148,374 research outputs found
A Scalable Architecture for Harvest-Based Digital Libraries - The ODU/Southampton Experiments
No full textThis paper discusses the requirements of current and emerging applications based on the Open Archives Initiative (OAI) and emphasizes the need for a common infrastructure to support them. Inspired by HTTP proxy, cache, gateway and web service concepts, a design for a scalable and reliable infrastructure that aims at satisfying these requirements is presented. Moreover it is shown how various applications can exploit the services included in the proposed infrastructure. The paper concludes by discussing the current status of several prototype implementations
Paikallisten binäärikuvioiden inspiroima tehokas syvä konvoluutioneuroverkko visuaalisten piirteiden oppimiseen
No full textAbstractIn the past decades, deep neural networks (DNNs) have revolutionized the computer vision community with their significant success in a wide range of computer vision tasks. Recent work has focused intensely on accuracy, which has resulted in a large number of huge and complex models designed in the community. However, with the ubiquitous use of edge devices like mobile phones, robots, and embedded systems, efficiency is gradually becoming more and more important for modern computer vision models. In computer vision, the quality of feature representation learning directly determines the quality of the whole machine learning model. The core challenge is to develop feature representation learning algorithms both effectively and efficiently.In this thesis, we put our efforts into the following matters to meet the challenge. On the one hand, we take the merit of traditional local binary pattern (LBP) descriptors of being computationally simple and efficient, and propose improvement in the learnability of LBP to extract more discriminative features. On the other hand, taking advantage of DNNs of high representational capacity, we target building compact DNN modules with less computational cost and model size. These two aspects are either separately developed or combined, and both are considered in this thesis.We start by extending traditional LBP to learnable descriptors, allowing the new descriptors to be learned from the data rather than handcrafted. Based on that, our model obtains a better trade-off than earlier LBP variants including distinctiveness, computational cost, and robustness. Next, we propose two novel types of convolutions that combine LBP and the convolution operation. The new convolutions enjoy the following benefits: capturing higher-order local differential information, being computationally efficient, and being able to be integrated well into existing DNNs. Then, we propose an efficient convolutional neural network (CNN) module that benefits from group convolution and dynamic execution. It shares the efficiency of the standard group convolution without losing representational ability. Finally, we develop a novel binary DNN module for robust point cloud analysis. The proposed point cloud models achieve both running efficiencies through network binarization and rotation invariance at the same time.Original papersOriginal papers are not included in the electronic version of the dissertation.Su, Z., Pietikäinen, M., & Liu, L. (2019). BIRD: Learning binary and illumination robust descriptor for face recognition. In 30th British Machine Visison Conference : BMVC 2019, 1–12.Self-archived versionSu, Z., Fang, L., Kang, W., Hu, D., Pietikäinen, M., & Liu, L. (2020). Dynamic group convolution for accelerating convolutional neural networks. In Computer Vision – ECCV 2020, 138–155. https://doi.org/10.1007/978-3-030-58539-6_9Self-archived versionSu, Z., Liu, W., Yu, Z., Hu, D., Liao, Q., Tian, Q., Pietikainen, M., & Liu, L. (2021). Pixel difference networks for efficient edge detection. