Autoencoder-Based Semantic Novelty Detection: Towards Dependable AI-Based Systems

Many autonomous systems, such as driverless taxis, perform safety-critical functions. Autonomous systems employ artificial intelligence (AI) techniques, specifically for environmental perception. Engineers cannot completely test or formally verify AI-based autonomous systems. The accuracy of AI-based systems depends on the quality of training data. Thus, novelty detection, that is, identifying data that differ in some respect from the data used for training, becomes a safety measure for system development and operation. In this study, we propose a new architecture for autoencoder-based semantic novelty detection with two innovations: architectural guidelines for a semantic autoencoder topology and a semantic error calculation as novelty criteria. We demonstrate that such a semantic novelty detection outperforms autoencoder-based novelty detection approaches known from the literature by minimizing false negatives.

Design of ALU based on Majority Logic for Low Power Applications

Approximate calculations are a new nanotechnology paradigm for improving efficiency and reducing energy use. Most of the logic extends to many contemporary nanotechnological developments and is used for the design of digital circuits in its basic portion (3 input plurality, MV). This paper suggests implementations of additional compressors and ML multiplicators. An additional bit discovery circuit is used for the proposed compressors. The size of the multiplier is calculated by a control factor for the importance of different extra bits. The designs proposed are tested with hardware (for example, time frame and port complexity) as well as with error calculation. These designs have superior performance in terms of area and delay. The validity of the proposed designs is also shown by case tests of the error resistance implementation.

CALCULATION OF THE BASIC AND ADDITIONAL ERRORS OF THE MOISTURE METER OF GRAIN AND GRAIN PRODUCTS

Calculation of the basic error of the moisture meter is determined by the error in measuring the frequency. The latter, in turn, consists of the error of the timer-counter when counting the frequency (quantization), the error in determining the coefficients a1and a2of the mathematical model of the measuring transducer during calibration, and the error due to the noise of the comparator-amplifier.Keywords:error, error calculation, measurements, result.

Distance-based exclusion method for parentage analysis using microsatellites (SSR) markers

Genotyping mistakes represent a challenge in parental assignment where even small errors can lead to significant amounts of unassigned siblings. Different parental assignment algorithms have been designed to approach this problem. The Exclusion method is the most applied for its reliability and biological meaning. However, the resolving power of this method is the lowest for data containing genotyping errors. We introduce a new distance-based approach which we coin as Distance-Based Exclusion (DBE). The DBE method calculates the distance between the offspring haplotype and haplotype of each of the potential fathers. The father with the lowest distance is then assigned as candidate father according to a distance ratio (α). We have tested the Exclusion and DBE methods using a real dataset of 1230 offsprings subdivided into families of 25 individuals. Each family had six potential fathers and one known mother. Compared with the Exclusion method, the DBE method is able to solve 4.7% more individuals (64.4% Exclusion vs 69.1% DBE) using the most restrictive α tested without errors. DBE method can also be used together with the Exclusion method for error calculation and to further solve unassigned individuals. Using a two-step approach, we were able to assign 98.1% of the offsprings with a total predicted error of 4.7%. Considering the results obtained, we propose the use of the DBE method in combination with the Exclusion method for parental assignment.