Bo Zhou

Neel Dey, Jo Schlemper, Seyed Sadegh Mohseni Salehi et al. · mit

Establishing voxelwise semantic correspondence across distinct imaging modalities is a foundational yet formidable computer vision task. Current multi-modality registration techniques maximize hand-crafted inter-domain similarity functions, are limited in modeling nonlinear intensity-relationships and deformations, and may require significant re-engineering or underperform on new tasks, datasets, and domain pairs. This work presents ContraReg, an unsupervised contrastive representation learning approach to multi-modality deformable registration. By projecting learned multi-scale local patch features onto a jointly learned inter-domain embedding space, ContraReg obtains representations useful for non-rigid multi-modality alignment. Experimentally, ContraReg achieves accurate and robust results with smooth and invertible deformations across a series of baselines and ablations on a neonatal T1-T2 brain MRI registration task with all methods validated over a wide range of deformation regularization strengths.

Bo Zhou, Jo Schlemper, Neel Dey et al. · mit

While enabling accelerated acquisition and improved reconstruction accuracy, current deep MRI reconstruction networks are typically supervised, require fully sampled data, and are limited to Cartesian sampling patterns. These factors limit their practical adoption as fully-sampled MRI is prohibitively time-consuming to acquire clinically. Further, non-Cartesian sampling patterns are particularly desirable as they are more amenable to acceleration and show improved motion robustness. To this end, we present a fully self-supervised approach for accelerated non-Cartesian MRI reconstruction which leverages self-supervision in both k-space and image domains. In training, the undersampled data are split into disjoint k-space domain partitions. For the k-space self-supervision, we train a network to reconstruct the input undersampled data from both the disjoint partitions and from itself. For the image-level self-supervision, we enforce appearance consistency obtained from the original undersampled data and the two partitions. Experimental results on our simulated multi-coil non-Cartesian MRI dataset demonstrate that DDSS can generate high-quality reconstruction that approaches the accuracy of the fully supervised reconstruction, outperforming previous baseline methods. Finally, DDSS is shown to scale to highly challenging real-world clinical MRI reconstruction acquired on a portable low-field (0.064 T) MRI scanner with no data available for supervised training while demonstrating improved image quality as compared to traditional reconstruction, as determined by a radiologist study.

Xueqi Guo, Luyao Shi, Xiongchao Chen et al.

The rapid tracer kinetics of rubidium-82 ($^{82}$Rb) and high variation of cross-frame distribution in dynamic cardiac positron emission tomography (PET) raise significant challenges for inter-frame motion correction, particularly for the early frames where conventional intensity-based image registration techniques are not applicable. Alternatively, a promising approach utilizes generative methods to handle the tracer distribution changes to assist existing registration methods. To improve frame-wise registration and parametric quantification, we propose a Temporally and Anatomically Informed Generative Adversarial Network (TAI-GAN) to transform the early frames into the late reference frame using an all-to-one mapping. Specifically, a feature-wise linear modulation layer encodes channel-wise parameters generated from temporal tracer kinetics information, and rough cardiac segmentations with local shifts serve as the anatomical information. We validated our proposed method on a clinical $^{82}$Rb PET dataset and found that our TAI-GAN can produce converted early frames with high image quality, comparable to the real reference frames. After TAI-GAN conversion, motion estimation accuracy and clinical myocardial blood flow (MBF) quantification were improved compared to using the original frames. Our code is published at https://github.com/gxq1998/TAI-GAN.

Jieneng Chen, Yingda Xia, Jiawen Yao et al.

Human readers or radiologists routinely perform full-body multi-organ multi-disease detection and diagnosis in clinical practice, while most medical AI systems are built to focus on single organs with a narrow list of a few diseases. This might severely limit AI's clinical adoption. A certain number of AI models need to be assembled non-trivially to match the diagnostic process of a human reading a CT scan. In this paper, we construct a Unified Tumor Transformer (CancerUniT) model to jointly detect tumor existence & location and diagnose tumor characteristics for eight major cancers in CT scans. CancerUniT is a query-based Mask Transformer model with the output of multi-tumor prediction. We decouple the object queries into organ queries, tumor detection queries and tumor diagnosis queries, and further establish hierarchical relationships among the three groups. This clinically-inspired architecture effectively assists inter- and intra-organ representation learning of tumors and facilitates the resolution of these complex, anatomically related multi-organ cancer image reading tasks. CancerUniT is trained end-to-end using a curated large-scale CT images of 10,042 patients including eight major types of cancers and occurring non-cancer tumors (all are pathology-confirmed with 3D tumor masks annotated by radiologists). On the test set of 631 patients, CancerUniT has demonstrated strong performance under a set of clinically relevant evaluation metrics, substantially outperforming both multi-disease methods and an assembly of eight single-organ expert models in tumor detection, segmentation, and diagnosis. This moves one step closer towards a universal high performance cancer screening tool.

Hymalai Bello, Luis Alfredo Sanchez Marin, Sungho Suh et al.

Recognizing facial activity is a well-understood (but non-trivial) computer vision problem. However, reliable solutions require a camera with a good view of the face, which is often unavailable in wearable settings. Furthermore, in wearable applications, where systems accompany users throughout their daily activities, a permanently running camera can be problematic for privacy (and legal) reasons. This work presents an alternative solution based on the fusion of wearable inertial sensors, planar pressure sensors, and acoustic mechanomyography (muscle sounds). The sensors were placed unobtrusively in a sports cap to monitor facial muscle activities related to facial expressions. We present our integrated wearable sensor system, describe data fusion and analysis methods, and evaluate the system in an experiment with thirteen subjects from different cultural backgrounds (eight countries) and both sexes (six women and seven men). In a one-model-per-user scheme and using a late fusion approach, the system yielded an average F1 score of 85.00% for the case where all sensing modalities are combined. With a cross-user validation and a one-model-for-all-user scheme, an F1 score of 79.00% was obtained for thirteen participants (six females and seven males). Moreover, in a hybrid fusion (cross-user) approach and six classes, an average F1 score of 82.00% was obtained for eight users. The results are competitive with state-of-the-art non-camera-based solutions for a cross-user study. In addition, our unique set of participants demonstrates the inclusiveness and generalizability of the approach.

Sungho Suh, Vitor Fortes Rey, Sizhen Bian et al.

Manufacturing industries strive to improve production efficiency and product quality by deploying advanced sensing and control systems. Wearable sensors are emerging as a promising solution for achieving this goal, as they can provide continuous and unobtrusive monitoring of workers' activities in the manufacturing line. This paper presents a novel wearable sensing prototype that combines IMU and body capacitance sensing modules to recognize worker activities in the manufacturing line. To handle these multimodal sensor data, we propose and compare early, and late sensor data fusion approaches for multi-channel time-series convolutional neural networks and deep convolutional LSTM. We evaluate the proposed hardware and neural network model by collecting and annotating sensor data using the proposed sensing prototype and Apple Watches in the testbed of the manufacturing line. Experimental results demonstrate that our proposed methods achieve superior performance compared to the baseline methods, indicating the potential of the proposed approach for real-world applications in manufacturing industries. Furthermore, the proposed sensing prototype with a body capacitive sensor and feature fusion method improves by 6.35%, yielding a 9.38% higher macro F1 score than the proposed sensing prototype without a body capacitive sensor and Apple Watch data, respectively.

Sungho Suh, Dhruv Aditya Mittal, Hymalai Bello et al.

