As individuals age, simultaneous declines in cognitive function, physical ability, and visual capacity (e.g., dynamic visual acuity) pose challenges to maintaining functional independence in later life. Despite advances in immersive virtual environments for age-related functional support, dynamic visual acuity remains underexplored, particularly in relation to its interaction with cognitive and motor processes. Meanwhile, existing interventions often rely solely on performance-based difficulty adjustment, neglecting the dual need to maintain engagement and promote skill progression. To address these gaps, we developed the Pareto-based Dynamic Difficulty Adjustment for Cross-domain Co-training in Virtual Reality (CCVR-PDDA) system, integrating psychology paradigm (e.g., Stroop task), upper-limb motor training (e.g., arm lifting and raising), and dynamic visual exercises (e.g., multi-directional eye movements). A 12-week longitudinal study involving 60 older adults (≥65 years) demonstrated significant cognitive improvements, sustained at six-month follow-up. Qualitative feedback further supported the system’s usability, motivation enhancement, and long-term engagement potential. These findings underscore the efficacy and acceptability of CCVR-PDDA in promoting cognitive health in aging populations.

This paper presents PipeFusion, an innovative parallel methodology to tackle the high latency issues associated with generating high-resolution images using diffusion transformers (DiTs) models. PipeFusion partitions images into patches and the model layers across multiple GPUs. It employs a patch-level pipeline parallel strategy to orchestrate communication and computation efficiently. By capitalizing on the high similarity between inputs from successive diffusion steps, PipeFusion reuses one-step stale feature maps to provide context for the current pipeline step. This approach notably reduces communication costs compared to existing DiTs inference parallelism, including tensor parallel, sequence parallel and DistriFusion. PipeFusion enhances memory efficiency through parameter distribution across devices, ideal for large DiTs like Flux.1. Experimental results demonstrate that PipeFusion achieves state-of-the-art performance on 8$\times$L40 PCIe GPUs for Pixart, Stable-Diffusion 3, and Flux.1 models. Our Source code is available at \url{https://github.com/xdit-project/xDiT}.

Early diagnosis of mild cognitive impairment (MCI) is pivotal in mitigating the risk of cognitive impairment and the onset of dementia. However, prevailing clinical cognitive screening tests and biomarker assessment approaches often suffer from drawbacks such as high cost, invasiveness, time consumption, or subjectivity. In China, existing digital cognitive assessment tools face multiple challenges due to their distinctive cultural, language, and healthcare landscape. Therefore, we embarked on a study to develop a digital cognitive assessment tool, evaluate its efficacy in distinguishing between healthy individuals and MCI patients, and examine its acceptability among Chinese older adults. Through a series of symposiums, programming, and interviews with stakeholders, we iteratively designed the "BrainNursing" mobile application. The system consists of eleven single tasks and three dual tasks, each taking only 1–3 minutes. Subsequently, we conducted statistical comparisons of movement kinetics and physiological signals recorded during cognitive testing in 181 older adults to investigate which parameters could serve as effective digital biomarkers for MCI screening. Leveraging machine learning classifiers and a majority voting principle, we evaluated the classification performance of the BrainNursing system in detecting MCI, yielding an accuracy rate of 90.3%. Furthermore, our analysis of movement kinetics revealed that time, score, and sequence features are crucial in cognitive assessment. Contrasting with their healthy counterparts, MCI patients exhibited decreased heart rate variability, increased sympathetic nervous system activity, and weakened autonomic nervous system regulation in response to stimuli during cognitive testing. Finally, user experience feedback indicated that participants universally perceived BrainNursing as a convenient and user-friendly cognitive screening tool and provided valuable insights for further improvement.

Web applications have increasingly adopted Deep Learning (DL) through in-browser inference , wherein DL inference performs directly within Web browsers. The actual performance of in-browser inference and its impacts on the Quality of Experience ( QoE ) remain unexplored, and urgently require new QoE measurements beyond traditional ones, e.g., mainly focusing on page load time. To bridge this gap, we make the first comprehensive performance measurement of in-browser inference to date. Our approach proposes new metrics to measure in-browser inference: responsiveness, smoothness, and inference accuracy. Our extensive analysis involves 9 representative DL models across Web browsers of 50 popular PC devices and 20 mobile devices. The results reveal that in-browser inference exhibits a substantial latency gap, averaging 16.9 times slower on CPU and 4.9 times slower on GPU compared to native inference on PC devices. The gap on mobile CPU and mobile GPU is 15.8 times and 7.8 times, respectively. Furthermore, we identify contributing factors to such latency gap, including underutilized hardware instruction sets, inherent overhead in the runtime environment, resource contention within the browser, and inefficiencies in software libraries and GPU abstractions. Additionally, in-browser inference imposes significant memory demands, at times exceeding 334.6 times the size of the DL models themselves, partly attributable to suboptimal memory management. We also observe that in-browser inference leads to a significant 67.2% increase in the time it takes for GUI components to render within Web browsers, significantly affecting the overall user QoE of Web applications reliant on this technology.

