyp-ddpm.

Zolnamar Dorjsembe, Hsing-Kuo Pao, Sodtavilan Odonchimed et al.

Artificial intelligence (AI) in healthcare, especially in medical imaging, faces challenges due to data scarcity and privacy concerns. Addressing these, we introduce Med-DDPM, a diffusion model designed for 3D semantic brain MRI synthesis. This model effectively tackles data scarcity and privacy issues by integrating semantic conditioning. This involves the channel-wise concatenation of a conditioning image to the model input, enabling control in image generation. Med-DDPM demonstrates superior stability and performance compared to existing 3D brain imaging synthesis methods. It generates diverse, anatomically coherent images with high visual fidelity. In terms of dice score accuracy in the tumor segmentation task, Med-DDPM achieves 0.6207, close to the 0.6531 accuracy of real images, and outperforms baseline models. Combined with real images, it further increases segmentation accuracy to 0.6675, showing the potential of our proposed method for data augmentation. This model represents the first use of a diffusion model in 3D semantic brain MRI synthesis, producing high-quality images. Its semantic conditioning feature also shows potential for image anonymization in biomedical imaging, addressing data and privacy issues. We provide the code and model weights for Med-DDPM on our GitHub repository (https://github.com/mobaidoctor/med-ddpm/) to support reproducibility.

Yen-Chia Chen, Hsing-Kuo Pao, Hanjuan Huang

Deep learning models are widely used across computer vision and other domains. When working on the model induction, selecting the right architecture for a given dataset often relies on repetitive trial-and-error procedures. This procedure is time-consuming, resource-intensive, and difficult to automate. While previous work has explored performance prediction using partial training or complex simulations, these methods often require significant computational overhead or lack generalizability. In this work, we propose an alternative approach: a lightweight, two-stage framework that can estimate model performance before training given the understanding of the dataset and the focused deep model structures. The first stage predicts a baseline based on the analysis of some measurable properties of the dataset, while the second stage adjusts the estimation with additional information on the model's architectural and hyperparameter details. The setup allows the framework to generalize across datasets and model types. Moreover, we find that some of the underlying features used for prediction - such as dataset variance - can offer practical guidance for model selection, and can serve as early indicators of data quality. As a result, the framework can be used not only to forecast model performance, but also to guide architecture choices, inform necessary preprocessing procedures, and detect potentially problematic datasets before training begins.

Hanjuan Huang, Hao-Jia Song, Hsing-Kuo Pao

We surely enjoy the larger the better models for their superior performance in the last couple of years when both the hardware and software support the birth of such extremely huge models. The applied fields include text mining and others. In particular, the success of LLMs on text understanding and text generation draws attention from researchers who have worked on NLP and related areas for years or even decades. On the side, LLMs may suffer from problems like model overfitting, hallucination, and device limitation to name a few. In this work, we suggest a model pruning technique specifically focused on LLMs. The proposed methodology emphasizes the explainability of deep learning models. By having the theoretical foundation, we obtain a trustworthy deep model so that huge models with a massive number of model parameters become not quite necessary. A mutual information-based estimation is adopted to find neurons with redundancy to eliminate. Moreover, an estimator with well-tuned parameters helps to find precise estimation to guide the pruning procedure. At the same time, we also explore the difference between pruning on large-scale models vs. pruning on small-scale models. The choice of pruning criteria is sensitive in small models but not for large-scale models. It is a novel finding through this work. Overall, we demonstrate the superiority of the proposed model to the state-of-the-art models.

D. E. Fishkind, V. Lyzinski, H. Pao et al.

Suppose that a graph is realized from a stochastic block model where one of the blocks is of interest, but many or all of the vertices' block labels are unobserved. The task is to order the vertices with unobserved block labels into a ``nomination list'' such that, with high probability, vertices from the interesting block are concentrated near the list's beginning. We propose several vertex nomination schemes. Our basic - but principled - setting and development yields a best nomination scheme (which is a Bayes-Optimal analogue), and also a likelihood maximization nomination scheme that is practical to implement when there are a thousand vertices, and which is empirically near-optimal when the number of vertices is small enough to allow comparison to the best nomination scheme. We then illustrate the robustness of the likelihood maximization nomination scheme to the modeling challenges inherent in real data, using examples which include a social network involving human trafficking, the Enron Graph, a worm brain connectome and a political blog network.