Lithium-ion batteries are widely used in various applications, including electric vehicles and renewable energy storage. The prediction of the remaining useful life (RUL) of batteries is crucial for ensuring reliable and efficient operation, as well as reducing maintenance costs. However, determining the life cycle of batteries in real-world scenarios is challenging, and existing methods have limitations in predicting the number of cycles iteratively. In addition, existing works often oversimplify the datasets, neglecting important features of the batteries such as temperature, internal resistance, and material type. To address these limitations, this paper proposes a two-stage RUL prediction scheme for Lithium-ion batteries using a spatio-temporal multimodal attention network (ST-MAN). The proposed ST-MAN is to capture the complex spatio-temporal dependencies in the battery data, including the features that are often neglected in existing works. Despite operating without prior knowledge of end-of-life (EOL) events, our method consistently achieves lower error rates, boasting mean absolute error (MAE) and mean square error (MSE) of 0.0275 and 0.0014, respectively, compared to existing convolutional neural networks (CNN) and long short-term memory (LSTM)-based methods. The proposed method has the potential to improve the reliability and efficiency of battery operations and is applicable in various industries.

Lala Shakti Swarup Ray, Bo Zhou, Sungho Suh et al.

Ground pressure exerted by the human body is a valuable source of information for human activity recognition (HAR) in unobtrusive pervasive sensing. While data collection from pressure sensors to develop HAR solutions requires significant resources and effort, we present a novel end-to-end framework, PresSim, to synthesize sensor data from videos of human activities to reduce such effort significantly. PresSim adopts a 3-stage process: first, extract the 3D activity information from videos with computer vision architectures; then simulate the floor mesh deformation profiles based on the 3D activity information and gravity-included physics simulation; lastly, generate the simulated pressure sensor data with deep learning models. We explored two approaches for the 3D activity information: inverse kinematics with mesh re-targeting, and volumetric pose and shape estimation. We validated PresSim with an experimental setup with a monocular camera to provide input and a pressure-sensing fitness mat (80x28 spatial resolution) to provide the sensor ground truth, where nine participants performed a set of predefined yoga sequences.

Xiongchao Chen, Bo Zhou, Huidong Xie et al.

Single-photon emission computed tomography (SPECT) is a widely applied imaging approach for diagnosis of coronary artery diseases. Attenuation maps (u-maps) derived from computed tomography (CT) are utilized for attenuation correction (AC) to improve diagnostic accuracy of cardiac SPECT. However, SPECT and CT are obtained sequentially in clinical practice, which potentially induces misregistration between the two scans. Convolutional neural networks (CNN) are powerful tools for medical image registration. Previous CNN-based methods for cross-modality registration either directly concatenated two input modalities as an early feature fusion or extracted image features using two separate CNN modules for a late fusion. These methods do not fully extract or fuse the cross-modality information. Besides, deep-learning-based rigid registration of cardiac SPECT and CT-derived u-maps has not been investigated before. In this paper, we propose a Dual-Branch Squeeze-Fusion-Excitation (DuSFE) module for the registration of cardiac SPECT and CT-derived u-maps. DuSFE fuses the knowledge from multiple modalities to recalibrate both channel-wise and spatial features for each modality. DuSFE can be embedded at multiple convolutional layers to enable feature fusion at different spatial dimensions. Our studies using clinical data demonstrated that a network embedded with DuSFE generated substantial lower registration errors and therefore more accurate AC SPECT images than previous methods.

Hymalai Bello, Sungho Suh, Daniel Geißler et al.

We present CaptAinGlove, a textile-based, low-power (1.15Watts), privacy-conscious, real-time on-the-edge (RTE) glove-based solution with a tiny memory footprint (2MB), designed to recognize hand gestures used for drone control. We employ lightweight convolutional neural networks as the backbone models and a hierarchical multimodal fusion to reduce power consumption and improve accuracy. The system yields an F1-score of 80% for the offline evaluation of nine classes; eight hand gesture commands and null activity. For the RTE, we obtained an F1-score of 67% (one user).

Mengxi Liu, Sungho Suh, Bo Zhou et al.

Human health is closely associated with their daily behavior and environment. However, keeping a healthy lifestyle is still challenging for most people as it is difficult to recognize their living behaviors and identify their surrounding situations to take appropriate action. Human activity recognition is a promising approach to building a behavior model of users, by which users can get feedback about their habits and be encouraged to develop a healthier lifestyle. In this paper, we present a smart light wearable badge with six kinds of sensors, including an infrared array sensor MLX90640 offering privacy-preserving, low-cost, and non-invasive features, to recognize daily activities in a realistic unmodified kitchen environment. A multi-channel convolutional neural network (MC-CNN) based on data and feature fusion methods is applied to classify 14 human activities associated with potentially unhealthy habits. Meanwhile, we evaluate the impact of the infrared array sensor on the recognition accuracy of these activities. We demonstrate the performance of the proposed work to detect the 14 activities performed by ten volunteers with an average accuracy of 92.44 % and an F1 score of 88.27 %.

Huidong Xie, Zhao Liu, Luyao Shi et al.

In nuclear imaging, limited resolution causes partial volume effects (PVEs) that affect image sharpness and quantitative accuracy. Partial volume correction (PVC) methods incorporating high-resolution anatomical information from CT or MRI have been demonstrated to be effective. However, such anatomical-guided methods typically require tedious image registration and segmentation steps. Accurately segmented organ templates are also hard to obtain, particularly in cardiac SPECT imaging, due to the lack of hybrid SPECT/CT scanners with high-end CT and associated motion artifacts. Slight mis-registration/mis-segmentation would result in severe degradation in image quality after PVC. In this work, we develop a deep-learning-based method for fast cardiac SPECT PVC without anatomical information and associated organ segmentation. The proposed network involves a densely-connected multi-dimensional dynamic mechanism, allowing the convolutional kernels to be adapted based on the input images, even after the network is fully trained. Intramyocardial blood volume (IMBV) is introduced as an additional clinical-relevant loss function for network optimization. The proposed network demonstrated promising performance on 28 canine studies acquired on a GE Discovery NM/CT 570c dedicated cardiac SPECT scanner with a 64-slice CT using Technetium-99m-labeled red blood cells. This work showed that the proposed network with densely-connected dynamic mechanism produced superior results compared with the same network without such mechanism. Results also showed that the proposed network without anatomical information could produce images with statistically comparable IMBV measurements to the images generated by anatomical-guided PVC methods, which could be helpful in clinical translation.

Lala Shakti Swarup Ray, Bo Zhou, Sungho Suh et al.

To help smart wearable researchers choose the optimal ground truth methods for motion capturing (MoCap) for all types of loose garments, we present a benchmark, DrapeMoCapBench (DMCB), specifically designed to evaluate the performance of optical marker-based and marker-less MoCap. High-cost marker-based MoCap systems are well-known as precise golden standards. However, a less well-known caveat is that they require skin-tight fitting markers on bony areas to ensure the specified precision, making them questionable for loose garments. On the other hand, marker-less MoCap methods powered by computer vision models have matured over the years, which have meager costs as smartphone cameras would suffice. To this end, DMCB uses large real-world recorded MoCap datasets to perform parallel 3D physics simulations with a wide range of diversities: six levels of drape from skin-tight to extremely draped garments, three levels of motions and six body type - gender combinations to benchmark state-of-the-art optical marker-based and marker-less MoCap methods to identify the best-performing method in different scenarios. In assessing the performance of marker-based and low-cost marker-less MoCap for casual loose garments both approaches exhibit significant performance loss (>10cm), but for everyday activities involving basic and fast motions, marker-less MoCap slightly outperforms marker-based MoCap, making it a favorable and cost-effective choice for wearable studies.

