Differentially private multivariate time series forecasting of aggregated human mobility with deep learning: Input or gradient perturbation?

This paper investigates the problem of forecasting multivariate aggregated human mobility while preserving the privacy of the individuals concerned. Differential privacy, a state-of-the-art formal notion, has been used as the privacy guarantee in two different and independent steps when training deep learning models. On one hand, we considered gradient perturbation, which uses the differentially private stochastic gradient descent algorithm to guarantee the privacy of each time series sample in the learning stage. On the other hand, we considered input perturbation, which adds differential privacy guarantees in each sample of the series before applying any learning. We compared four state-of-the-art recurrent neural networks: Long Short-Term Memory, Gated Recurrent Unit, and their Bidirectional architectures, i.e., Bidirectional-LSTM and Bidirectional-GRU. Extensive experiments were conducted with a real-world multivariate mobility dataset, which we published openly along with this paper. As shown in the results, differentially private deep learning models trained under gradient or input perturbation achieve nearly the same performance as non-private deep learning models, with loss in performance varying between 0.57 and 2.8 % . The contribution of this paper is significant for those involved in urban planning and decision-making, providing a solution to the human mobility multivariate forecast problem through differentially private deep learning models.

Ancestral Reconstruction and Investigations of Genomic Recombination on some Pentapetalae Chloroplasts.

In this article, we propose a semi-automated method to rebuild genome ancestors of chloroplasts by taking into account gene duplication. Two methods have been used in order to achieve this work: a naked eye investigation using homemade scripts, whose results are considered as a basis of knowledge, and a dynamic programming based approach similar to Needleman-Wunsch. The latter fundamentally uses the Gestalt pattern matching method of sequence matcher to evaluate the occurrences probability of each gene in the last common ancestor of two given genomes. The two approaches have been applied on chloroplastic genomes from Apiales, Asterales, and Fabids orders, the latter belonging to Pentapetalae group. We found that Apiales species do not undergo indels, while they occur in the Asterales and Fabids orders. A series of experiments was then carried out to extensively verify our findings by comparing the obtained ancestral reconstruction results with the latest released approach called MLGO (Maximum Likelihood for Gene-Order analysis).

On the reconstruction of the ancestral bacterial genomes in genus Mycobacterium and Brucella.

BACKGROUND: To reconstruct the evolution history of DNA sequences, novel models of increasing complexity regarding the number of free parameters taken into account in the sequence evolution, as well as faster and more accurate algorithms, and statistical and computational methods, are needed. More particularly, as the principal forces that have led to major structural changes are genome rearrangements (such as translocations, fusions, and so on), understanding their underlying mechanisms, among other things via the ancestral genome reconstruction, are essential. In this problem, since finding the ancestral genomes that minimize the number of rearrangements in a phylogenetic tree is known to be NP-hard for three or more genomes, heuristics are commonly chosen to obtain approximations of the exact solution. The aim of this work is to show that another path is possible. RESULTS: Various algorithms and software already deal with the difficult nature of the problem of reconstruction of the ancestral genome, but they do not function with precision, in particular when indels or single nucleotide polymorphisms fall into repeated sequences. In this article, and despite the theoretical NP-hardness of the ancestral reconstruction problem, we show that an exact solution can be found in practice in various cases, encompassing organelles and some bacteria. A practical example proves that an accurate reconstruction, which also allows to highlight homoplasic events, can be obtained. This is illustrated by the reconstruction of ancestral genomes of two bacterial pathogens, belonging in Mycobacterium and Brucella genera. CONCLUSIONS: By putting together automatically reconstructed ancestral regions with handmade ones for problematic cases, we show that an accurate reconstruction of ancestors of the Brucella genus and of the Mycobacterium tuberculosis complex is possible. By doing so, we are able to investigate the evolutionary history of each pathogen by computing their common ancestors. They can be investigated extensively, by studying the gene content evolution over time, the resistance acquisition, and the impacts of mobile elements on genome plasticity.

Comparison of metaheuristics to measure gene effects on phylogenetic supports and topologies.

BACKGROUND: A huge and continuous increase in the number of completely sequenced chloroplast genomes, available for evolutionary and functional studies in plants, has been observed during the past years. Consequently, it appears possible to build large-scale phylogenetic trees of plant species. However, building such a tree that is well-supported can be a difficult task, even when a subset of close plant species is considered. Usually, the difficulty raises from a few core genes disturbing the phylogenetic information, due for example from problems of homoplasy. Fortunately, a reliable phylogenetic tree can be obtained once these problematic genes are identified and removed from the analysis.Therefore, in this paper we address the problem of finding the largest subset of core genomes which allows to build the best supported tree. RESULTS: As an exhaustive study of all core genes combination is untractable in practice, since the combinatorics of the situation made it computationally infeasible, we investigate three well-known metaheuristics to solve this optimization problem. More precisely, we design and compare distributed approaches using genetic algorithm, particle swarm optimization, and simulated annealing. The latter approach is a new contribution and therefore is described in details, whereas the two former ones have been already studied in previous works. They have been designed de novo in a new platform, and new experiments have been achieved on a larger set of chloroplasts, to compare together these three metaheuristics. CONCLUSIONS: The ways genes affect both tree topology and supports are assessed using statistical tools like Lasso or dummy logistic regression, in an hybrid approach of the genetic algorithm. By doing so, we are able to provide the most supported trees based on the largest subsets of core genes.

Neural networks and chaos: construction, evaluation of chaotic networks, and prediction of chaos with multilayer feedforward networks.

Many research works deal with chaotic neural networks for various fields of application. Unfortunately, up to now, these networks are usually claimed to be chaotic without any mathematical proof. The purpose of this paper is to establish, based on a rigorous theoretical framework, an equivalence between chaotic iterations according to Devaney and a particular class of neural networks. On the one hand, we show how to build such a network, on the other hand, we provide a method to check if a neural network is a chaotic one. Finally, the ability of classical feedforward multilayer perceptrons to learn sets of data obtained from a dynamical system is regarded. Various boolean functions are iterated on finite states. Iterations of some of them are proven to be chaotic as it is defined by Devaney. In that context, important differences occur in the training process, establishing with various neural networks that chaotic behaviors are far more difficult to learn.