Deep neural networks (DNNs) are vulnerable to the attacks of adversarial examples, which bring serious security risks to the learning systems. In this paper, we propose a new defense method to ...improve the adversarial robustness of DNNs based on stochastic neural networks (SNNs), termed as Margin-SNN. The proposed Margin-SNN mainly includes two modules, i.e., feature uncertainty learning module and label embedding module. The first module introduces uncertainty to the latent feature space by giving each sample a distributional representation rather than a fixed point representation, and leverages the advantages of variational information bottleneck method in achieving good intra-class compactness in latent space. The second module develops a label embedding mechanism to take advantage of the semantic information underlying the labels, which maps the labels into the same latent space with the features, in order to capture the similarity between sample and its class centroid, where a penalty term is equipped to elegantly enlarge the margin between different classes for better inter-class separability. Since no adversarial information is introduced, the proposed model can be learned in standard training to improve adversarial robustness, which is much more efficient than adversarial training. Extensive experiments on data sets MNIST, FASHION MNIST, CIFAR10, CIFAR100 and SVHN demonstrate superior defensive ability of the proposed method. Our code is available at https://github.com/humeng24/Margin-SNN.
In human populations, changes in genetic variation are driven not only by genetic processes, but can also arise from cultural or social changes. An abrupt population bottleneck specific to human ...males has been inferred across several Old World (Africa, Europe, Asia) populations 5000-7000 BP. Here, bringing together anthropological theory, recent population genomic studies and mathematical models, we propose a sociocultural hypothesis, involving the formation of patrilineal kin groups and intergroup competition among these groups. Our analysis shows that this sociocultural hypothesis can explain the inference of a population bottleneck. We also show that our hypothesis is consistent with current findings from the archaeogenetics of Old World Eurasia, and is important for conceptions of cultural and social evolution in prehistory.
The relationship between population size, inbreeding, loss of genetic variation and evolutionary potential of fitness traits is still unresolved, and large-scale empirical studies testing theoretical ...expectations are surprisingly scarce. Here we present a highly replicated experimental evolution setup with 120 lines of Drosophila melanogaster having experienced inbreeding caused by low population size for a variable number of generations. Genetic variation in inbred lines and in outbred control lines was assessed by genotyping-by-sequencing (GBS) of pooled samples consisting of 15 males per line. All lines were reared on a novel stressful medium for 10 generations during which body mass, productivity, and extinctions were scored in each generation. In addition, we investigated egg-to-adult viability in the benign and the stressful environments before and after rearing at the stressful conditions for 10 generations. We found strong positive correlations between levels of genetic variation and evolutionary response in all investigated traits, and showed that genomic variation was more informative in predicting evolutionary responses than population history reflected by expected inbreeding levels. We also found that lines with lower genetic diversity were at greater risk of extinction. For viability, the results suggested a trade-off in the costs of adapting to the stressful environments when tested in a benign environment. This work presents convincing support for long-standing evolutionary theory, and it provides novel insights into the association between genetic variation and evolutionary capacity in a gradient of diversity rather than dichotomous inbred/outbred groups.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Introduced species often colonize regions that have vastly different ecological and environmental conditions than those found in their native range. As such, species introductions can provide a ...deeper understanding into the process of adaptive evolution. In the 1880s, steelhead trout (Oncorhynchus mykiss) from California were introduced into Lake Michigan (Laurentian Great Lakes, North America) where they established naturally reproducing populations. In their native range, steelhead hatch in rivers, migrate to the ocean and return to freshwater to spawn. Steelhead in Lake Michigan continue to swim up rivers to spawn, but now treat the freshwater environment of the Great Lakes as a surrogate ocean. To examine the effects of this introduction, we sequenced the genomes of 264 fish. By comparing steelhead from Lake Michigan to steelhead from their ancestral range, we determined that the introduction led to consistent reductions in genetic diversity across all 29 chromosomes. Despite this reduction in genetic diversity, three chromosomal regions were associated with rapid genetic adaptation to the novel environment. The first region contained functional changes to ceramide kinase, which likely altered metabolic and wound‐healing rates in Lake Michigan steelhead. The second and third regions encoded carbonic anhydrases and a solute carrier protein, both of which are critical for osmoregulation, and demonstrate how steelhead physiologically adapted to freshwater. Furthermore, the contemporary release of diverse hatchery strains into the lake increased genetic diversity but reduced the signature of genetic adaptation. This study illustrates that species can rapidly adapt to novel environments despite genome‐wide reductions in genetic diversity.
see also the Perspective by Seeb et al.