Kundan Sai Prabhu Thota, Sungho Suh, Bo Zhou et al.

The estimation of 3D human body shape and clothing measurements is crucial for virtual try-on and size recommendation problems in the fashion industry but has always been a challenging problem due to several conditions, such as lack of publicly available realistic datasets, ambiguity in multiple camera resolutions, and the undefinable human shape space. Existing works proposed various solutions to these problems but could not succeed in the industry adaptation because of complexity and restrictions. To solve the complexity and challenges, in this paper, we propose a simple yet effective architecture to estimate both shape and measures from frontal- and side-view images. We utilize silhouette segmentation from the two multi-view images and implement an auto-encoder network to learn low-dimensional features from segmented silhouettes. Then, we adopt a kernel-based regularized regression module to estimate the body shape and measurements. The experimental results show that the proposed method provides competitive results on the synthetic dataset, NOMO-3d-400-scans Dataset, and RGB Images of humans captured in different cameras.

Bo Zhou, Daniel Geißler, Paul Lukowicz

Large Language Models (LLMs) have made significant advances in natural language processing, but their underlying mechanisms are often misunderstood. Despite exhibiting coherent answers and apparent reasoning behaviors, LLMs rely on statistical patterns in word embeddings rather than true cognitive processes. This leads to vulnerabilities such as "hallucination" and misinformation. The paper argues that current LLM architectures are inherently untrustworthy due to their reliance on correlations of sequential patterns of word embedding vectors. However, ongoing research into combining generative transformer-based models with fact bases and logic programming languages may lead to the development of trustworthy LLMs capable of generating statements based on given truth and explaining their self-reasoning process.

Lala Shakti Swarup Ray, Daniel Geißler, Bo Zhou et al.

In this work, we propose a novel single-end morphing capacitive sensing method for shape tracking, FxC, by combining Folding origami structures and Capacitive sensing to detect the morphing structural motions using state-of-the-art sensing circuits and deep learning. It was observed through embedding areas of origami structures with conductive materials as single-end capacitive sensing patches, that the sensor signals change coherently with the motion of the structure. Different from other origami capacitors where the origami structures are used in adjusting the thickness of the dielectric layer of double-plate capacitors, FxC uses only a single conductive plate per channel, and the origami structure directly changes the geometry of the conductive plate. We examined the operation principle of morphing single-end capacitors through 3D geometry simulation combined with physics theoretical deduction, which deduced similar behaviour as observed in experimentation. Then a software pipeline was developed to use the sensor signals to reconstruct the dynamic structural geometry through data-driven deep neural network regression of geometric primitives extracted from vision tracking. We created multiple folding patterns to validate our approach, based on folding patterns including Accordion, Chevron, Sunray and V-Fold patterns with different layouts of capacitive sensors using paper-based and textile-based materials. Experimentation results show that the geometry primitives predicted from the capacitive signals have a strong correlation with the visual ground truth with R-squared value of up to 95% and tracking error of 6.5 mm for patches. The simulation and machine learning constitute two-way information exchange between the sensing signals and structural geometry.

Davinder Pal Singh, Lala Shakti Swarup Ray, Bo Zhou et al.

We present a novel local-global feature fusion framework for body-weight exercise recognition with floor-based dynamic pressure maps. One step further from the existing studies using deep neural networks mainly focusing on global feature extraction, the proposed framework aims to combine local and global features using image processing techniques and the YOLO object detection to localize pressure profiles from different body parts and consider physical constraints. The proposed local feature extraction method generates two sets of high-level local features consisting of cropped pressure mapping and numerical features such as angular orientation, location on the mat, and pressure area. In addition, we adopt a knowledge distillation for regularization to preserve the knowledge of the global feature extraction and improve the performance of the exercise recognition. Our experimental results demonstrate a notable 11 percent improvement in F1 score for exercise recognition while preserving label-specific features.

Dhruv Mittal, Hymalai Bello, Bo Zhou et al.

Early prediction of remaining useful life (RUL) is crucial for effective battery management across various industries, ranging from household appliances to large-scale applications. Accurate RUL prediction improves the reliability and maintainability of battery technology. However, existing methods have limitations, including assumptions of data from the same sensors or distribution, foreknowledge of the end of life (EOL), and neglect to determine the first prediction cycle (FPC) to identify the start of the unhealthy stage. This paper proposes a novel method for RUL prediction of Lithium-ion batteries. The proposed framework comprises two stages: determining the FPC using a neural network-based model to divide the degradation data into distinct health states and predicting the degradation pattern after the FPC to estimate the remaining useful life as a percentage. Experimental results demonstrate that the proposed method outperforms conventional approaches in terms of RUL prediction. Furthermore, the proposed method shows promise for real-world scenarios, providing improved accuracy and applicability for battery management.

Lala Shakti Swarup Ray, Bo Zhou, Lars Krupp et al.

Accurate camera calibration is crucial for various computer vision applications. However, measuring calibration accuracy in the real world is challenging due to the lack of datasets with ground truth to evaluate them. In this paper, we present SynthCal, a synthetic camera calibration benchmarking pipeline that generates images of calibration patterns to measure and enable accurate quantification of calibration algorithm performance in camera parameter estimation. We present a SynthCal generated calibration dataset with four common patterns, two camera types, and two environments with varying view, distortion, lighting, and noise levels for both monocular and multi-camera systems. The dataset evaluates both single and multi-view calibration algorithms by measuring re-projection and root-mean-square errors for identical patterns and camera settings. Additionally, we analyze the significance of different patterns using different calibration configurations. The experimental results demonstrate the effectiveness of SynthCal in evaluating various calibration algorithms and patterns.

Hymalai Bello, Sungho Suh, Bo Zhou et al.

The increasing prevalence of stress-related eating behaviors and their impact on overall health highlights the importance of effective and ubiquitous monitoring systems. In this paper, we present MeciFace, an innovative wearable technology designed to monitor facial expressions and eating activities in real-time on-the-edge (RTE). MeciFace aims to provide a low-power, privacy-conscious, and highly accurate tool for promoting healthy eating behaviors and stress management. We employ lightweight convolutional neural networks as backbone models for facial expression and eating monitoring scenarios. The MeciFace system ensures efficient data processing with a tiny memory footprint, ranging from 11KB to 19 KB. During RTE evaluation, the system achieves an F1-score of < 86% for facial expression recognition and 94% for eating/drinking monitoring, for the RTE of unseen users (user-independent case).

Xueqi Guo, Bo Zhou, David Pigg et al.

Subject motion in whole-body dynamic PET introduces inter-frame mismatch and seriously impacts parametric imaging. Traditional non-rigid registration methods are generally computationally intense and time-consuming. Deep learning approaches are promising in achieving high accuracy with fast speed, but have yet been investigated with consideration for tracer distribution changes or in the whole-body scope. In this work, we developed an unsupervised automatic deep learning-based framework to correct inter-frame body motion. The motion estimation network is a convolutional neural network with a combined convolutional long short-term memory layer, fully utilizing dynamic temporal features and spatial information. Our dataset contains 27 subjects each under a 90-min FDG whole-body dynamic PET scan. With 9-fold cross-validation, compared with both traditional and deep learning baselines, we demonstrated that the proposed network obtained superior performance in enhanced qualitative and quantitative spatial alignment between parametric $K_{i}$ and $V_{b}$ images and in significantly reduced parametric fitting error. We also showed the potential of the proposed motion correction method for impacting downstream analysis of the estimated parametric images, improving the ability to distinguish malignant from benign hypermetabolic regions of interest. Once trained, the motion estimation inference time of our proposed network was around 460 times faster than the conventional registration baseline, showing its potential to be easily applied in clinical settings.