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 5097–5107. https://doi.org/10.1109/ICCV48922.2021.00507Self-archived versionSu, Z., Welling, M., Pietikäinen, M., & Liu, L. (2022). SVNet: Where SO(3) equivariance meets binarization on point cloud representation. In 2022 International Conference on 3D Vision (3DV), 547–556. https://doi.org/10.1109/3DV57658.2022.00084Self-archived versionSu, Z., Zhang, J., Wang, L., Zhang, H., Liu, Z., Pietikäinen, M., & Liu, L. (2023). Lightweight pixel difference networks for efficient visual representation learning. IEEE Transactions on Pattern Analysis and Machine Intelligence. Advance online publication. https://doi.org/10.1109/TPAMI.2023.3300513Su, Z., Müller, M., Wofk, D., Pietikäinen, M., & Liu, L. (2023). Spatial and temporal difference network for real-time salient object detection. Manuscript submitted for publication.TiivistelmäViime vuosikymmeninä syvät neuroverkot ovat mullistaneet konenäköä suurella menestyksellä useissa eri tehtävissä. Viimeaikaisin tutkimus on keskittynyt menetelmien tarkkuuteen, mikä on johtanut suuren määrään valtavan kokoisia ja kompleksisia malleja. Kaikella läsnä olevalla tekniikalla, kuten puhelimilla, roboteilla ja sulautetuilla järjestelmillä konenäkö on kasvavassa määrin tärkeämpää ja täten mallien tehokkuus on myös tärkeämpää. Konenäössä piirteiden oppiminen määrittää suoraan koko konenäkö mallin laadun. Keskeinen haaste on kehittää piirteiden oppimisalgoritmeja tehokkaasti.Tässä väitöskirjassa esitetään seuraavia ratkaisuja mallien tehokkuuden ongelmaan. Ensiksi parannetaan laskennallisesti tehokasta ja yksinkertaista paikallista binäärikuva -menetelmää lisäämällä sen piirteiden määrää. Toiseksi hyödynnetään suurta neuroverkkojen piirteiden kapasiteettia kehittämällä laskennallisesti tehokkaampia ja pienempiä moduuleja. Kumpiakin tekniikkoja käytetään erikseen ja yhdessä tässä väitöskirjassa.Perinteisestä paikallisesta binäärikuvio -menetelmästä tehdään oppiva, jolloin uusia piirteitä voidaan oppia datasta, sen sijaan että ne määriteltäisiin algoritmillisesti. Uusi kehitetty oppiva versio on laskennallisesti tehokkaampi, robustimpi ja erottelevaisempi. Seuraavaksi esitellään tekniikka, joka yhdistää paikallisen binäärikuvion ja konvoluution. Kehitetty konvoluutio pystyy irrottamaan korkeamman asteen paikallista informaatiota, se on laskennallisesti tehokas ja se voidaan integroida olemassa oleviin neuroverkkoihin vaivattomasti. Sen jälkeen esitellään konvoluutioneuroverkon moduuli, joka käyttää hyväkseen ryhmäkonvoluutiota ja dynaamista suoritusta. Moduuli pitää normaalin konvoluution piirteidenirrotus kyvyn ollen kuitenkin yhtä tehokas ryhmäkonvoluution kanssa laskennallisesti. Lopuksi esitetään binäärineuroverkko moduuli robustiin pistepilven analyysiin. Kehitetty moduuli parantaa laskennallista tehokkuutta hyödyntämällä sekä binarisointia että pyörimis-invarianttia konvoluutiota.OsajulkaisutOsajulkaisut eivät sisälly väitöskirjan elektroniseen versioon.Su, Z., Pietikäinen, M., & Liu, L. (2019). BIRD: Learning binary and illumination robust descriptor for face recognition. In 30th British Machine Visison Conference : BMVC 2019, 1–12.Rinnakkaistallennettu versioSu, Z., Fang, L., Kang, W., Hu, D., Pietikäinen, M., & Liu, L. (2020). Dynamic group convolution for accelerating convolutional neural networks. In Computer Vision – ECCV 2020, 138–155. https://doi.org/10.1007/978-3-030-58539-6_9Rinnakkaistallennettu versioSu, Z., Liu, W., Yu, Z., Hu, D., Liao, Q., Tian, Q., Pietikainen, M., & Liu, L. (2021). Pixel difference networks for efficient edge detection. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 5097–5107. https://doi.org/10.1109/ICCV48922.2021.00507Rinnakkaistallennettu versioSu, Z., Welling, M., Pietikäinen, M., & Liu, L. (2022). SVNet: Where SO(3) equivariance meets binarization on point cloud representation. In 2022 International Conference on 3D Vision (3DV), 547–556. https://doi.org/10.1109/3DV57658.2022.00084Rinnakkaistallennettu versioSu, Z., Zhang, J., Wang, L., Zhang, H., Liu, Z., Pietikäinen, M., & Liu, L. (2023). Lightweight pixel difference networks for efficient visual representation learning. IEEE Transactions on Pattern Analysis and Machine Intelligence. Advance online publication. https://doi.org/10.1109/TPAMI.2023.3300513Su, Z., Müller, M., Wofk, D., Pietikäinen, M., & Liu, L. (2023). Spatial and temporal difference network for real-time salient object detection. Manuscript submitted for publication.Academic dissertation to be presented with the assent of the Doctoral Programme Committee of Information Technology and Electrical Engineering of the University of Oulu for public defence in the OP auditorium (L10), Linnanmaa, on 16 October 2023, at 12 noonAbstract
In the past decades, deep neural networks (DNNs) have revolutionized the computer vision community with their significant success in a wide range of computer vision tasks. Recent work has focused intensely on accuracy, which has resulted in a large number of huge and complex models designed in the community. However, with the ubiquitous use of edge devices like mobile phones, robots, and embedded systems, efficiency is gradually becoming more and more important for modern computer vision models. In computer vision, the quality of feature representation learning directly determines the quality of the whole machine learning model. The core challenge is to develop feature representation learning algorithms both effectively and efficiently.
In this thesis, we put our efforts into the following matters to meet the challenge. On the one hand, we take the merit of traditional local binary pattern (LBP) descriptors of being computationally simple and efficient, and propose improvement in the learnability of LBP to extract more discriminative features. On the other hand, taking advantage of DNNs of high representational capacity, we target building compact DNN modules with less computational cost and model size. These two aspects are either separately developed or combined, and both are considered in this thesis.
We start by extending traditional LBP to learnable descriptors, allowing the new descriptors to be learned from the data rather than handcrafted. Based on that, our model obtains a better trade-off than earlier LBP variants including distinctiveness, computational cost, and robustness. Next, we propose two novel types of convolutions that combine LBP and the convolution operation. The new convolutions enjoy the following benefits: capturing higher-order local differential information, being computationally efficient, and being able to be integrated well into existing DNNs. Then, we propose an efficient convolutional neural network (CNN) module that benefits from group convolution and dynamic execution. It shares the efficiency of the standard group convolution without losing representational ability. Finally, we develop a novel binary DNN module for robust point cloud analysis. The proposed point cloud models achieve both running efficiencies through network binarization and rotation invariance at the same time.Tiivistelmä
Viime vuosikymmeninä syvät neuroverkot ovat mullistaneet konenäköä suurella menestyksellä useissa eri tehtävissä. Viimeaikaisin tutkimus on keskittynyt menetelmien tarkkuuteen, mikä on johtanut suuren määrään valtavan kokoisia ja kompleksisia malleja. Kaikella läsnä olevalla tekniikalla, kuten puhelimilla, roboteilla ja sulautetuilla järjestelmillä konenäkö on kasvavassa määrin tärkeämpää ja täten mallien tehokkuus on myös tärkeämpää. Konenäössä piirteiden oppiminen määrittää suoraan koko konenäkö mallin laadun. Keskeinen haaste on kehittää piirteiden oppimisalgoritmeja tehokkaasti.
Tässä väitöskirjassa esitetään seuraavia ratkaisuja mallien tehokkuuden ongelmaan. Ensiksi parannetaan laskennallisesti tehokasta ja yksinkertaista paikallista binäärikuva -menetelmää lisäämällä sen piirteiden määrää. Toiseksi hyödynnetään suurta neuroverkkojen piirteiden kapasiteettia kehittämällä laskennallisesti tehokkaampia ja pienempiä moduuleja. Kumpiakin tekniikkoja käytetään erikseen ja yhdessä tässä väitöskirjassa.
