The rapid pace of environmental change has prompted pressing concerns about the persistence of wild populations. For plants, because they move from one place to another only passively between generations, their persistence is especially likely to depend on their capacity for ongoing adaptive evolution. There are numerous examples of rapid adaptation in the recent past, but evidence about rates of adaptation in the wild is limited. Previously, to assess the capacity for genetic adaptation of three wild plant populations growing in their source locations, we have estimated their additive genetic variance for fitness in three successive years. Here, we present the actual difference between successive generations in their average absolute fitness. We partition this change into components resulting from genetic change and due to environmental difference, as well as a residual component. In each of six cases of intergenerational change, we have detected evolutionary adaptation as genetic increase in average fitness during the first generation, while also finding generally greater effects of differences in environment between years. Nevertheless, we show that when environmental change reduces a population's average fitness, these adaptive genetic responses often substantively ameliorate its deleterious impact.
Federated fine-tuning of on-device large language models (LLMs) mitigates privacy concerns by preventing raw data sharing. However, the intensive computational and memory demands pose significant challenges for resource-constrained edge devices. To overcome these limitations, split federated learning (SFL) emerges as a promising solution that partitions the model into lightweight client-side and compute-intensive server-side sub-models, thus offloading the primary training workload to a powerful server. Nevertheless, high-dimensional activation exchanges in SFL lead to excessive communication overhead. To overcome this, we propose SplitCom, a communication-efficient SFL framework for LLMs that exploits temporal redundancy in activations across consecutive training epochs. Inspired by video compression, the core innovation of our framework lies in selective activation uploading only when a noticeable deviation from previous epochs occurs. To balance communication efficiency and learning performance, we introduce two adaptive threshold control schemes based on 1) bang-bang control or 2) deep deterministic policy gradient (DDPG)-based reinforcement learning. Moreover, we implement dimensionality reduction techniques to alleviate client-side memory requirements. Furthermore, we extend SplitCom to the U-shape architecture, ensuring the server never accesses clients' labels. Extensive simulations and laboratory experiments demonstrate that SplitCom reduces uplink communication costs by up to 98.6 % in its standard configuration and total communication costs by up to 95.8 % in its U-shape variant without noticeably compromising model performance.
Dark matter (DM) halos form hierarchically in the Universe through a series of merger events. By pairing halo finder and postprocessing codes, merger series in cosmological simulations can be conveniently represented as a graph-like "tree" structure. Previous work has shown that these merger trees are sensitive to cosmological simulation parameters, but as structures comprised of DM halos, the outstanding question of their sensitivity to DM models remains unanswered. In this work, we investigate the feasibility of deep learning methods trained on merger trees to infer warm dark matter (WDM) particle masses from the DREAMS simulation suite. We organize the merger trees from 1024 zoom-in simulations into graphs with nodes representing halos at different epochs and edges denoting hereditary links. We vary the complexity of the node features included in the graphs, ranging from a single node feature up through an array of several galactic properties. We train a graph neural network (GNN) to predict the WDM mass using the graph representation of the merger tree as input. We find that the GNN can predict the mass of the WDM particle, with success depending on the graph complexity and node features. We extend the same methods to supernovae (SNe) and active galactic nuclei feedback parameters, successfully inferring the SNe parameters. With reduced accuracy, the GNN can even infer the WDM mass from merger tree histories without any node features, indicating that the structure of merger trees alone inherits information about the cosmological parameters of the simulations from which they form.
Arsenic exposure is a major global health challenge. In addition to well-documented toxic effects in exposed people and animals, there is evidence that exposure to arsenic may lead to transgenerational effects. Transgenerational effects of low levels of exposure are challenging to study in species with long generation times. The model organism Caenorhabditis elegans offers the ability to quickly carry out transgenerational experiments with very large sample sizes of isogenic animals, reducing variation, and numerous biological replicates, to increase statistical rigor. An important challenge historically associated with this species for such work is uncertainty about internal dosimetry and toxicokinetics. Here, we report a 4-generation experiment in which C. elegans were exposed during larval development to sodium arsenite concentrations in the parental generation at concentrations resulting in no or mild growth inhibition up to significant growth inhibition. These exposures resulted in internal concentrations between 0.4 and 6.7 nM and rapid excretion (t1/2 = 3 hours), despite the lack of arsenic methylation in this species. These exposures had strong and significant effects on the exposed generation later in life, but no transgenerational effects were detected. We discuss possible reasons for this "negative" result. We also report strong similarity of the nematode transcriptomic, metabolomic, and fat accumulation responses in the exposed generation to responses reported in other organisms, including persistent alterations in cysteine and fatty acid metabolism, phase II and III metabolic processes, and increased adiposity. Finally, we discuss ways to take advantage of this species difference in arsenic metabolism for the use of C. elegans in toxicology testing.
Online surveys have become increasingly popular for academic research due to their cost-effectiveness and ability to reach large populations. However, the use of monetary incentives to encourage participation has led to widespread infiltration by automated bots designed to complete surveys fraudulently. This study presents a transparent, multi-filter framework to detect and remove bot responses, using an incentivized online transportation survey on cycling preferences as a case study. The survey received 12,020 responses over 94 days, with clear evidence of bot activity including 3951 submissions (32.9