Mengxi Liu, Sizhen Bian, Bo Zhou et al.

This paper presents a novel beverage intake monitoring system that can accurately recognize beverage kinds and freshness. By mounting carbon electrodes on the commercial cup, the system measures the electrochemical impedance spectrum of the fluid in the cup. We studied the frequency sensitivity of the electrochemical impedance spectrum regarding distinct beverages and the importance of features like amplitude, phase, and real and imaginary components for beverage classification. The results show that features from a low-frequency domain (100 Hz to 1000 Hz) provide more meaningful information for beverage classification than the higher frequency domain. Twenty beverages, including carbonated drinks and juices, were classified with nearly perfect accuracy using a supervised machine learning approach. The same performance was also observed in the freshness recognition, where four different kinds of milk and fruit juice were studied.

Lala Shakti Swarup Ray, Vitor Fortes Rey, Bo Zhou et al.

We propose PressureTransferNet, a novel method for Human Activity Recognition (HAR) using ground pressure information. Our approach generates body-specific dynamic ground pressure profiles for specific activities by leveraging existing pressure data from different individuals. PressureTransferNet is an encoder-decoder model taking a source pressure map and a target human attribute vector as inputs, producing a new pressure map reflecting the target attribute. To train the model, we use a sensor simulation to create a diverse dataset with various human attributes and pressure profiles. Evaluation on a real-world dataset shows its effectiveness in accurately transferring human attributes to ground pressure profiles across different scenarios. We visually confirm the fidelity of the synthesized pressure shapes using a physics-based deep learning model and achieve a binary R-square value of 0.79 on areas with ground contact. Validation through classification with F1 score (0.911$\pm$0.015) on physical pressure mat data demonstrates the correctness of the synthesized pressure maps, making our method valuable for data augmentation, denoising, sensor simulation, and anomaly detection. Applications span sports science, rehabilitation, and bio-mechanics, contributing to the development of HAR systems.

Bo Zhou, Yingda Xia, Jiawen Yao et al.

Pancreatic cancer is one of the leading causes of cancer-related death. Accurate detection, segmentation, and differential diagnosis of the full taxonomy of pancreatic lesions, i.e., normal, seven major types of lesions, and other lesions, is critical to aid the clinical decision-making of patient management and treatment. However, existing works focus on segmentation and classification for very specific lesion types (PDAC) or groups. Moreover, none of the previous work considers using lesion prevalence-related non-imaging patient information to assist the differential diagnosis. To this end, we develop a meta-information-aware dual-path transformer and exploit the feasibility of classification and segmentation of the full taxonomy of pancreatic lesions. Specifically, the proposed method consists of a CNN-based segmentation path (S-path) and a transformer-based classification path (C-path). The S-path focuses on initial feature extraction by semantic segmentation using a UNet-based network. The C-path utilizes both the extracted features and meta-information for patient-level classification based on stacks of dual-path transformer blocks that enhance the modeling of global contextual information. A large-scale multi-phase CT dataset of 3,096 patients with pathology-confirmed pancreatic lesion class labels, voxel-wise manual annotations of lesions from radiologists, and patient meta-information, was collected for training and evaluations. Our results show that our method can enable accurate classification and segmentation of the full taxonomy of pancreatic lesions, approaching the accuracy of the radiologist's report and significantly outperforming previous baselines. Results also show that adding the common meta-information, i.e., gender and age, can boost the model's performance, thus demonstrating the importance of meta-information for aiding pancreatic disease diagnosis.

Huidong Xie, Bo Zhou, Xiongchao Chen et al.

Cardiovascular disease (CVD) is the leading cause of death worldwide, and myocardial perfusion imaging using SPECT has been widely used in the diagnosis of CVDs. The GE 530/570c dedicated cardiac SPECT scanners adopt a stationary geometry to simultaneously acquire 19 projections to increase sensitivity and achieve dynamic imaging. However, the limited amount of angular sampling negatively affects image quality. Deep learning methods can be implemented to produce higher-quality images from stationary data. This is essentially a few-view imaging problem. In this work, we propose a novel 3D transformer-based dual-domain network, called TIP-Net, for high-quality 3D cardiac SPECT image reconstructions. Our method aims to first reconstruct 3D cardiac SPECT images directly from projection data without the iterative reconstruction process by proposing a customized projection-to-image domain transformer. Then, given its reconstruction output and the original few-view reconstruction, we further refine the reconstruction using an image-domain reconstruction network. Validated by cardiac catheterization images, diagnostic interpretations from nuclear cardiologists, and defect size quantified by an FDA 510(k)-cleared clinical software, our method produced images with higher cardiac defect contrast on human studies compared with previous baseline methods, potentially enabling high-quality defect visualization using stationary few-view dedicated cardiac SPECT scanners.

Bo Zhou, Yu-Jung Tsai, Jiazhen Zhang et al.

Patient motion during PET is inevitable. Its long acquisition time not only increases the motion and the associated artifacts but also the patient's discomfort, thus PET acceleration is desirable. However, accelerating PET acquisition will result in reconstructed images with low SNR, and the image quality will still be degraded by motion-induced artifacts. Most of the previous PET motion correction methods are motion type specific that require motion modeling, thus may fail when multiple types of motion present together. Also, those methods are customized for standard long acquisition and could not be directly applied to accelerated PET. To this end, modeling-free universal motion correction reconstruction for accelerated PET is still highly under-explored. In this work, we propose a novel deep learning-aided motion correction and reconstruction framework for accelerated PET, called Fast-MC-PET. Our framework consists of a universal motion correction (UMC) and a short-to-long acquisition reconstruction (SL-Reon) module. The UMC enables modeling-free motion correction by estimating quasi-continuous motion from ultra-short frame reconstructions and using this information for motion-compensated reconstruction. Then, the SL-Recon converts the accelerated UMC image with low counts to a high-quality image with high counts for our final reconstruction output. Our experimental results on human studies show that our Fast-MC-PET can enable 7-fold acceleration and use only 2 minutes acquisition to generate high-quality reconstruction images that outperform/match previous motion correction reconstruction methods using standard 15 minutes long acquisition data.

Zefang Zong, Dingwei Chen, Yang Li et al.

LLM agents have emerged as powerful systems for tackling multi-turn tasks by interleaving internal reasoning and external tool interactions. Agentic Reinforcement Learning has recently drawn significant research attention as a critical post-training paradigm to further refine these capabilities. In this paper, we present AT$^2$PO (Agentic Turn-based Policy Optimization via Tree Search), a unified framework for multi-turn agentic RL that addresses three core challenges: limited exploration diversity, sparse credit assignment, and misaligned policy optimization. AT$^2$PO introduces a turn-level tree structure that jointly enables Entropy-Guided Tree Expansion for strategic exploration and Turn-wise Credit Assignment for fine-grained reward propagation from sparse outcomes. Complementing this, we propose Agentic Turn-based Policy Optimization, a turn-level learning objective that aligns policy updates with the natural decision granularity of agentic interactions. ATPO is orthogonal to tree search and can be readily integrated into any multi-turn RL pipeline. Experiments across seven benchmarks demonstrate consistent improvements over the state-of-the-art baseline by up to 1.84 percentage points in average, with ablation studies validating the effectiveness of each component. Our code is available at https://github.com/zzfoutofspace/ATPO.

