On energy complexity of fully-connected layers.

The massive increase in the size of deep neural networks (DNNs) is accompanied by a significant increase in energy consumption of their hardware implementations which is critical for their widespread deployment in low-power mobile devices. In our previous work, an abstract hardware-independent model of energy complexity for convolutional neural networks (CNNs) has been proposed and experimentally validated. Based on this model, we provide a theoretical analysis of energy complexity related to the computation of a fully-connected layer when its inputs, outputs, and weights are transferred between two kinds of memories (DRAM and Buffer). First, we establish a general lower bound on this energy complexity. Then, we present two dataflows and calculate their energy costs to achieve the corresponding upper bounds. In the case of a partitioned Buffer, we prove by the weak duality theorem from linear programming that the lower and upper bounds coincide up to an additive constant, and therefore establish the optimal energy complexity. Finally, the asymptotically optimal quadratic energy complexity of fully-connected layers is experimentally validated by estimating their energy consumption on the Simba and Eyeriss hardware.

Energy Complexity of Convolutional Neural Networks.

The energy efficiency of hardware implementations of convolutional neural networks (CNNs) is critical to their widespread deployment in low-power mobile devices. Recently, a number of methods have been proposed for providing energy-optimal mappings of CNNs onto diverse hardware accelerators. Their estimated energy consumption is related to specific implementation details and hardware parameters, which does not allow for machine-independent exploration of CNN energy measures. In this letter, we introduce a simplified theoretical energy complexity model for CNNs, based on only a two-level memory hierarchy that captures asymptotically all important sources of energy consumption for different CNN hardware implementations. In this model, we derive a simple energy lower bound and calculate the energy complexity of evaluating a CNN layer for two common data flows, providing corresponding upper bounds. According to statistical tests, the theoretical energy upper and lower bounds we present fit asymptotically very well with the real energy consumption of CNN implementations on the Simba and Eyeriss hardware platforms, estimated by the Timeloop/Accelergy program, which validates the proposed energy complexity model for CNNs.

Rotation-based schedules in elementary schools to prevent COVID-19 spread: a simulation study.

Rotations of schoolchildren were considered as a non-pharmacological intervention in the COVID-19 pandemic. This study investigates the impact of different rotation and testing schedules.We built an agent-based model of interactions among pupils and teachers based on a survey in an elementary school in Prague, Czechia. This model contains 624 schoolchildren and 55 teachers and about 27 thousands social contacts in 10 layers. The layers reflect different types of contacts (classroom, cafeteria, etc.) in the survey. On this multi-graph structure we run a modified SEIR model of covid-19 infection. The parameters of the model are calibrated on data from the outbreak in the Czech Republic in spring 2020. Weekly rotations of in-class and distance learning are an effective preventative measure in schools reducing the spread of covid-19 by 75-81% . Antigen testing twice a week or PCR once a week significantly reduces infections even when using tests with a lower sensitivity. The structure of social contacts between pupils and teachers strongly influences the transmission. While the density of contact graphs for older pupils is 1.5 times higher than for younger pupils, the teachers' network is an order of magnitude denser. Teachers moreover act as bridges between groups of children, responsible for 14-18% of infections in the secondary school compared to 8-11% in the primary school. Weekly rotations with regular testing are a highly effective non-pharmacological intervention for the prevention of covid-19 spread in schools and a way to keep schools open during an epidemic.

On the contact tracing for COVID-19: A simulation study.

BACKGROUND: Contact tracing is one of the most effective non-pharmaceutical interventions in the COVID-19 pandemic. This study uses a multi-agent model to investigate the impact of four types of contact tracing strategies to prevent the spread of COVID-19. METHODS: In order to analyse individual contact tracing in a reasonably realistic setup, we construct an agent-based model of a small municipality with about 60.000 inhabitants (nodes) and about 2.8 million social contacts (edges) in 30 different layers. Those layers reflect demographic, geographic, sociological and other patterns of the TTWA (Travel-to-work-area) Hodonín in Czechia. Various data sources such as census, land register, transport data or data reflecting the shopping behaviour, were employed to meet this purpose. On this multi-graph structure we run a modified SEIR model of the COVID-19 dynamics. The parameters of the model are calibrated on data from the outbreak in the Czech Republic in the period March to June 2020. The simplest type of contact tracing follows just the family, the second tracing version tracks the family and all the work contacts, the third type finds all contacts with the family, work contacts and friends (leisure activities). The last one is a complete (digital) tracing capable of recalling any and all contacts. We evaluate the performance of these contact tracing strategies in four different environments. First, we consider an environment without any contact restrictions (benchmark); second with strict contact restriction (replicating the stringent non-pharmaceutical interventions employed in Czechia in the spring 2020); third environment, where the measures were substantially relaxed, and, finally an environment with weak contact restrictions and superspreader events (replicating the situation in Czechia in the summer 2020). FINDINGS: There are four main findings in our paper. 1. In general, local closures are more effective than any type of tracing. 2. In an environment with strict contact restrictions there are only small differences among the four contact tracing strategies. 3. In an environment with relaxed contact restrictions the effectiveness of the tracing strategies differs substantially. 4. In the presence of superspreader events only complete contact tracing can stop the epidemic. INTERPRETATION: In situations, where many other non-pharmaceutical interventions are in place, the specific extent of contact tracing may not have a large influence on their effectiveness. In a more relaxed setting with few contact restrictions and larger events the effectiveness of contact tracing depends heavily on their extent.

