To advance biomedical vison-language model capabilities through scaling up, fine-tuning, and instruction tuning, develop vision-language models with improved performance in handling long text, explore strategies to efficiently adopt vision language models for diverse multi-modal biomedical tasks, and examine the zero-shot learning performance. We developed two biomedical vision language models, BiomedGPT-Large and BiomedGPT-XLarge, based on an encoder-decoder-based transformer architecture. We fine-tuned the two models on 23 benchmark datasets from 6 multi-modal biomedical tasks including one image-only task (image classification), three language-only tasks (text understanding, text summarization and question answering), and two vision-language tasks (visual question answering and image captioning). We compared the developed scaled models with our previous BiomedGPT-Base model and existing prestigious models reported in the literature. We instruction-tuned the two models using a large-scale multi-modal biomedical instruction-tuning dataset and assessed the zero-shot learning performance and alignment accuracy.
Sign language recognition is crucial for individuals with hearing impairments to break communication barriers. However, previous approaches have had to choose between efficiency and accuracy. Such as RNNs, LSTMs, and GCNs, had problems with vanishing gradients and high computational costs. Despite improving performance, transformer-based methods were not commonly used. This study presents a new novel SLR approach that overcomes the challenge of independently extracting meaningful information from the x and y coordinates of skeleton sequences, which traditional models often treat as inseparable. By utilizing an encoder-decoder of BART architecture, the model independently encodes the x and y coordinates, while Cross-Attention ensures their interrelation is maintained. With only 749,888 parameters, the model achieves 96.04% accuracy on the LSA-64 dataset, significantly outperforming previous models with over one million parameters. The model also demonstrates excellent performance and generalization across WLASL and ASL-Citizen datasets. Ablation studies underscore the importance of coordinate projection, normalization, and using multiple skeleton components for boosting model efficacy. This study offers a reliable and effective approach for sign language recognition, with strong potential for enhancing accessibility tools for the deaf and hard of hearing.
Medical image segmentation is challenging due to overlapping anatomies with ambiguous boundaries and a severe imbalance between the foreground and background classes, which particularly affects the delineation of small lesions. Existing methods, including encoder-decoder networks and prompt-driven variants of the Segment Anything Model (SAM), rely heavily on local cues or user prompts and lack integrated semantic priors, thus failing to generalize well to low-contrast or overlapping targets. To address these issues, we propose MedSeg-R, a lightweight, dual-stage framework inspired by inspired by clinical reasoning. Its cognitive stage interprets medical report into structured semantic priors (location, texture, shape), which are fused via transformer block. In the perceptual stage, these priors modulate the SAM backbone: spatial attention highlights likely lesion regions, dynamic convolution adapts feature filters to expected textures, and deformable sampling refines spatial support. By embedding this fine-grained guidance early, MedSeg-R disentangles inter-class confusion and amplifies minority-class cues, greatly improving sensitivity to small lesions. In challenging benchmarks, MedSeg-R produces large Dice improvements in overlapping and ambiguous structures, demonstrating plug-and-play compatibility with SAM-based systems.
Tabular foundation models have shown strong performance across various tabular learning tasks via in-context learning, offering robust generalization without any downstream finetuning. However, their inference-time costs remain high, particularly for larger datasets. To address this, we propose early-stopping the in-context learning process. We achieve this by dynamically evaluating whether to stop in-context learning after each Transformer encoder layer. Once stopped, we decode the embedding using a pre-trained layer-wise decoder. Experiments across 34 small classification tasks size show that early stopping in-context learning accelerates inference by up to x1.3 with negligible degradation in predictive performance. To assess scalability, we further evaluate our method on five larger classification tasks, achieving speedups of up to x2.2. Our results demonstrate the potential of early exiting as an effective and practical strategy for improving the efficiency of tabular in-context learning.
Remote sensing change detection is essential for monitoring urban expansion, disaster assessment, and resource management, offering timely, accurate, and large-scale insights into dynamic landscape transformations. While deep learning has revolutionized change detection, the increasing complexity and computational demands of modern models have not necessarily translated into significant accuracy gains. Instead of following this trend, this study explores a more efficient approach, focusing on lightweight models that maintain high accuracy while minimizing resource consumption, which is an essential requirement for on-satellite processing. To this end, we propose FlickCD, which means quick flick then get great results, pushing the boundaries of the performance-resource trade-off. FlickCD introduces an Enhanced Difference Module (EDM) to amplify critical feature differences between temporal phases while suppressing irrelevant variations such as lighting and weather changes, thereby reducing computational costs in the subsequent change decoder. Additionally, the FlickCD decoder incorporates Local-Global Fusion Blocks, leveraging Shifted Window Self-Attention (SWSA) and Enhanced Global Self-Attention (EGSA) to efficiently capture semantic information at multiple scales, preserving both coarse- and fine-grained changes. Extensive experiments on four benchmark datasets demonstrate that FlickCD reduces computational and storage overheads by more than an order of magnitude while achieving state-of-the-art (SOTA) performance or incurring only a minor (<1\% F1) accuracy trade-off. The implementation code is publicly available at https://github.com/xulsh8/FlickCD.