Bo Zhou, Huidong Xie, Qiong Liu et al.

Low-count PET is an efficient way to reduce radiation exposure and acquisition time, but the reconstructed images often suffer from low signal-to-noise ratio (SNR), thus affecting diagnosis and other downstream tasks. Recent advances in deep learning have shown great potential in improving low-count PET image quality, but acquiring a large, centralized, and diverse dataset from multiple institutions for training a robust model is difficult due to privacy and security concerns of patient data. Moreover, low-count PET data at different institutions may have different data distribution, thus requiring personalized models. While previous federated learning (FL) algorithms enable multi-institution collaborative training without the need of aggregating local data, addressing the large domain shift in the application of multi-institutional low-count PET denoising remains a challenge and is still highly under-explored. In this work, we propose FedFTN, a personalized federated learning strategy that addresses these challenges. FedFTN uses a local deep feature transformation network (FTN) to modulate the feature outputs of a globally shared denoising network, enabling personalized low-count PET denoising for each institution. During the federated learning process, only the denoising network's weights are communicated and aggregated, while the FTN remains at the local institutions for feature transformation. We evaluated our method using a large-scale dataset of multi-institutional low-count PET imaging data from three medical centers located across three continents, and showed that FedFTN provides high-quality low-count PET images, outperforming previous baseline FL reconstruction methods across all low-count levels at all three institutions.

Hymalai Bello, Daniel Geißler, Sungho Suh et al.

Smaller machine learning models, with less complex architectures and sensor inputs, can benefit wearable sensor-based human activity recognition (HAR) systems in many ways, from complexity and cost to battery life. In the specific case of smart factories, optimizing human-robot collaboration hinges on the implementation of cutting-edge, human-centric AI systems. To this end, workers' activity recognition enables accurate quantification of performance metrics, improving efficiency holistically. We present a two-stage semantic-aware knowledge distillation (KD) approach, TSAK, for efficient, privacy-aware, and wearable HAR in manufacturing lines, which reduces the input sensor modalities as well as the machine learning model size, while reaching similar recognition performance as a larger multi-modal and multi-positional teacher model. The first stage incorporates a teacher classifier model encoding attention, causal, and combined representations. The second stage encompasses a semantic classifier merging the three representations from the first stage. To evaluate TSAK, we recorded a multi-modal dataset at a smart factory testbed with wearable and privacy-aware sensors (IMU and capacitive) located on both workers' hands. In addition, we evaluated our approach on OpenPack, the only available open dataset mimicking the wearable sensor placements on both hands in the manufacturing HAR scenario. We compared several KD strategies with different representations to regulate the training process of a smaller student model. Compared to the larger teacher model, the student model takes fewer sensor channels from a single hand, has 79% fewer parameters, runs 8.88 times faster, and requires 96.6% less computing power (FLOPS).

Lala Shakti Swarup Ray, Daniel Geißler, Mengxi Liu et al.

Despite the widespread integration of ambient light sensors (ALS) in smart devices commonly used for screen brightness adaptation, their application in human activity recognition (HAR), primarily through body-worn ALS, is largely unexplored. In this work, we developed ALS-HAR, a robust wearable light-based motion activity classifier. Although ALS-HAR achieves comparable accuracy to other modalities, its natural sensitivity to external disturbances, such as changes in ambient light, weather conditions, or indoor lighting, makes it challenging for daily use. To address such drawbacks, we introduce strategies to enhance environment-invariant IMU-based activity classifications through augmented multi-modal and contrastive classifications by transferring the knowledge extracted from the ALS. Our experiments on a real-world activity dataset for three different scenarios demonstrate that while ALS-HAR's accuracy strongly relies on external lighting conditions, cross-modal information can still improve other HAR systems, such as IMU-based classifiers.Even in scenarios where ALS performs insufficiently, the additional knowledge enables improved accuracy and macro F1 score by up to 4.2 % and 6.4 %, respectively, for IMU-based classifiers and even surpasses multi-modal sensor fusion models in two of our three experiment scenarios. Our research highlights the untapped potential of ALS integration in advancing sensor-based HAR technology, paving the way for practical and efficient wearable ALS-based activity recognition systems with potential applications in healthcare, sports monitoring, and smart indoor environments.

Hymalai Bello, Daniel Geißler, Lala Ray et al.

Artificial Intelligence (AI) methods are powerful tools for various domains, including critical fields such as avionics, where certification is required to achieve and maintain an acceptable level of safety. General solutions for safety-critical systems must address three main questions: Is it suitable? What drives the system's decisions? Is it robust to errors/attacks? This is more complex in AI than in traditional methods. In this context, this paper presents a comprehensive mind map of formal AI certification in avionics. It highlights the challenges of certifying AI development with an example to emphasize the need for qualification beyond performance metrics.

Ziqiao Weng, Jiancheng Yang, Kangxian Xie et al.

Pulmonary trees extracted from CT images frequently exhibit topological incompleteness, such as missing or disconnected branches, which substantially degrades downstream anatomical analysis and limits the applicability of existing pulmonary tree modeling pipelines. Current approaches typically rely on dense volumetric processing or explicit graph reasoning, leading to limited efficiency and reduced robustness under realistic structural corruption. We propose TopoField, a topology-aware implicit modeling framework that treats topology repair as a first-class modeling problem and enables unified multi-task inference for pulmonary tree analysis. TopoField represents pulmonary anatomy using sparse surface and skeleton point clouds and learns a continuous implicit field that supports topology repair without relying on complete or explicit disconnection annotations, by training on synthetically introduced structural disruptions over \textit{already} incomplete trees. Building upon the repaired implicit representation, anatomical labeling and lung segment reconstruction are jointly inferred through task-specific implicit functions within a single forward pass.Extensive experiments on the Lung3D+ dataset demonstrate that TopoField consistently improves topological completeness and achieves accurate anatomical labeling and lung segment reconstruction under challenging incomplete scenarios. Owing to its implicit formulation, TopoField attains high computational efficiency, completing all tasks in just over one second per case, highlighting its practicality for large-scale and time-sensitive clinical applications. Code and data will be available at https://github.com/HINTLab/TopoField.

Bo Zhou, Qiuxia Lai, Zeren Sun et al.

Robust 3D representation learning forms the perceptual foundation of spatial intelligence, enabling downstream tasks in scene understanding and embodied AI. However, learning such representations directly from unposed multi-view images remains challenging. Recent self-supervised methods attempt to unify geometry, appearance, and semantics in a feed-forward manner, but they often suffer from weak geometry induction, limited appearance detail, and inconsistencies between geometry and semantics. We introduce UniSplat, a feed-forward framework designed to address these limitations through three complementary components. First, we propose a dual-masking strategy that strengthens geometry induction in the encoder. By masking both encoder and decoder tokens, and targeting decoder masks toward geometry-rich regions, the model is forced to infer structural information from incomplete visual cues, yielding geometry-aware representations even under unposed inputs. Second, we develop a coarse-to-fine Gaussian splatting strategy that reduces appearance-semantics inconsistencies by progressively refining the radiance field. Finally, to enforce geometric-semantic consistency, we introduce a pose-conditioned recalibration mechanism that interrelates the outputs of multiple heads by re-projecting predicted 3D point and semantic maps into the image plane using estimated camera parameters, and aligning them with corresponding RGB and semantic predictions to ensure cross-task consistency, thereby resolving geometry-semantic mismatches. Together, these components yield unified 3D representations that are robust to unposed, sparse-view inputs and generalize across diverse tasks, laying a perceptual foundation for spatial intelligence.