Perinteisestä paikallisesta binäärikuvio -menetelmästä tehdään oppiva, jolloin uusia piirteitä voidaan oppia datasta, sen sijaan että ne määriteltäisiin algoritmillisesti. Uusi kehitetty oppiva versio on laskennallisesti tehokkaampi, robustimpi ja erottelevaisempi. Seuraavaksi esitellään tekniikka, joka yhdistää paikallisen binäärikuvion ja konvoluution. Kehitetty konvoluutio pystyy irrottamaan korkeamman asteen paikallista informaatiota, se on laskennallisesti tehokas ja se voidaan integroida olemassa oleviin neuroverkkoihin vaivattomasti. Sen jälkeen esitellään konvoluutioneuroverkon moduuli, joka käyttää hyväkseen ryhmäkonvoluutiota ja dynaamista suoritusta. Moduuli pitää normaalin konvoluution piirteidenirrotus kyvyn ollen kuitenkin yhtä tehokas ryhmäkonvoluution kanssa laskennallisesti. Lopuksi esitetään binäärineuroverkko moduuli robustiin pistepilven analyysiin. Kehitetty moduuli parantaa laskennallista tehokkuutta hyödyntämällä sekä binarisointia että pyörimis-invarianttia konvoluutiota
Pyrolysis of trifluoroacetaldehyde, initiated by di-tertiary-butyl peroxide decomposition
No full textPT: J; CR: ARTHUR NL, 1965, AUST J CHEM, V18, P1561 AYSCOUGH PB, 1956, J CHEM PHYS, V24, P994 BATT L, 1977, INT J CHEM KINET, V9, P141 COME GM, 1968, REV I R PETROLE, V23, P1365 COX DL, 1966, J CHEM SOC B, P245 DODD RE, 1957, J CHEM SOC, P1465 FERGUSON JM, 1965, J CHEM SOC, P4416 GRAY P, 1971, CHEM REV, V71, P247 HIATT R, 1972, INT J CHEM KINET, V4, P479 HIATT R, 1978, INT J CHEM KINET, V10, P185 HOOPER DG, 1975, J CHEM EDUC, V52, P131 LIU MH, 1973, CAN J CHEM, V51, P2292 LIU MTH, 1968, CAN J CHEM, V46, P479 LIU MTH, 1977, INT J CHEM KINET, V9, P589 MORRIS ER, 1967, T FARADAY SOC, V63, P2470 MORRIS ER, 1968, T FARADAY SOC, V64, P3027 PEARCE C, 1971, J CHEM SOC CHEM COMM, P1464 SHAW DH, 1968, CAN J CHEM, V46, P2721 SHEPP A, 1956, J CHEM PHYS, V24, P939 WIJNEN MHJ, 1960, J AM CHEM SOC, V82, P1847 YEEQUEE MJ, 1968, J PHYSICAL CHEMISTRY, V72, P2824 YEEQUEE MJ, 1968, T FARADAY SOC, V64, P1296; NR: 22; TC: 10; J9: CAN J CHEM; PG: 10; GA: HN360Source type: Electronic(1
A Subband-Selective Broadband GSC with Cosine-Modulated Blocking Matrix
No full textIn this paper, a novel subband-selective generalized sidelobe canceller (GSC) for partially adaptive broadband beamforming is proposed. The columns of the blocking matrix are derived from a prototype vector by cosine-modulation, and the broadside constraint is incorporated by imposing zeros on the prototype vector appropriately. These columns constitute a series of bandpass filters, which select signals with specific angles of arrival and frequencies. This results in highpass-type bandlimited spectra of the blocking matrix outputs, which is further exploited by subbands decomposition and suitably discarding the low-pass subbands prior to running independent unconstrained adaptive filters in each non-redundant subband. By these steps, the computational complexity of a GSC implementation is greatly reduced compared to fully adaptive GSC schemes, while performance is comparable or even enhanced due to subband decorrelation in both spatial and temporal domains
Study of rare B-c(+)-> D(d,s)((*)+)l(l)over-bar decays
