Accurate prediction of inflammatory bowel disease (IBD) from gut microbiome data offers a promising path toward non-invasive diagnosis. In this study, we propose IBDPred, an enhanced stacking ensemble framework designed to predict IBD using species-level microbial abundance features. Microbial profiles were derived from publicly available human gut metagenomic datasets and normalized through a standardized preprocessing pipeline. To identify informative microbial taxa, two complementary feature selection strategies were compared: random forest–based importance ranking and a mutual information–variance (MI-VAR) method. The selected features from each strategy were used to train a two-layer stacking model composed of SVM, XGBoost, and Random Forest as base learners, with logistic regression serving as the meta-learner. Furthermore, a weighted fusion mechanism incorporating accuracy and inter-model diversity was introduced to enhance ensemble coordination and robustness. Experimental results show that IBDPred achieved superior performance compared with traditional models. On an independent test cohort, it attained an AUC of 0.8108 and an accuracy of 0.7683, demonstrating strong generalization. These findings suggest that IBDPred provides an effective and interpretable framework for microbiome-based IBD risk prediction.

Reasoning plays a crucial role in advancing Multimodal Large Language Models (MLLMs) toward Artificial General Intelligence. However, existing MLLM benchmarks often fall short in precisely and comprehensively evaluating long-chain reasoning abilities from three key aspects: (1) lack of difficulty and diversity, (2) susceptibility to guessability and memorization, (3) inadequate assessment of intermediate reasoning steps. To fill this gap, we introduce MMReason, a new benchmark designed to precisely and comprehensively evaluate MLLM long-chain reasoning capability with diverse, open-ended, challenging questions. First, we curate challenging questions requiring multi-step reasoning from various fields (i.e., 6 disciplines) and multiple difficulty levels (i.e., from pre-university to university, and from foundational to competition tiers). Second, these questions are reformulated into an open-ended format and filtered using a multi-model voting technique to eliminate shortcut cases related to guessing and memorization, ensuring robust reasoning evaluations. Third, we annotate the questions with detailed step-by-step solutions, and design a reference-based ternary scoring mechanism to reliably assess intermediate reasoning steps. With MMReason, we benchmark popular leading MLLMs and provide an in-depth analysis of their reasoning capabilities. We hope MMReason will serve as a valuable resource for advancing MLLM reasoning research. Code will be available at https://github.com/HJYao00/MMReason.

Metrics that aggregate the state of neighboring nodes are frequently used in wireless sensor networks. In this paper, we present two primitives that exploit simultaneous communications in 802.15.4 radios to enable a polling node to calculate with low power the number (or set) of its neighbors where some state predicate of interest holds. In both primitives, the poller assigns transmission powers and response lengths to its respective neighbors for their simultaneous response to each of its poll requests. The two primitives adopt complementary schemes for power assignment such that the Received Signal Strength Indicator (RSSI) of the respective signal from each neighbor is significantly different from that of all others in one primitive and nearly equivalent to that of the others in the other. The first primitive, LinearPoll, suits sparse networks and consumes energy that is linear in the size of its neighborhood, whereas the second primitive, LogPoll, suits dense networks and consumes constant energy. Compared to the state-of-the-art solutions that use multiple sub-carriers, our primitives are simpler and more compute-efficient while provide estimation with comparable quality. Compared to single-carrier solutions, our primitives achieve comparable quality at less than half the energy cost or richer information at comparable energy cost. They are also compatible with other radio physical layers. Based on our implementation for CC2420 radios on the TelosB platform, we evaluate the primitives in different wireless environments and neighborhood topologies to study their performance, the tradeoff between their estimation accuracy and energy cost, and methods for tuning their critical parameters, and we compare them with baseline counting protocols.

Capacity studies for wireless networks typically focus on scheduling of transmissions but rarely on scheduling of radio sleep-wakeup. We show that in a multi-hop network setting the duty-cycling of radios results in a per-node throughput capacity of Θ(W ψ√log n/√n) as opposed to Θ(W ψ/√ n log n) where ψ is the fraction of time each node radio is active. Likewise, in a single-hop clique network the duty-cycling of radios results in a per-node throughput capacity of Θ (Wψ), as opposed to Θ(Wψ/n). These capacity gains of Θ(log n) and Θ(n) respectively result from spreading interference over time. They emphasizes the importance of efficient co-scheduling of transmissions and sleep-wakeup, which is normally a function of the Medium Access Control (MAC) layer. We also examine how well canonical classes of duty-cycled MAC protocols achieve throughput capacity. In particular, we abstract four schedulers, each of which represents several well-used MAC schedulers, namely sender-centric synchronous (e.g. S-MAC, T-MAC, SCP-MAC), receiver-centric synchronous (e.g. O-MAC), sender-centric asynchronous (e.g. B-MAC, X-MAC, BoX-MAC), and receiver-centric asynchronous (e.g. RI-MAC). We compare the achievable throughput of these schedulers analytically; extensive experiments are then performed to corroborate our MAC analysis. The findings strongly suggest that out of the four classes receiver-centric synchronous scheme yields the smallest gap between existing MACs and the optimal scheduler.

The energy efficiency of the widely used convergecast pattern depends substantially on the choice of medium access control (MAC) and routing protocol. In this paper, we formalize the maximization of convergecast energy efficiency with respect to its MAC and routing as a resource constrained optimization problem. We then analytically show that this maximization problem is linear in the context of two prototypical MACs - a locally synchronized wakeup (as in S-MAC) and a locally staggered wakeup MAC (as in O-MAC) - assuming low, uniform traffic that is delivered reliably and without interference. With this insight, we present a centralized algorithm, MeeCast, that solves the optimization problem utilizing linear programming techniques. We also design a distributed version of MeeCast, for the case where the traffic is ultra-low, and prove that it achieves optimality as well as fast convergence time. Notably, this version is self-stabilizing, so it autonomically handles changes in traffic load, network topology, loss of coordination and state corruption. In comparison with Dozer, a state-of-the-art convergecast protocol, MeeCast achieves better energy efficiency and application lifetime in the context of S-MAC and identical energy efficiency but better application lifetime in the context of O-MAC.