Mengxi Liu, Sizhen Bian, Vitor Fortes et al.

Kolmogorov-Arnold Networks (KANs) have demonstrated an exceptional ability to learn complex functions on clean, low-dimensional data but struggle to maintain performance on noisy and imperfect real-world datasets. In contrast, conventional multi-layer perceptrons (MLPs) are far more tolerant to noise and computationally efficient. Replacing all MLP components with KANs in HAR models often degrades accuracy and computation efficiency, highlighting an open challenge: how to combine KANs' precision with MLPs' noise robustness and efficiency. To address this, we systematically explore various placements of KAN modules within deep HAR networks and propose a hybrid architecture that strategically synergizes the strengths of both paradigms, which uses a KAN-based input embedding layer, retains MLP layers for intermediate feature mixing, and introduces a specialized LarctanKAN module for final activity classification. Across eight public HAR datasets, the hybrid KAN-MLP model achieves an average macro F1 score relative improvement of 5.33\% compared pure-MLP model, significantly outperforming standalone KAN and MLP baselines. Furthermore, integrating this hybrid strategy into other state-of-the-art HAR architectures consistently boosts their performance. Our findings demonstrate that a carefully orchestrated combination of KAN, MLP, or other conventional neural components yields more robust and accurate HAR models for real-world wearable sensing environments.

Yuanqi Du, Botao Yu, Tianyu Liu et al.

There has been unprecedented interest in developing agents that expand the boundary of scientific discovery, primarily by optimizing quantitative objective functions specified by scientists. However, for grand challenges in science , these objectives are only imperfect proxies. We argue that automating objective function design is a central, yet unmet requirement for scientific discovery agents. In this work, we introduce the Scientific Autonomous Goal-evolving Agent (SAGA) to amend this challenge. SAGA employs a bi-level architecture in which an outer loop of LLM agents analyzes optimization outcomes, proposes new objectives, and converts them into computable scoring functions, while an inner loop performs solution optimization under the current objectives. This bi-level design enables systematic exploration of the space of objectives and their trade-offs, rather than treating them as fixed inputs. We demonstrate the framework through a broad spectrum of applications, including antibiotic design, inorganic materials design, functional DNA sequence design, and chemical process design, showing that automating objective formulation can substantially improve the effectiveness of scientific discovery agents.

Yiwen Pang, Bo Zhou, Changjin Li et al.

Robotic laboratories play a critical role in autonomous scientific discovery by enabling scalable, continuous experimental execution. Recent vision-language-action (VLA) models offer a promising foundation for robotic laboratories. However, scientific experiments typically involve long-horizon tasks composed of multiple atomic tasks, posing a fundamental challenge to existing VLA models. While VLA models fine-tuned for scientific tasks can reliably execute atomic experimental actions seen during training, they often fail to perform composite tasks formed by reordering and composing these known atomic actions. This limitation arises from a distributional mismatch between training-time atomic tasks and inference-time composite tasks, which prevents VLA models from executing necessary transitional operations between atomic tasks. To address this challenge, we propose an Agentic VLA Inference Plugin for Long-Horizon Tasks in Scientific Experiments. It introduces an LLM-based agentic inference mechanism that intervenes when executing sequential manipulation tasks. By performing explicit transition inference and generating transitional robotic action code, the proposed plugin guides VLA models through missing transitional steps, enabling reliable execution of composite scientific workflows without any additional training. This inference-only intervention makes our method computationally efficient, data-efficient, and well-suited for open-ended and long-horizon robotic laboratory tasks. We build 3D assets of scientific instruments and common scientific operating scenes within an existing simulation environment. In these scenes, we have verified that our method increases the average success rate per atomic task by 42\% during inference. Furthermore, we show that our method can be easily transferred from the simulation to real scientific laboratories.

Deepika Gurung, Lala Shakti Swarup Ray, Mengxi Liu et al.

Real time sensor based applications in pervasive computing require edge deployable models to ensure low latency privacy and efficient interaction. A prime example is sensor based human activity recognition where models must balance accuracy with stringent resource constraints. Yet many deep learning approaches treat temporal sensor signals as black box sequences overlooking spectral temporal structure while demanding excessive computation. We present SPECTRA a deployment first co designed spectral temporal architecture that integrates short time Fourier transform STFT feature extraction depthwise separable convolutions and channel wise self attention to capture spectral temporal dependencies under real edge runtime and memory constraints. A compact bidirectional GRU with attention pooling summarizes within window dynamics at low cost reducing downstream model burden while preserving accuracy. Across five public HAR datasets SPECTRA matches or approaches larger CNN LSTM and Transformer baselines while substantially reducing parameters latency and energy. Deployments on a Google Pixel 9 smartphone and an STM32L4 microcontroller further demonstrate end to end deployable realtime private and efficient HAR.

Xue Gong, Qi Yi, Ziyuan Nan et al.

Training Large Language Models (LLMs) for reasoning tasks is increasingly driven by Reinforcement Learning with Verifiable Rewards (RLVR), where Proximal Policy Optimization (PPO) provides a principled framework for stable policy updates. However, the practical application of PPO is hindered by unreliable advantage estimation in the sparse-reward RLVR regime. This issue arises because the sparse rewards in RLVR lead to inaccurate intermediate value predictions, which in turn introduce significant bias when aggregated at every token by Generalized Advantage Estimation (GAE). To address this, we introduce Segmental Advantage Estimation (SAE), which mitigates the bias that GAE can incur in RLVR. Our key insight is that aggregating $n$-step advantages at every token(as in GAE) is unnecessary and often introduces excessive bias, since individual tokens carry minimal information. Instead, SAE first partitions the generated sequence into coherent sub-segments using low-probability tokens as heuristic boundaries. It then selectively computes variance-reduced advantage estimates only from these information-rich segment transitions, effectively filtering out noise from intermediate tokens. Our experiments demonstrate that SAE achieves superior performance, with marked improvements in final scores, training stability, and sample efficiency. These gains are shown to be consistent across multiple model sizes, and a correlation analysis confirms that our proposed advantage estimator achieves a higher correlation with an approximate ground-truth advantage, justifying its superior performance.

Jinkui Hao, Gorkem Durak, Halil Ertugrul Aktas et al.

Coronary artery disease, the leading cause of cardiovascular mortality worldwide, can be assessed non-invasively by coronary computed tomography angiography (CCTA). Despite progress in automated CCTA analysis using deep learning, clinical translation is constrained by the scarcity of expert-annotated datasets. Furthermore, widely adopted label-free pretraining strategies, such as masked image modeling, are intrinsically biased toward global anatomical statistics, frequently failing to capture the spatially localized pathological features of coronary plaques. Here, we introduce CORA, a 3D vision foundation model for comprehensive cardiovascular risk assessment. CORA learns directly from volumetric CCTA via a pathology-centric, synthesis-driven self-supervised framework. By utilizing an anatomy-guided lesion synthesis engine, the model is explicitly trained to detect simulated vascular abnormalities, biasing representation learning toward clinically relevant disease features rather than dominant background anatomy. We trained CORA on a large-scale cohort of 12,801 unlabeled CCTA volumes and comprehensively evaluated the model across multi-center datasets from nine independent hospitals. Across diagnostic and anatomical tasks, including plaque characterization, stenosis detection, and coronary artery segmentation, CORA consistently outperformed the state-of-the-art 3D vision foundation models, achieving up to a 29\% performance gain. Crucially, by coupling the imaging encoder with a large language model, we extended CORA into a multimodal framework that significantly improved 30-day major adverse cardiac event (MACE) risk stratification. Our results establish CORA as a scalable and extensible foundation for unified anatomical assessment and cardiovascular risk prediction.