Cloud servers are highly prone to resource bottleneck failures. In this work, we propose an ensemble learning model to build LSTM-based multiclass classifier for resource bottleneck identification. ...The workload at cloud servers is highly dynamic in nature. To support continuous online learning of resource bottleneck identification models, we propose relevance feedback based online learning of proposed ensemble model. Here we propose to analyse, catastrophe forgetting and incremental architectural evolution as two fundamental challenges associated with online adaptation of LSTM-based multiclass classifier models. To avoid catastrophic forgetting, we propose a combination of distillation loss and the standard crossentropy loss. For architectural evolution, we propose and analyse three different alternatives to update the architecture of the bottleneck identification model on the fly.
We evaluate the proposed approaches on a real world dataset collected in an industrial case study and on a dataset collected in a virtual environment setup using Docker containers. The experimental results show that the proposed approaches outperform existing state-of-the-art methods for bottleneck identification.
Ciprofloxacin is an important antibacterial drug targeting Type II topoisomerases, highly active against Gram-negatives including Escherichia coli. The evolution of resistance to ciprofloxacin in E. ...coli always requires multiple genetic changes, usually including mutations affecting two different drug target genes, gyrA and parC. Resistant mutants selected in vitro or in vivo can have many different mutations in target genes and efflux regulator genes that contribute to resistance. Among resistant clinical isolates the genotype, gyrA S83L D87N, parC S80I is significantly overrepresented suggesting that it has a selective advantage. However, the evolutionary or functional significance of this high frequency resistance genotype is not fully understood. By combining experimental data and mathematical modeling, we addressed the reasons for the predominance of this specific genotype. The experimental data were used to model trajectories of mutational resistance evolution under different conditions of drug exposure and population bottlenecks. We identified the order in which specific mutations are selected in the clinical genotype, showed that the high frequency genotype could be selected over a range of drug selective pressures, and was strongly influenced by the relative fitness of alternative mutations and factors affecting mutation supply. Our data map for the first time the fitness landscape that constrains the evolutionary trajectories taken during the development of clinical resistance to ciprofloxacin and explain the predominance of the most frequently selected genotype. This study provides strong support for the use of in vitro competition assays as a tool to trace evolutionary trajectories, not only in the antibiotic resistance field.
•A market-driven, permit-based scheme for managing traffic on single-bottleneck roadways.•An integrated framework imbedding the permit-based traffic management scheme.•A mathematical programming ...model and a heuristic algorithm for endowment of the permits.•An iterative auction mechanism for optimizing the revenue outcome of the permit-based scheme.•The strategy-proof scheme guaranteeing minimum equilibrium market prices.
We study the problem of managing traffic on single-bottleneck roadways using a mobility permit (MP)-based method—a scheme that becomes increasingly relevant to shape the new mobility culture. To identify the main concerns of the involved stakeholders, we present an integrated framework that embeds our MP-based traffic management scheme, in which the mobility permit users are allowed to choose the options best matching their travel needs. Furthermore, to find the most effective market prices, the proposed permit allocation approach is integrated into an iterative auction process to achieve the best equilibrium state in terms of permit prices that is suitable for users, mitigating potential efficiency loss. Our computational results indicate the effectiveness of the proposed scheme as an alternative solution for permit-based mobility management on single-bottleneck roadways.
Abstract
Background
Persian walnut,
Juglans regia
, occurs naturally from Greece to western China, while its closest relative, the iron walnut,
Juglans sigillata
, is endemic in southwest China; both ...species are cultivated for their nuts and wood. Here, we infer their demographic histories and the time and direction of possible hybridization and introgression between them.