No full text[[abstract]]We study the rare decays of B-c(+)-->D(q)((*)+)l (l) over bar (q=d,s and l=nu(l),e,mu,tau) in the standard model. The form factors are evaluated in the light front and constituent quark models, respectively. We find that the decay branching ratios calculated in the two models for B-c(+)-->D(q)(+)l (l) over bar agree well with each other, whereas those for B-c(+)-->D(q)(*+)l (l) over bar are different.[[fileno]]2010129010062[[department]]物理
Insights into the antibacterial mechanism of PEGylated nano-bacitracin A against Streptococcus pneumonia: both penicillin-sensitive and penicillin-resistant strains [Corrigendum]
No full textInsights into the antibacterial mechanism of PEG ylated nano-bacitracin A against Streptococcus pneumonia: both penicillin-sensitive and penicillinresistant strains [Corrigendum]Hong W, Liu L, Zhang Z, Zhao Y, Zhang D, Liu M. Int J Nanomedicine. 2018;13:6297–6309.The author wishes to advise that on page 6306, Figure 7 parts A and B are incorrect. The correct figure parts are included below:Read the original articl
Database of Human Osteoblast Gene Expression without and with ACTH
No full textThese data are used in multiple grants and papers. For description of methods and some specific results, please see Blair HC, Larrouture QC, Li Y, Lin H, Beer-Stoltz D, Liu L, Tuan RS, Robinson LJ, Schlesinger PH, Nelson DJ. Osteoblast Differentiation and Bone Matrix Formation In Vivo and In Vitro. Tissue Eng Part B Rev23(3):268-280, 2017. doi: 10.1089/ten.TEB.2016.0454
Study of decay dynamics and CP asymmetry in D+ -> K(L)(0)e(+)nu(e) decay
No full textUsing 2.92 fb(-1) of electron-positron annihilation data collected at root s = 3.773 GeV with the BESIII detector, we obtain the first measurements of the absolute branching fraction B(D+ -> K(L)(0)e(+)nu(e)) = (4.481 +/- 0.027(stat) +/- 0.103(sys))% and the CP asymmetry A(CP)(D+-> KL0e+nu e) = (-0.59 +/- 0.60(stat) +/- 1.48(sys))%. From the D+ -> K(L)(0)e(+)nu(e) differential decay rate distribution, the product of the hadronic form factor and the magnitude of the Cabibbo-Kobayashi-Maskawa matrix element, f(+)(K)(0)vertical bar V-cs vertical bar, is determined to be 0.728 +/- 0.006(stat) +/- 0.011(sys). Using vertical bar V-cs vertical bar from the SM constrained fit with the measured f(+)(K)(0)vertical bar V-cs vertical bar, f(+)(K)(0) = 0.748 +/- 0.007(stat) +/- 0.012(sys) is obtained, and utilizing the unquenched Lattice QCD (LQCD) calculation for f(+)(K)(0), vertical bar V-cs vertical bar = 0.975 +/- 0.008(stat) +/- 0.015(sys) +/- 0.025(LQCD)
Difference based Ridge and Liu type Estimators in Semiparametric Regression Models
No full textWe consider a difference based ridge regression estimator and a Liu type estimator of the regression parameters in the partial linear semiparametric regression model, y = Xβ + f + ε. Both estimators are analysed and compared in the sense of mean-squared error. We consider the case of independent errors with equal variance and give conditions under which the proposed estimators are superior to the unbiased difference based estimation technique. We extend the results to account for heteroscedasticity and autocovariance in the error terms. Finally, we illustrate the performance of these estimators with an application to the determinants of electricity consumption in Germany.Difference based estimator; Differencing estimator, Differencing matrix, Liu estimator, Liu type estimator, Multicollinearity, Ridge regression estimator, Semiparametric model
- …