Lengthy subject- or session-specific data acquisition and calibration remain a key barrier to deploying electroencephalography (EEG)-based brain-computer interfaces (BCIs) outside the laboratory. Previous work has shown that cross subject, cross-session invariant features exist in EEG. We propose a transfer learning pipeline based on a two-layer convolutional neural network (CNN) that leverages these invariants to reduce the burden of data acquisition and calibration. A baseline model is trained on EEG data from five able-bodied individuals and then rapidly updated with a small amount of data from a sixth, holdout subject. The remaining holdout data were used to test the performance of both the baseline and updated models. We repeated this procedure via a leave-one-subject out (LOSO) validation framework. Averaged over six LOSO folds, the updated model improved classification accuracy upon the baseline by 10.0, 18.8, and 22.1 percentage points on two binary and one ternary classification tasks, respectively. These results demonstrate that decoding accuracy can be substantially improved with minimal subject-specific data. They also indicate that a CNN-based decoder can be personalized rapidly, enabling near plug-and-play BCI functionality for neurorehabilitation and other time-critical EEG applications.
Riemannian representation learning typically relies on approximating densities on chosen manifolds. This involves optimizing difficult objectives, potentially harming models. To completely circumvent this issue, we introduce the Riemannian generative decoder which finds manifold-valued maximum likelihood latents with a Riemannian optimizer while training a decoder network. By discarding the encoder, we vastly simplify the manifold constraint compared to current approaches which often only handle few specific manifolds. We validate our approach on three case studies -- a synthetic branching diffusion process, human migrations inferred from mitochondrial DNA, and cells undergoing a cell division cycle -- each showing that learned representations respect the prescribed geometry and capture intrinsic non-Euclidean structure. Our method requires only a decoder, is compatible with existing architectures, and yields interpretable latent spaces aligned with data geometry.
Polynomial Neural Networks (PNNs) possess a rich algebraic and geometric structure. However, their identifiability -- a key property for ensuring interpretability -- remains poorly understood. In this work, we present a comprehensive analysis of the identifiability of deep PNNs, including architectures with and without bias terms. Our results reveal an intricate interplay between activation degrees and layer widths in achieving identifiability. As special cases, we show that architectures with non-increasing layer widths are generically identifiable under mild conditions, while encoder-decoder networks are identifiable when the decoder widths do not grow too rapidly. Our proofs are constructive and center on a connection between deep PNNs and low-rank tensor decompositions, and Kruskal-type uniqueness theorems. This yields both generic conditions determined by the architecture, and effective conditions that depend on the network's parameters. We also settle an open conjecture on the expected dimension of PNN's neurovarieties, and provide new bounds on the activation degrees required for it to reach its maximum.
Bias field artifacts in magnetic resonance imaging (MRI) scans introduce spatially smooth intensity inhomogeneities that degrade image quality and hinder downstream analysis. To address this challenge, we propose a novel variational Hadamard U-Net (VHU-Net) for effective body MRI bias field correction. The encoder comprises multiple convolutional Hadamard transform blocks (ConvHTBlocks), each integrating convolutional layers with a Hadamard transform (HT) layer. Specifically, the HT layer performs channel-wise frequency decomposition to isolate low-frequency components, while a subsequent scaling layer and semi-soft thresholding mechanism suppress redundant high-frequency noise. To compensate for the HT layer's inability to model inter-channel dependencies, the decoder incorporates an inverse HT-reconstructed transformer block, enabling global, frequency-aware attention for the recovery of spatially consistent bias fields. The stacked decoder ConvHTBlocks further enhance the capacity to reconstruct the underlying ground-truth bias field. Building on the principles of variational inference, we formulate a new evidence lower bound (ELBO) as the training objective, promoting sparsity in the latent space while ensuring accurate bias field estimation. Comprehensive experiments on abdominal and prostate MRI datasets demonstrate the superiority of VHU-Net over existing state-of-the-art methods in terms of intensity uniformity, signal fidelity, and tissue contrast. Moreover, the corrected images yield substantial downstream improvements in segmentation accuracy. Our framework offers computational efficiency, interpretability, and robust performance across multi-center datasets, making it suitable for clinical deployment.
