The application of loss reweighting in modern deep learning presents a nuanced picture. While it fails to alter the terminal learning phase in overparameterized deep neural networks (DNNs) trained on high-dimensional datasets, empirical evidence consistently shows it offers significant benefits early in training. To transparently demonstrate and analyze this phenomenon, we introduce a small-scale model (SSM). This model is specifically designed to abstract the inherent complexities of both the DNN architecture and the input data, while maintaining key information about the structure of imbalance within its spectral components. On the one hand, the SSM reveals how vanilla empirical risk minimization preferentially learns to distinguish majority classes over minorities early in training, consequently delaying minority learning. In stark contrast, reweighting restores balanced learning dynamics, enabling the simultaneous learning of features associated with both majorities and minorities.

Power management, while indispensable for the working of battery-free devices on fragile ambient energy, unfortunately, also entails excessive workloads that consume the scarce harvested energy. Existing efforts aimed at addressing this typically manage to tackle only a fraction of the challenges, leaving power management as a painful Achilles’ heel for battery-free devices. In this paper, we systematically analyze the full-flow of power management and propose μ Man, a painless architecture with no extra workload on battery-free devices. That is, we shift the entire workload of power management from the resource-constrained battery-free devices to the resource-rich gateway. For this goal, we design a near-zero-power sampling-free monitoring mechanism to transparently piggyback the power status of the device directly onto the uplink signal waveform. Based on these real-time statuses, the gateway can take over the required computation and issue the resultant energy allocations back to devices. The design is fully transparent to the devices, and the devices can even remain in deep sleep during the whole process to minimize energy consumption. The experiments show that μ Man can reduce the energy consumption of power management by 97.2%, improve the power efficiency by 53%, and reduce the minimum energy requirements for the device start-up by 5.8 ×.

Battery-free nodes harvest ambient energy, accelerating large-scale IoT (Internet of Things) deployment. However, sporadic beginnings and ends of power failures impede collaboration, obstructing the execution of complex applications. The prior collaborative protocols have high energy demands and lack scalability. This paper introduces Cupid, a novel scheduling architecture that employs a coordinator device to circumvent the collaborative energy bottleneck, enhancing the scalability of battery-free node collaboration. Cupid employs an efficient crosslayer communication protocol to offload energy-intensive tasks to the coordinator. To reduce latency from non-local execution, we propose a predictive scheduling algorithm based on curve fitting. Additionally, we implement a circuit on the node side for ultra-low-power upload and download capabilities. We implement a prototype and conduct extensive evaluations. Compared to the state-of-the-art, it is the first to achieve intermittent coordination in medium-scale EH-WSNs, reducing latency by 94.56%.

Storing open-source fine-tuned models separately introduces redundancy and increases response times in applications utilizing multiple models. Delta-parameter pruning (DPP), particularly the random drop and rescale (DARE) method proposed by Yu et al., addresses this by pruning the majority of delta parameters—the differences between fine-tuned and pre-trained model weights—while typically maintaining minimal performance loss. However, DARE fails when either the pruning rate or the magnitude of the delta parameters is large. We highlight two key reasons for this failure: (1) an excessively large rescaling factor as pruning rates increase, and (2) high mean and variance in the delta parameters. To push DARE’s limits, we introduce DAREx (DARE the eXtreme), which features two algorithmic improvements: (1) DAREx-q, a rescaling factor modification that significantly boosts performance at high pruning rates (e.g., > 30% on COLA and SST2 for encoder models, with even greater gains in decoder models), and (2) DAREx-L2, which combines DARE with AdamR, an in-training method that applies appropriate delta regularization before DPP. We also demonstrate that DAREx-q can be seamlessly combined with vanilla parameter-efficient fine-tuning techniques like LoRA and can facilitate structural DPP. Additionally, we revisit the application of importance-based pruning techniques within DPP, demonstrating that they outperform random-based methods when delta parameters are large. Through this comprehensive study, we develop a pipeline for selecting the most appropriate DPP method under various practical scenarios.

The Internet of Things (IoT) can only thrive if IoT sensor nodes can be effortlessly deployed and maintained without compromising their general-purpose nature. However, existing low-power sensor systems fail to strike a balance between these two issues, leaving the widespread of IoT sensor nodes as an open problem. In this paper, we propose LEGO+ as a minimalist yet general-purpose sensing edge-end architecture. Instead of running embedded software on a redundant general-purpose microprocessor, LEGO+ can directly construct the desired control functionality for various IoT sensing applications through hardware-level logic orchestration. To achieve this, we first conduct an in-depth analysis of the underlying unit behaviors within IoT sensor systems and, based on this, abstract a uniform logic orchestration model. Next, to enable sensor nodes to comprehend and execute the generated logic, we devise a hierarchical atomic control circuit with negligible overheads. Finally, we develop a task state prediction scheme to further improve the overall operation efficiency among multiple nodes. We prototype LEGO+ for proof-of-concept and conduct comprehensive experiments, and the results demonstrate that LEGO+ can reduce the overall power consumption of sensor nodes by 86% and enhance task efficiency by 49%, thereby facilitating a wider array of IoT sensing applications.

Discovering reliable and informative relationships among brain regions from functional magnetic resonance imaging (fMRI) signals is essential in phenotypic predictions in neuroscience. Most of the current methods fail to accurately characterize those interactions because they only focus on pairwise connections and overlook the high-order relationships of brain regions. We propose that these high-order relationships should be *maximally informative and minimally redundant* (MIMR). However, identifying such high-order relationships is challenging and under-explored due to the exponential search space and the absence of a tractable objective. In response to this gap, we propose a novel method named HyBRiD, which aims to extract MIMR high-order relationships from fMRI data. HyBRiD employs a Constructor to identify hyperedge structures, and a Weighter to compute a weight for each hyperedge, which avoids searching in exponential space. HyBRiD achieves the MIMR objective through an innovative information bottleneck framework named multi-head drop-bottleneck with theoretical guarantees. Our comprehensive experiments demonstrate the effectiveness of our model. Our model outperforms the state-of-the-art predictive model by an average of 11.2%, regarding the quality of hyperedges measured by CPM, a standard protocol for studying brain connections.