Results
We use whole-genome resequencing data, different population-genetic approaches (PSMC and GONE), and isolation-with-migration models (IMa3) on individuals from Europe, Iran, Kazakhstan, Pakistan, and China. IMa3 analyses indicate that the two species diverged from each other by 0.85 million years ago, with unidirectional gene flow from eastern
J. regia
and its ancestor into
J. sigillata
, including the shell-thickness gene. Within
J. regia
, a western group, located from Europe to Iran, and an eastern group with individuals from northern China, experienced dramatically declining population sizes about 80 generations ago (roughly 2400 to 4000 years), followed by an expansion at about 40 generations, while
J. sigillata
had a constant population size from about 100 to 20 generations ago, followed by a rapid decline.
Conclusions
Both
J. regia
and
J. sigillata
appear to have suffered sudden population declines during their domestication, suggesting that the bottleneck scenario of plant domestication may well apply in at least some perennial crop species. Introgression from introduced
J. regia
appears to have played a role in the domestication of
J. sigillata.
Fine-grained visual categorization (FGVC) aims to discriminate similar subcategories, whose main challenge is the large intraclass diversities and subtle inter-class differences. Existing FGVC ...methods usually select discriminant regions found by a trained model, which is prone to neglect other potential discriminant information. On the other hand, the massive interactions between the sequence of image patches in ViT make the resulting class token contain lots of redundant information, which may also impact FGVC performance. In this paper, we present a novel approach for FGVC, which can simultaneously make use of partial yet sufficient discriminative information in environmental cues and also compress the redundant information in class-token with respect to the target. Specifically, our model calculates the ratio of high-weight regions in a batch, adaptively adjusts the masking threshold, and achieves moderate extraction of background information in the input space. Moreover, we also use the Information Bottleneck (IB) approach to guide our network to learn a minimum sufficient representations in the feature space. Experimental results on three widely-used benchmark datasets verify that our approach can achieve better performance than other state-of-the-art approaches and baseline models. The code of our model is available at: https://github.com/SYe-hub/R-2-Trans.
•Batch-based dynamic masking for input redundancy.•Information bottleneck for feature redundancy.•Performance witnessed an increase across three datasets.
•Investigate the congestion management issues of a bottleneck-constrained highway system with manned vehicles (MVs) and autonomous vehicles (AVs)•Incorporate the effects of in-vehicle activity ...utilities and activity type choices of AV commuters in the model.•Differentiated and non-differentiated (or anonymous) step tolling schemes for the mixed MV and AV traffic system are analyzed and compared.•Queuing segregation phenomenon occurs between MV and AV commuters.•Optimal AV proportion solution that minimizes total social cost of the mixed MV and AV system is derived.•Differentiated step tolling scheme outperforms non-differentiated step tolling scheme in terms of total social cost and queuing removal rate.
This paper addresses the congestion management issues of a bottleneck-constrained highway system with manned vehicles (MVs) and autonomous vehicles (AVs). A novel model that incorporates the in-vehicle activity utilities and activity type choices of AV commuters is presented for analyzing the equilibrium of the bi-modal bottleneck system with MVs and AVs. Time-varying and step tolling schemes accounting for the effects of in-vehicle activity utilities are analytically explored. In particular, a differentiated and a non-differentiated (or anonymous) step tolling scheme for the mixed MV and AV traffic system are analyzed and compared. The differentiated step tolling scheme means different toll levels across MV and AV commuters, whereas the non-differentiated step tolling scheme means an identical toll level for both the MV and AV commuters. The results show that a queuing segregation phenomenon occurs between the MV and AV commuters. The optimal AV proportion solution that minimizes the total social cost of the mixed MV and AV system exists. For a pure MV or AV system, the optimal step toll scheme can eliminate exactly half of the total queuing delay with the no-toll equilibrium for each of the differentiated and non-differentiated step tolling schemes. However, for a mixed MV and AV system, the queuing removal rate of the optimal differentiated step tolling scheme would exceed a half, whereas that of the optimal non-differentiated step tolling scheme may be lower than a half. The differentiated step tolling scheme outperforms the non-differentiated step tolling scheme in terms of total social cost and queuing removal rate.