Xinran Gong, Gorkem Durak, Halil Ertugrul Aktas et al.

Coronary artery calcium (CAC) scoring from chest CT is a well-established tool to stratify and refine clinical cardiovascular disease risk estimation. CAC quantification relies on the accurate delineation of calcified lesions, but is oftentimes affected by artifacts introduced by cardiac and respiratory motion. ECG-gated cardiac CTs substantially reduce motion artifacts, but their use in population screening and routine imaging remains limited due to gating requirements and lack of insurance coverage. Although identification of incidental CAC from non-gated chest CT is increasingly considered for it offers an accessible and widely available alternative, this modality is limited by more severe motion artifacts. We present ProDM (Property-aware Progressive Correction Diffusion Model), a generative diffusion framework that restores motion-free calcified lesions from non-gated CTs. ProDM introduces three key components: (1) a CAC motion simulation data engine that synthesizes realistic non-gated acquisitions with diverse motion trajectories directly from cardiac-gated CTs, enabling supervised training without paired data; (2) a property-aware learning strategy incorporating calcium-specific priors through a differentiable calcium consistency loss to preserve lesion integrity; and (3) a progressive correction scheme that reduces artifacts gradually across diffusion steps to enhance stability and calcium fidelity. Experiments on real patient datasets show that ProDM significantly improves CAC scoring accuracy, spatial lesion fidelity, and risk stratification performance compared with several baselines. A reader study on real non-gated scans further confirms that ProDM suppresses motion artifacts and improves clinical usability. These findings highlight the potential of progressive, property-aware frameworks for reliable CAC quantification from routine chest CT imaging.

Minda Hu, Zexuan Qiu, Zenan Xu et al.

Recent breakthroughs in Large Reasoning Models (LRMs) have demonstrated that extensive Chain-of-Thought (CoT) generation is critical for enabling intricate cognitive behaviors, such as self-verification and backtracking, to solve complex tasks. However, this capability often leads to ``overthinking'', where models generate redundant reasoning paths that inflate computational costs without improving accuracy. While Supervised Fine-Tuning (SFT) on reasoning traces is a standard paradigm for the 'cold start' phase, applying existing compression techniques to these traces often compromises logical coherence or incurs prohibitive sampling costs. In this paper, we introduce ConMax (Confidence-Maximizing Compression), a novel reinforcement learning framework designed to automatically compress reasoning traces while preserving essential reasoning patterns. ConMax formulates compression as a reward-driven optimization problem, training a policy to prune redundancy by maximizing a weighted combination of answer confidence for predictive fidelity and thinking confidence for reasoning validity through a frozen auxiliary LRM. Extensive experiments across five reasoning datasets demonstrate that ConMax achieves a superior efficiency-performance trade-off. Specifically, it reduces inference length by 43% over strong baselines at the cost of a mere 0.7% dip in accuracy, proving its effectiveness in generating high-quality, efficient training data for LRMs.

Yueqi Zhu, Yan Pan, Chufan Rui et al.

In safety-critical scenarios, the protection level of the autonomous navigation system is crucial for enabling mobile robots to perform safe tasks. However, existing studies on probabilistic navigation systems for robots usually perform offline accuracy evaluations using limited datasets and assume that the results can be applied to unknown real-world environments. As a result, current autonomous mobile robots often lack protection levels for online safety assessment. To fill this gap, we propose a safety-critical LiDAR-inertial odometry (LIO) that provides deterministic protection levels based on on-manifold deterministic state estimation. By adopting the unknown but bounded assumption, we derive a neat closed-form relationship between point cloud noise and the uncertainty of the estimation from the iterated closest point algorithm. Using this relationship, we design an on-manifold ellipsoidal set-membership filter and implement it within the LIO system. Leveraging the properties of the set-membership filter, our system offers the feasible sets of the estimated locations as the deterministic protection levels, serving as safety references for the robots' downstream autonomous operations. The experimental results show that our system can provide effective deterministic online safety references for diverse robots in various environments.

Zifei Dong, Wenjie Wu, Jinkui Hao et al.

Robust anatomical segmentation of chest X-rays (CXRs) remains challenging due to the scarcity of comprehensive annotations and the substantial variability of real-world acquisition conditions. We propose AnyCXR, a unified framework that enables generalizable multi-organ segmentation across arbitrary CXR projection angles using only synthetic supervision. The method combines a Multi-stage Domain Randomization (MSDR) engine, which generates over 100,000 anatomically faithful and highly diverse synthetic radiographs from 3D CT volumes, with a Conditional Joint Annotation Regularization (CAR) learning strategy that leverages partial and imperfect labels by enforcing anatomical consistency in a latent space. Trained entirely on synthetic data, AnyCXR achieves strong zero-shot generalization on multiple real-world datasets, providing accurate delineation of 54 anatomical structures in PA, lateral, and oblique views. The resulting segmentation maps support downstream clinical tasks, including automated cardiothoracic ratio estimation, spine curvature assessment, and disease classification, where the incorporation of anatomical priors improves diagnostic performance. These results demonstrate that AnyCXR establishes a scalable and reliable foundation for anatomy-aware CXR analysis and offers a practical pathway toward reducing annotation burdens while improving robustness across diverse imaging conditions.

Taiqiang Wu, Zenan Xu, Bo Zhou et al.

Large Language Models (LLMs) consistently benefit from scaled Chain-of-Thought (CoT) reasoning, but also suffer from heavy computational overhead. To address this issue, efficient reasoning aims to incentivize short yet accurate thinking trajectories, typically through reward shaping with Reinforcement Learning (RL). In this paper, we systematically investigate the mechanics of efficient reasoning for LLMs. For comprehensive evaluation, we advocate for more fine-grained metrics, including length distribution conditioned on correctness and performance across a wide spectrum of token budgets ranging from 2k to 32k. First, we reveal that the training process follows a two-stage paradigm: length adaptation and reasoning refinement. After that, we conduct extensive experiments (about 0.2 million GPU hours) in a unified protocol, deconstructing training prompts and rollouts, reward shaping, and optimization strategies. In particular, a key finding is to train on relatively easier prompts, ensuring the density of positive reward signals and thus avoiding the length collapse. Meanwhile, the learned length bias can be generalized across domains. We distill all findings into valuable insights and practical guidelines, and further validate them across the Qwen3 series, ranging from 0.6B to 30B, demonstrating the robustness and generalization.

