Population Diversity in Ant-inspired Optimization Algorithms

Finding a balance between exploration and exploitation is very important in the case of metaheuristics optimization, especially in the systems leveraging population of individuals expressing (as in Evolutionary Algorithms, etc.) or constructing (as in Ant Colony Optimization) solutions. Premature convergence is a real problem and finding means of its automatic detection and counteracting are of great importance. Measuring diversity in Evolutionary Algorithms working in real-value search space is often computationally complex, but feasible while measuring diversity in combinatorial domain is practically impossible (cf. Closest String Problem). Nevertheless, we propose several practical and feasible diversity measurement techniques dedicated to Ant Colony Optimization algorithms, leveraging the fact that even though analysis of the search space is at least an NP problem, we can focus on the pheromone table, where the direct outcomes of the search are expressed and can be analyzed. Besides proposing the measurement techniques, we apply them to assess the diversity of several variants of ACO, and closely analyze their features for the classic ACO. The discussion of the results is the first step towards applying the proposed measurement techniques in auto-adaptation of the parameters affecting directly the exploitation and exploration features in ACO in the future.

Negative media portrayals of immigrants increase ingroup favoritism and hostile physiological and emotional reactions

Anti-immigration rhetoric in the mass media has intensified over the last two decades, potentially decreasing prosocial behavior and increasing outgroup hostility toward immigrants, and fostering ingroup favoritism toward natives. We aim to understand the effects of negative and positive discourses about immigration on prosociality at different levels of societal ethnic diversity. In two studies (student sample, nationally representative sample), we conduct a survey and a 3X3 between-subject experiment, including money-incentivized behavioral games measuring prosociality. We manipulate media representations of immigrants and the probability of interacting with immigrants (the latter measuring diversity). Results show that negative news affects prosociality as a function of the probability of interacting with immigrants. Negative portrayals increase altruism and trustworthiness in ethnically homogenous settings relative to unknown and ethnically-mixed contexts. These results are stronger for right-wing and high-prejudice respondents. Moreover, negative media portrayals of immigrants increase the testosterone-cortisol ratio, which is a proxy for proneness to social aggression. Negative news also increases outgroup-related perceived health risk, outgroup anxiety and outgroup threat less in ethnically-homogeneous contexts. Overall, negative portrayals of immigrants generate physiological and emotional hostility toward the outgroup, and ingroup favoritism in economic transactions, possibly determining efficiency losses in ethnically-diverse markets, relative to ethnically-homogeneous markets.

Media diversity and the analysis of qualitative variation

Diversity is recognised as a significant criterion for appraising the democratic performance of media systems. This article begins by considering key conceptual debates that help differentiate types and levels of diversity. It then addresses a core methodological challenge in measuring diversity: how do we model statistical variation and difference when many measures of source and content diversity only attain the nominal level of measurement? We identify a range of obscure statistical indices developed in other fields that measure the strength of ‘qualitative variation’. Using original data, we compare the performance of five diversity indices and, on this basis, propose the creation of a more effective diversity average measure. The article concludes by outlining innovative strategies for drawing statistical inferences from these measures, using bootstrapping and permutation testing resampling. All statistical procedures are supported by a unique online resource developed for this article.

Measuring diversity from space: a global view of the free and open source rasterdiv R package under a coding perspective

The variation of species diversity over space and time has been widely recognised as a key challenge in ecology. However, measuring species diversity over large areas might be difficult for logistic reasons related to both time and cost savings for sampling, as well as accessibility of remote ecosystems. In this paper, we present a new package - - to calculate diversity indices based on remotely sensed data, by discussing the theory behind the developed algorithms. Obviously, measures of diversity from space should not be viewed as a replacement of in situ data on biological diversity, but they are rather complementary to existing data and approaches. In practice, they integrate available information of Earth surface properties, including aspects of functional (structural, biophysical and biochemical), taxonomic, phylogenetic and genetic diversity. Making use of the package can result useful in making multiple calculations based on reproducible open source algorithms, robustly rooted in Information Theory.

Assessment of Mixed Plasmodium falciparum sera5 Infection in Endemic Burkitt Lymphoma: A Case-Control Study in Malawi

Background: Endemic Burkitt lymphoma (eBL) is the most common childhood cancer in Africa and is linked to Plasmodium falciparum (Pf) malaria infection, one of the most common and deadly childhood infections in Africa; however, the role of Pf genetic diversity is unclear. A potential role of Pf genetic diversity in eBL has been suggested by a correlation of age-specific patterns of eBL with the complexity of Pf infection in Ghana, Uganda, and Tanzania, as well as a finding of significantly higher Pf genetic diversity, based on a sensitive molecular barcode assay, in eBL cases than matched controls in Malawi. We examined this hypothesis by measuring diversity in Pf-serine repeat antigen-5 (Pfsera5), an antigenic target of blood-stage immunity to malaria, among 200 eBL cases and 140 controls, all Pf polymerase chain reaction (PCR)-positive, in Malawi. Methods: We performed Pfsera5 PCR and sequencing (~3.3 kb over exons II–IV) to determine single or mixed PfSERA5 infection status. The patterns of Pfsera5 PCR positivity, mixed infection, sequence variants, and haplotypes among eBL cases, controls, and combined/pooled were analyzed using frequency tables. The association of mixed Pfsera5 infection with eBL was evaluated using logistic regression, controlling for age, sex, and previously measured Pf genetic diversity. Results: Pfsera5 PCR was positive in 108 eBL cases and 70 controls. Mixed Pf SERA5 infection was detected in 41.7% of eBL cases versus 24.3% of controls; the odds ratio (OR) was 2.18, and the 95% confidence interval (CI) was 1.12–4.26, which remained significant in adjusted results (adjusted odds ratio [aOR] of 2.40, 95% CI of 1.11–5.17). A total of 29 nucleotide variations and 96 haplotypes were identified, but these were unrelated to eBL. Conclusions: Our results increase the evidence supporting the hypothesis that infection with mixed Pf infection is increased with eBL and suggest that measuring Pf genetic diversity may provide new insights into the role of Pf infection in eBL.

Analyzing and Controlling Inter-Head Diversity in Multi-Head Attention

Multi-head attention, a powerful strategy for Transformer, is assumed to utilize information from diverse representation subspaces. However, measuring diversity between heads’ representations or exploiting the diversity has been rarely studied. In this paper, we quantitatively analyze inter-head diversity of multi-head attention by applying recently developed similarity measures between two deep representations: Singular Vector Canonical Correlation Analysis (SVCCA) and Centered Kernel Alignment (CKA). By doing so, we empirically show that multi-head attention does diversify representation subspaces of each head as the number of heads increases. Based on our analysis, we hypothesize that there exists an optimal inter-head diversity with which a model can achieve better performance. To examine our hypothesis, we deeply inspect three techniques to control the inter-head diversity; (1) Hilbert-Schmidt Independence Criterion regularizer among representation subspaces, (2) Orthogonality regularizer, and (3) Drophead as zero-outing each head randomly in every training step. In our experiments on various machine translation and language modeling tasks, we show that controlling inter-head diversity leads to the best performance among baselines.