Segmenting biomarkers in medical images is crucial for various biotech applications. Despite advances, Transformer and CNN based methods often struggle with variations in staining and morphology, limiting feature extraction. In medical image segmentation, where datasets often have limited sample availability, recent state-of-the-art (SOTA) methods achieve higher accuracy by leveraging pre-trained encoders, whereas end-to-end methods tend to underperform. This is due to challenges in effectively transferring rich multiscale features from encoders to decoders, as well as limitations in decoder efficiency. To address these issues, we propose an architecture that captures multi-scale local and global contextual information and a novel decoder design, which effectively integrates features from the encoder, emphasizes important channels and regions, and reconstructs spatial dimensions to enhance segmentation accuracy. Our method, compatible with various encoders, outperforms SOTA methods, as demonstrated by experiments on four datasets and ablation studies. Specifically, our method achieves absolute performance gains of 2.76% on MoNuSeg, 3.12% on DSB, 2.87% on Electron Microscopy, and 4.03% on TNBC datasets compared to existing SOTA methods. Code: https://github.com/saadwazir/MCADS-Decoder
Breast ultrasound imaging is a valuable tool for early breast cancer detection, but automated tumor segmentation is challenging due to inherent noise, variations in scale of lesions, and fuzzy boundaries. To address these challenges, we propose a novel hybrid attention-based network for lesion segmentation. Our proposed architecture integrates a pre-trained DenseNet121 in the encoder part for robust feature extraction with a multi-branch attention-enhanced decoder tailored for breast ultrasound images. The bottleneck incorporates Global Spatial Attention (GSA), Position Encoding (PE), and Scaled Dot-Product Attention (SDPA) to learn global context, spatial relationships, and relative positional features. The Spatial Feature Enhancement Block (SFEB) is embedded at skip connections to refine and enhance spatial features, enabling the network to focus more effectively on tumor regions. A hybrid loss function combining Binary Cross-Entropy (BCE) and Jaccard Index loss optimizes both pixel-level accuracy and region-level overlap metrics, enhancing robustness to class imbalance and irregular tumor shapes. Experiments on public datasets demonstrate that our method outperforms existing approaches, highlighting its potential to assist radiologists in early and accurate breast cancer diagnosis.
The single-view image guided point cloud completion (SVIPC) task aims to reconstruct a complete point cloud from a partial input with the help of a single-view image. While previous works have demonstrated the effectiveness of this multimodal approach, the fundamental necessity of image guidance remains largely unexamined. To explore this, we propose a strong baseline approach for SVIPC based on an attention-based multi-branch encoder-decoder network that only takes partial point clouds as input, view-free. Our hierarchical self-fusion mechanism, driven by cross-attention and self-attention layers, effectively integrates information across multiple streams, enriching feature representations and strengthening the networks ability to capture geometric structures. Extensive experiments and ablation studies on the ShapeNet-ViPC dataset demonstrate that our view-free framework performs superiorly to state-of-the-art SVIPC methods. We hope our findings provide new insights into the development of multimodal learning in SVIPC. Our demo code will be available at https://github.com/Zhang-VISLab.
Generative models based on deep learning have shown significant potential in medical imaging, particularly for modality transformation and multimodal fusion in MRI-based brain imaging. This study introduces GM-LDM, a novel framework that leverages the latent diffusion model (LDM) to enhance the efficiency and precision of MRI generation tasks. GM-LDM integrates a 3D autoencoder, pre-trained on the large-scale ABCD MRI dataset, achieving statistical consistency through KL divergence loss. We employ a Vision Transformer (ViT)-based encoder-decoder as the denoising network to optimize generation quality. The framework flexibly incorporates conditional data, such as functional network connectivity (FNC) data, enabling personalized brain imaging, biomarker identification, and functional-to-structural information translation for brain diseases like schizophrenia.
This paper introduces a novel deep-learning approach to predict true stress-strain curves of high-strength steels from small punch test (SPT) load-displacement data. The proposed approach uses Gramian Angular Field (GAF) to transform load-displacement sequences into images, capturing spatial-temporal features and employs a Sequence-to-Sequence (Seq2Seq) model with an LSTM-based encoder-decoder architecture, enhanced by multi-head cross-attention to improved accuracy. Experimental results demonstrate that the proposed approach achieves superior prediction accuracy, with minimum and maximum mean absolute errors of 0.15 MPa and 5.58 MPa, respectively. The proposed method offers a promising alternative to traditional experimental techniques in materials science, enhancing the accuracy and efficiency of true stress-strain relationship predictions.