Given the prevalence of cognitive impairment in older adults and its frequent misdiagnosis or delayed diagnosis, non-pharmacological interventions have been proposed as solutions, which include cognitive training, prevention or risk reduction of dementia using virtual reality (VR) technology. This study aimed to investigate the effects of a virtual reality cognitive-motor training intervention (VRCMTI) on improving cognitive and physical function in older adults with cognitive impairment. We co-designed the VRCMTI system with multiple stakeholders by organizing symposiums and conducting pilot evaluations. The VRCMTI consisted of three virtual cognitive tasks and three upper limb movement tasks focused on improving working memory, spatial cognition, attention shifting, executive control, joint flexibility, and coordination. One hour of brain cognition and upper limb motor training was performed each week during the 12-week intervention. Sixty older adults were included in the study and randomly assigned to either the VR group or the control group. Participants in the control group received usual care and would not undergo any intervention training. Task independence, accuracy and time were measured during each session. The results showed that the VR group significantly improved global cognitive ability scores compared to the control group, especially in attention and verbal cognition. In addition, older adults in the VR group also showed significant improvements in upper limb motor skills, driven primarily by movement quality and processing speed. The findings suggest that VRCMTI could improve cognitive function and enhance motor performance. When implemented with routine health care for older adults, this practical and effective intervention may be an appropriate complementary strategy for maintaining cognitive health and preventing motor deterioration.

Mild cognitive impairment is associated with many neurodegenerative diseases. It is essential to detect mild cognitive impairment on time to reduce the prevalence of such disorders. Nevertheless, present clinically employed test scales and biomarkers are time-consuming, user-unfriendly, and expensive. Hence, we developed a neuropsychological test system based on virtual reality in this study, the Virtual Reality Neuropsychological Mild Cognitive Impairment Test (VRNPT). The diagnosis and classification of MCI were achieved by effectively combining digital cognitive parameters and EEG signal features obtained during the VRNPT cognitive task. The VRNPT contains three head-mounted display-based cognitive tasks that assess participants’ attention, memory, spatial perception, working memory, and visuospatial executive ability across multiple cognitive domains of functioning. We investigated how to design and optimize these tasks. We conducted a field study by recruiting 80 participants (40 MCI patients and 40 normal older adults). The results showed that the classification accuracy of combining digitized cognitive parameters and EEG signals during VRNPT was 91.3%, higher than using only digitized parameters from VRNPT and applying EEG signals alone, demonstrating the validity and feasibility of this method for diagnosing MCI. The user satisfaction survey showed that the subjects were satisfied with VRNPT.

Network binarization emerges as one of the most promising compression approaches offering extraordinary computation and memory savings by minimizing the bit-width. However, recent research has shown that applying existing binarization algorithms to diverse tasks, architectures, and hardware in realistic scenarios is still not straightforward. Common challenges of binarization, such as accuracy degradation and efficiency limitation, suggest that its attributes are not fully understood. To close this gap, we present BiBench, a rigorously designed benchmark with in-depth analysis for network binarization. We first carefully scrutinize the requirements of binarization in the actual production and define evaluation tracks and metrics for a comprehensive and fair investigation. Then, we evaluate and analyze a series of milestone binarization algorithms that function at the operator level and with extensive influence. Our benchmark reveals that 1) the binarized operator has a crucial impact on the performance and deployability of binarized networks; 2) the accuracy of binarization varies significantly across different learning tasks and neural architectures; 3) binarization has demonstrated promising efficiency potential on edge devices despite the limited hardware support. The results and analysis also lead to a promising paradigm for accurate and efficient binarization. We believe that BiBench will contribute to the broader adoption of binarization and serve as a foundation for future research. The code for our BiBench is released https://github.com/htqin/BiBench .

Among various supervised deep metric learning methods proxy-based approaches have achieved high retrieval accuracies. Proxies, which are class-representative points in an embedding space, receive updates based on proxy-sample similarities in a similar manner to sample representations. In existing methods, a relatively small number of samples can produce large gradient magnitudes (i.e., hard samples), and a relatively large number of samples can produce small gradient magnitudes (i.e., easy samples); these can play a major part in updates. Assuming that acquiring too much sensitivity to such extreme sets of samples would deteriorate the generalizability of a method, we propose a novel proxy-based method called Informative Sample-Aware Proxy (Proxy-ISA), which directly modifies a gradient weighting factor for each sample using a scheduled threshold function, so that the model is more sensitive to the informative samples. Extensive experiments on the CUB-200-2011, Cars-196, Stanford Online Products and In-shop Clothes Retrieval datasets demonstrate the superiority of Proxy-ISA compared with the state-of-the-art methods.