Wyniki wyszukiwania - BazTech

1

An impact of tensor-based data compression methods on deep neural network accuracy

Grabek Jakub, Cyganek Bogusław

Annals of Computer Science and Information Systems

|

2021

|

Vol. 26

3--11

EN

The emergence of the deep neural architectures greatly influenced the contemporary big data revolution. How-ever, requirements on large datasets even increased a necessity for efficient data storage. The storage problem is present at all stages, from the dataset creation up to the training and prediction stages. However, compression algorithms can significantly deteriorate the quality of data and in effect the classification models. In this article, an in-depth analysis of the influence of the tensor-based lossy data compression on the performance of the various deep neural architectures is presented. We show that the Tucker and the Tensor Train decomposition methods, with properly selected parameters, allow for very high compression ratios, while conveying enough information in the decompressed data to achieve only a negligible or very small drop in the accuracy. The measurements were performed on the popular deep neural architectures: AlexNet, ResNet, VGG, and MNASNet. We show that further augmentation of the tensor decompositions with the ZFP floating-point compression algorithm allows for finding optimal parameters and even higher compressions ratios at the same recognition accuracy. Our experiments show data compressions of 94%-97% that result in less than 1% accuracy drop.

2

Effect of compression ratio, peak temperature of the cycle and specific heat of working fluid on the performance of diesel cycle

Khan M. N.

International Journal of Applied Mechanics and Engineering

|

2011

|

Vol. 16, no 1

271-276

EN

This paper is an analytical analysis of performance of irreversible diesel cycle by varying its compression ratio, specific heat of working fluid and peak temperature of the cycle simultaneously. The relation between the power output, compression ratio and specific heat of the working fluid, between the thermal efficiency, compression ratio and the specific heat of the working fluid and between the thermal efficiency and power output of the cycle are analyzed under the particular range of cycle peak temperatures of the cycle. The results show that there are significant effects of the specific heat of the working fluid, cycle peak temperature on the performance of diesel cycle. The conclusion can be useful for design of diesel engines as well as for predicating the performance of irreversible Diesel cycle under practical conditions.

3

Poprawa zależności poziomu listków bocznych od współczynnika kompresji dla sygnałów złożonych z małą bazą

Łuszczyk M.

Elektronika : konstrukcje, technologie, zastosowania

|

2010

|

Vol. 51, nr 11

75-78

PL

Najczęściej wykorzystywanym sygnałem w radiolokacji jest sygnał złożony z wewnątrzimpulsową modulacją częstotliwości (LFM). Podstawowym parametrem tego typu sygnałów jest baza sygnału, którą wyznacza się jako iloczyn czasu trwania impulsu oraz dewiacji sygnału zmodulowanego częstotliwościowo. Oczekiwany poziom czasowych listków bocznych sygnału po kompresji dla sygnałów LFM uzyskuje się dla współczynników kompresji powyżej 100. W przypadku stosowania sygnałów LFM 0 mniejszej bazie (np. krótkim czasie trwania) następuje znaczna redukcja poziomu czasowych listków bocznych. Istotne obniżenia listków bocznych można osiągnąć poprzez zastosowanie nieliniowej modulacji częstotliwości (NLFM). W artykule przedstawiono algorytm syntezy sygnałów NLFM oraz zaprezentowano wyniki badań symulacyjnych. Proponowana metoda jest nowym podejściem do problemu zapewnienia wymaganej rozdzielczości odległościowej w warunkach minimalizacji strefy martwej radaru.

EN

The mostly applicable radar signal is pulse signal with linear frequency modulation (LFM). The main feature such signal is time-bandwidth product which is calculated as product of pulse duration T and signal bandwidth B. Theoretical peak to sidelobe level is achievable with time-bandwidth product value grater than 100. For time-bandwidth product smaller then 100 (Le. short pulse duration or narrow deviation of FM modulation) peak to-sidelobe level (PSL) and pulse compression coefficient are significantly reduced. Non-linear frequency modulated (NLFM) radar signal for small time-bandwidth product features better PSL and pulse compression coefficient. The NLFM signal synthesis algorithm and simulation results are presented in the paper. LFM and NLFM signals with small time-bandwidth product are compared and results are discussed in aspect of radar resolution improvement.