Guanhua Huang, Tingqiang Xu, Mingze Wang et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has propelled Large Language Models in complex reasoning, yet its scalability is often hindered by a training bottleneck where performance plateaus as policy entropy collapses, signaling a loss of exploration. Previous methods typically address this by maintaining high policy entropy, yet the precise mechanisms that govern meaningful exploration have remained underexplored. Our analysis suggests that an unselective focus on entropy risks amplifying irrelevant tokens and destabilizing training. This paper investigates the exploration dynamics within RLVR and identifies a key issue: the gradual elimination of valuable low-probability exploratory tokens, which we term \textbf{\textit{reasoning sparks}}. We find that while abundant in pre-trained models, these sparks are systematically extinguished during RLVR due to over-penalization, leading to a degeneracy in exploration. To address this, we introduce Low-probability Regularization (Lp-Reg). Its core mechanism regularizes the policy towards a heuristic proxy distribution. This proxy is constructed by filtering out presumed noise tokens and re-normalizing the distribution over the remaining candidates. The result is a less-noisy proxy where the probability of \textit{reasoning sparks} is amplified, which then serves as a soft regularization target to shield these valuable tokens from elimination via KL divergence. Experiments show that Lp-Reg enables stable on-policy RL, sustaining continuous scaling across $3,000$ training steps and $81,204$ GPU-hours, where baseline entropy-control methods collapse. This sustained exploration leads to state-of-the-art performance, achieving a $60.17\%$ average accuracy on five math benchmarks, an improvement of $2.66\%$ over prior methods. Code is available at https://github.com/CarlanLark/Lp-Reg.

Pixi Kang, Julian Moosmann, Mengxi Liu et al.

Human Activity Recognition (HAR) with different sensing modalities requires both strong generalization across diverse users and efficient personalization for individuals. However, conventional HAR models often fail to generalize when faced with user-specific variations, leading to degraded performance. To address this challenge, we propose a novel on-device few-shot learning framework that bridges generalization and personalization in HAR. Our method first trains a generalizable representation across users and then rapidly adapts to new users with only a few labeled samples, updating lightweight classifier layers directly on resource-constrained devices. This approach achieves robust on-device learning with minimal computation and memory cost, making it practical for real-world deployment. We implement our framework on the energy-efficient RISC-V GAP9 microcontroller and evaluate it on three benchmark datasets (RecGym, QVAR-Gesture, Ultrasound-Gesture). Across these scenarios, post-deployment adaptation improves accuracy by 3.73\%, 17.38\%, and 3.70\%, respectively. These results demonstrate that few-shot on-device learning enables scalable, user-aware, and energy-efficient wearable human activity recognition by seamlessly uniting generalization and personalization. The related framework is open sourced for further research\footnote{https://github.com/kangpx/onlineTiny2023}.

Xiongchao Chen, Bo Zhou, Xueqi Guo et al.

Single-Photon Emission Computed Tomography (SPECT) is widely applied for the diagnosis of coronary artery diseases. Low-dose (LD) SPECT aims to minimize radiation exposure but leads to increased image noise. Limited-view (LV) SPECT, such as the latest GE MyoSPECT ES system, enables accelerated scanning and reduces hardware expenses but degrades reconstruction accuracy. Additionally, Computed Tomography (CT) is commonly used to derive attenuation maps ($μ$-maps) for attenuation correction (AC) of cardiac SPECT, but it will introduce additional radiation exposure and SPECT-CT misalignments. Although various methods have been developed to solely focus on LD denoising, LV reconstruction, or CT-free AC in SPECT, the solution for simultaneously addressing these tasks remains challenging and under-explored. Furthermore, it is essential to explore the potential of fusing cross-domain and cross-modality information across these interrelated tasks to further enhance the accuracy of each task. Thus, we propose a Dual-Domain Coarse-to-Fine Progressive Network (DuDoCFNet), a multi-task learning method for simultaneous LD denoising, LV reconstruction, and CT-free $μ$-map generation of cardiac SPECT. Paired dual-domain networks in DuDoCFNet are cascaded using a multi-layer fusion mechanism for cross-domain and cross-modality feature fusion. Two-stage progressive learning strategies are applied in both projection and image domains to achieve coarse-to-fine estimations of SPECT projections and CT-derived $μ$-maps. Our experiments demonstrate DuDoCFNet's superior accuracy in estimating projections, generating $μ$-maps, and AC reconstructions compared to existing single- or multi-task learning methods, under various iterations and LD levels. The source code of this work is available at https://github.com/XiongchaoChen/DuDoCFNet-MultiTask.

Bo Zhou, Qianglong Chen, Tianyu Wang et al.

To fully evaluate the overall performance of different NLP models in a given domain, many evaluation benchmarks are proposed, such as GLUE, SuperGLUE and CLUE. The fi eld of natural language understanding has traditionally focused on benchmarks for various tasks in languages such as Chinese, English, and multilingua, however, there has been a lack of attention given to the area of classical Chinese, also known as "wen yan wen", which has a rich history spanning thousands of years and holds signifi cant cultural and academic value. For the prosperity of the NLP community, in this paper, we introduce the WYWEB evaluation benchmark, which consists of nine NLP tasks in classical Chinese, implementing sentence classifi cation, sequence labeling, reading comprehension, and machine translation. We evaluate the existing pre-trained language models, which are all struggling with this benchmark. We also introduce a number of supplementary datasets and additional tools to help facilitate further progress on classical Chinese NLU. The github repository is https://github.com/baudzhou/WYWEB.

Bo Zhou, Zachary Augenfeld, Julius Chapiro et al.

Multimodal image registration has many applications in diagnostic medical imaging and image-guided interventions, such as Transcatheter Arterial Chemoembolization (TACE) of liver cancer guided by intraprocedural CBCT and pre-operative MR. The ability to register peri-procedurally acquired diagnostic images into the intraprocedural environment can potentially improve the intra-procedural tumor targeting, which will significantly improve therapeutic outcomes. However, the intra-procedural CBCT often suffers from suboptimal image quality due to lack of signal calibration for Hounsfield unit, limited FOV, and motion/metal artifacts. These non-ideal conditions make standard intensity-based multimodal registration methods infeasible to generate correct transformation across modalities. While registration based on anatomic structures, such as segmentation or landmarks, provides an efficient alternative, such anatomic structure information is not always available. One can train a deep learning-based anatomy extractor, but it requires large-scale manual annotations on specific modalities, which are often extremely time-consuming to obtain and require expert radiological readers. To tackle these issues, we leverage annotated datasets already existing in a source modality and propose an anatomy-preserving domain adaptation to segmentation network (APA2Seg-Net) for learning segmentation without target modality ground truth. The segmenters are then integrated into our anatomy-guided multimodal registration based on the robust point matching machine. Our experimental results on in-house TACE patient data demonstrated that our APA2Seg-Net can generate robust CBCT and MR liver segmentation, and the anatomy-guided registration framework with these segmenters can provide high-quality multimodal registrations. Our code is available at https://github.com/bbbbbbzhou/APA2Seg-Net.

Anton Smerdov, Bo Zhou, Paul Lukowicz et al.

Proper training and analytics in eSports require accurately collected and annotated data. Most eSports research focuses exclusively on in-game data analysis, and there is a lack of prior work involving eSports athletes' psychophysiological data. In this paper, we present a dataset collected from professional and amateur teams in 22 matches in League of Legends video game with more than 40 hours of recordings. Recorded data include the players' physiological activity, e.g. movements, pulse, saccades, obtained from various sensors, self-reported aftermatch survey, and in-game data. An important feature of the dataset is simultaneous data collection from five players, which facilitates the analysis of sensor data on a team level. Upon the collection of dataset we carried out its validation. In particular, we demonstrate that stress and concentration levels for professional players are less correlated, meaning more independent playstyle. Also, we show that the absence of team communication does not affect the professional players as much as amateur ones. To investigate other possible use cases of the dataset, we have trained classical machine learning algorithms for skill prediction and player re-identification using 3-minute sessions of sensor data. Best models achieved 0.856 and 0.521 (0.10 for a chance level) accuracy scores on a validation set for skill prediction and player re-id problems, respectively. The dataset is available at https://github.com/smerdov/eSports Sensors Dataset.