Bluetooth Low Energy (BLE) Mesh is a pivotal multi-hop self-organizing network in the Internet of Things (IoT) domain, offering low power consumption, low cost, and robustness. This paper presents a comprehensive study on the communication performance of BLE-Mesh using commercial off-the-shelf devices, focusing on the impact of key mesh parameters such as transmission power, packet interval, and network structure on performance. Through extensive indoor and outdoor experiments, we quantify the impact of these parameters and conduct a detailed study. Our findings provide insights into the actual communication range of BLE-Mesh, the effect of node design on overall network performance, and the configuration for optimal performance. The research contributes to the establishment of a BLE-Mesh network in real-world environments, answering critical questions for practitioners, and offering a reference for future BLE-Mesh deployments. This work furthers our understanding of the characteristics, challenges, and future directions of BLE-Mesh, setting the stage for advancements in IoT applications such as smart offices and homes.

Graph theoretical measures have frequently been used to study disrupted connectivity in Alzheimer's disease human brain connectomes. However, prior studies have noted that differences in graph creation methods are confounding factors that may alter the topological observations found in these measures. In this study, we conduct a novel investigation regarding the effect of parcellation scale on graph theoretical measures computed for fiber density networks derived from diffusion tensor imaging. We computed 4 network-wide graph theoretical measures of average clustering coefficient, transitivity, characteristic path length, and global efficiency, and we tested whether these measures are able to consistently identify group differences among healthy control (HC), mild cognitive impairment (MCI), and AD groups in the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort across 5 scales of the Lausanne parcellation. We found that the segregative measure of transtivity offered the greatest consistency across scales in distinguishing between healthy and diseased groups, while the other measures were impacted by the selection of scale to varying degrees. Global efficiency was the second most consistent measure that we tested, where the measure could distinguish between HC and MCI in all 5 scales and between HC and AD in 3 out of 5 scales. Characteristic path length was highly sensitive to the variation in scale, corroborating previous findings, and could not identify group differences in many of the scales. Average clustering coefficient was also greatly impacted by scale, as it consistently failed to identify group differences in the higher resolution parcellations. From these results, we conclude that many graph theoretical measures are sensitive to the selection of parcellation scale, and further development in methodology is needed to offer a more robust characterization of AD's relationship with disrupted connectivity.

Support vector machine (SVM) is a popular classification method for the analysis of wide range of data including big data. Many SVM methods with feature selection have been developed under frequentist regularization or Bayesian shrinkage frameworks. On the other hand, the importance of incorporating a priori known biological knowledge, such as gene pathway information which stems from the gene regulatory network, into the statistical analysis of genomic data has been recognized in recent years. In this article, we propose a new Bayesian SVM approach that enables the feature selection to be guided by the knowledge on the graphical structure among predictors. The proposed method uses the spike-and-slab prior for feature selection, combined with the Ising prior that encourages group-wise selection of the predictors adjacent to each other on the known graph. Gibbs sampling algorithm is used for Bayesian inference. The performance of our method is evaluated and compared with existing SVM methods in terms of prediction and feature selection in extensive simulation settings. In addition, our method is illustrated in the analysis of genomic data from a cancer study, demonstrating its advantage in generating biologically meaningful results and identifying potentially important features.

Ultra-high dimensional variable selection has become increasingly important in analysis of neuroimaging data. For example, in the Autism Brain Imaging Data Exchange (ABIDE) study, neuroscientists are interested in identifying important biomarkers for early detection of the autism spectrum disorder (ASD) using high resolution brain images that include hundreds of thousands voxels. However, most existing methods are not feasible for solving this problem due to their extensive computational costs. In this work, we propose a novel multiresolution variable selection procedure under a Bayesian probit regression framework. It recursively uses posterior samples for coarser-scale variable selection to guide the posterior inference on finer-scale variable selection, leading to very efficient Markov chain Monte Carlo (MCMC) algorithms. The proposed algorithms are computationally feasible for ultra-high dimensional data. Also, our model incorporates two levels of structural information into variable selection using Ising priors: the spatial dependence between voxels and the functional connectivity between anatomical brain regions. Applied to the resting state functional magnetic resonance imaging (R-fMRI) data in the ABIDE study, our methods identify voxel-level imaging biomarkers highly predictive of the ASD, which are biologically meaningful and interpretable. Extensive simulations also show that our methods achieve better performance in variable selection compared to existing methods.

MOTIVATION: Network marker selection on genome-scale networks plays an important role in the understanding of biological mechanisms and disease pathologies. Recently, a Bayesian nonparametric mixture model has been developed and successfully applied for selecting genes and gene sub-networks. Hence, extending this method to a unified approach for network-based feature selection on general large-scale networks and creating an easy-to-use software package is on demand. RESULTS: We extended the method and developed an R package, the Bayesian network feature finder (BANFF), providing a package of posterior inference, model comparison and graphical illustration of model fitting. The model was extended to a more general form, and a parallel computing algorithm for the Markov chain Monte Carlo -based posterior inference and an expectation maximization-based algorithm for posterior approximation were added. Based on simulation studies, we demonstrate the use of BANFF on analyzing gene expression on a protein-protein interaction network. AVAILABILITY: https://cran.r-project.org/web/packages/BANFF/index.html CONTACT: [email protected], [email protected] information: Supplementary data are available at Bioinformatics online.