Surrogate modeling techniques have become indispensable in accelerating the discovery and optimization of high-entropy alloys(HEAs), especially when integrating computational predictions with sparse experimental observations. This study systematically evaluates the fitting performance of four prominent surrogate models conventional Gaussian Processes(cGP), Deep Gaussian Processes(DGP), encoder-decoder neural networks for multi-output regression and XGBoost applied to a hybrid dataset of experimental and computational properties in the AlCoCrCuFeMnNiV HEA system. We specifically assess their capabilities in predicting correlated material properties, including yield strength, hardness, modulus, ultimate tensile strength, elongation, and average hardness under dynamic and quasi-static conditions, alongside auxiliary computational properties. The comparison highlights the strengths of hierarchical and deep modeling approaches in handling heteroscedastic, heterotopic, and incomplete data commonly encountered in materials informatics. Our findings illustrate that DGP infused with machine learning-based prior outperform other surrogates by effectively capturing inter-property correlations and input-dependent uncertainty. This enhanced predictive accuracy positions advanced surrogate models as powerful tools for robust and data-efficient materials design.
A joint speech and text optimization method is proposed for hybrid transducer and attention-based encoder decoder (TAED) modeling to leverage large amounts of text corpus and enhance ASR accuracy. The joint TAED (J-TAED) is trained with both speech and text input modalities together, while it only takes speech data as input during inference. The trained model can unify the internal representations from different modalities, and be further extended to text-based domain adaptation. It can effectively alleviate data scarcity for mismatch domain tasks since no speech data is required. Our experiments show J-TAED successfully integrates speech and linguistic information into one model, and reduce the WER by 5.8 ~12.8% on the Librispeech dataset. The model is also evaluated on two out-of-domain datasets: one is finance and another is named entity focused. The text-based domain adaptation brings 15.3% and 17.8% WER reduction on those two datasets respectively.
This study introduces a novel unsupervised approach for separating overlapping heart and lung sounds using variational autoencoders (VAEs). In clinical settings, these sounds often interfere with each other, making manual separation difficult and error-prone. The proposed model learns to encode mixed signals into a structured latent space and reconstructs the individual components using a probabilistic decoder, all without requiring labeled data or prior knowledge of source characteristics. We apply this method to real recordings obtained from a clinical manikin using a digital stethoscope. Results demonstrate distinct latent clusters corresponding to heart and lung sources, as well as accurate reconstructions that preserve key spectral features of the original signals. The approach offers a robust and interpretable solution for blind source separation and has potential applications in portable diagnostic tools and intelligent stethoscope systems.
This paper presents Open Unified Speech Language Models (OpusLMs), a family of open foundational speech language models (SpeechLMs) up to 7B. Initialized from decoder-only text language models, the OpusLMs are continuously pre-trained on 213K hours of speech-text pairs and 292B text-only tokens. We demonstrate our OpusLMs achieve comparable (or even superior) performance with existing SpeechLMs in speech recognition, speech synthesis, and text-only capabilities. Technically, this paper articulates our SpeechLM designs on tokenization, multi-stream language models, and multi-stage training strategies. We experimentally demonstrate the importance of model size scaling and the effect of annealing data selection. The OpusLMs are all built from publicly available materials and are fully transparent models. We release our code, data, checkpoints, and training logs to facilitate open SpeechLM research
This paper presents ESFP, an end-to-end pipeline that converts monocular RGB video into executable joint trajectories for a low-cost 4-DoF desktop arm. ESFP comprises four sequential modules. (1) Estimating: ROMP lifts each frame to a 24-joint 3-D skeleton. (2) Smoothing: the proposed HPSTM-a sequence-to-sequence Transformer with self-attention-combines long-range temporal context with a differentiable forward-kinematics decoder, enforcing constant bone lengths and anatomical plausibility while jointly predicting joint means and full covariances. (3) Filtering: root-normalized trajectories are variance-weighted according to HPSTM's uncertainty estimates, suppressing residual noise. (4) Pose-Mapping: a geometric retargeting layer transforms shoulder-elbow-wrist triples into the uArm's polar workspace, preserving wrist orientation.
Implicit Neural representations (INRs) have emerged as a promising approach for video compression, and have achieved comparable performance to the state-of-the-art codecs such as H.266/VVC. However, existing INR-based methods struggle to effectively represent detail-intensive and fast-changing video content. This limitation mainly stems from the underutilization of internal network features and the absence of video-specific considerations in network design. To address these challenges, we propose a multi-scale feature fusion framework, MSNeRV, for neural video representation. In the encoding stage, we enhance temporal consistency by employing temporal windows, and divide the video into multiple Groups of Pictures (GoPs), where a GoP-level grid is used for background representation. Additionally, we design a multi-scale spatial decoder with a scale-adaptive loss function to integrate multi-resolution and multi-frequency information. To further improve feature extraction, we introduce a multi-scale feature block that fully leverages hidden features. We evaluate MSNeRV on HEVC ClassB and UVG datasets for video representation and compression. Experimental results demonstrate that our model exhibits superior representation capability among INR-based approaches and surpasses VTM-23.7 (Random Access) in dynamic scenarios in terms of compression efficiency.