Delays, Masks, the Elderly, and Schools: First Covid-19 Wave in the Czech Republic.

Running across the globe for nearly 2 years, the Covid-19 pandemic keeps demonstrating its strength. Despite a lot of understanding, uncertainty regarding the efficiency of interventions still persists. We developed an age-structured epidemic model parameterized with epidemiological and sociological data for the first Covid-19 wave in the Czech Republic and found that (1) starting the spring 2020 lockdown 4 days earlier might prevent half of the confirmed cases by the end of lockdown period, (2) personal protective measures such as face masks appear more effective than just a realized reduction in social contacts, (3) the strategy of sheltering just the elderly is not at all effective, and (4) leaving schools open is a risky strategy. Despite vaccination programs, evidence-based choice and timing of non-pharmaceutical interventions remains an effective weapon against the Covid-19 pandemic.

Importance of vaccine action and availability and epidemic severity for delaying the second vaccine dose.

Following initial optimism regarding potentially rapid vaccination, delays and shortages in vaccine supplies occurred in many countries during spring 2021. Various strategies to counter this gloomy reality and speed up vaccination have been set forth, of which the most popular has been to delay the second vaccine dose for a longer period than originally recommended by the manufacturers. Controversy has surrounded this strategy, and overly simplistic models have been developed to shed light on this issue. Here we use three different epidemic models, all accounting for then actual COVID-19 epidemic in the Czech Republic, including the real vaccination rollout, to explore when delaying the second vaccine dose by another 3 weeks from 21 to 42 days is advantageous. Using COVID-19-related deaths as a quantity to compare various model scenarios, we find that the way of vaccine action at the beginning of the infection course (preventing infection and symptoms appearance), mild epidemic and sufficient vaccine supply rate call for the original inter-dose period of 21 days regardless of vaccine efficacy. On the contrary, for the vaccine action at the end of infection course (preventing severe symptoms and death), severe epidemic and low vaccine supply rate, the 42-day inter-dose period is preferable, at any plausible vaccine efficacy.

Effective Automatic Method Selection for Nonlinear Regression Modeling.

Metalearning, an important part of artificial intelligence, represents a promising approach for the task of automatic selection of appropriate methods or algorithms. This paper is interested in recommending a suitable estimator for nonlinear regression modeling, particularly in recommending either the standard nonlinear least squares estimator or one of such available alternative estimators, which is highly robust with respect to the presence of outliers in the data. The authors hold the opinion that theoretical considerations will never be able to formulate such recommendations for the nonlinear regression context. Instead, metalearning is explored here as an original approach suitable for this task. In this paper, four different approaches for automatic method selection for nonlinear regression are proposed and computations over a training database of 643 real publicly available datasets are performed. Particularly, while the metalearning results may be harmed by the imbalanced number of groups, an effective approach yields much improved results, performing a novel combination of supervised feature selection by random forest and oversampling by synthetic minority oversampling technique (SMOTE). As a by-product, the computations bring arguments in favor of the very recent nonlinear least weighted squares estimator, which turns out to outperform other (and much more renowned) estimators in a quite large percentage of datasets.

Delays, masks, the elderly, and schools: first COVID-19 wave in the Czech Republic.

Running across the globe for more than a year, the COVID-19 pandemic keeps demonstrating its strength. Despite a lot of understanding, uncertainty regarding the efficiency of interventions still persists. We developed an age-structured epidemic model parameterized with sociological data for the Czech Republic and found that (1) delaying the spring 2020 lockdown by four days produced twice as many confirmed cases by the end of the lockdown period, (2) personal protective measures such as face masks appear more effective than just a reduction of social contacts, (3) only sheltering the elderly is by no means effective, and (4) leaving schools open is a risky strategy. Despite the onset of vaccination, an evidence-based choice and timing of non-pharmaceutical interventions still remains the most important weapon against the COVID-19 pandemic.

Vulnerability of classifiers to evolutionary generated adversarial examples.

This paper deals with the vulnerability of machine learning models to adversarial examples and its implication for robustness and generalization properties. We propose an evolutionary algorithm that can generate adversarial examples for any machine learning model in the black-box attack scenario. This way, we can find adversarial examples without access to model's parameters, only by querying the model at hand. We have tested a range of machine learning models including deep and shallow neural networks. Our experiments have shown that the vulnerability to adversarial examples is not only the problem of deep networks, but it spreads through various machine learning architectures. Rather, it depends on the type of computational units. Local units, such as Gaussian kernels, are less vulnerable to adversarial examples.

Comparison of behavior-based and planning techniques on the small robot maze exploration problem.

A comparison of behavior-based and planning approaches of robot control is presented in this paper. We focus on miniature mobile robotic agents with limited sensory abilities. Two reactive control mechanisms for an agent are considered-a radial basis function neural network trained by evolutionary algorithm and a traditional reinforcement learning algorithm over a finite agent state space. The control architecture based on localization and planning is compared to the former method.