the initial setup of the scene is performed only once at the beginning, a few seconds of processing time is well within the acceptable range. Our experiments also showed that the proposed methods provided an increase in accuracy compared to previous or standard solutions. Thus the proposed algorithms were shown to be efficient for the problem described above, furthermore, they were shown not to be limited to a single application.
In conclusion, our methods provide a feasible scheme for automatically pairing vir-tual objects to real placeholders in an Adaptive Mixed Reality environment. It is worth noting, that while the error rate of our method is relatively low (0.62% on the real image-based dataset), it still provides erroneous assignments at times. These, however only present a mild inconvenience to the user, as the occasional error could be easily corrected manually. Without automatic pairing, manual assignment would have to be performed for all virtual objects. For reproducibility, the implementa-tion of these methods, tests, as well as the datasets used for evaluaimplementa-tion are available online at Szemenyei [37].
problematic in low power embedded systems. Notably, neural network-based graph classification methods failed in our case at least partly due to small dataset sizes.
Nowadays, however, both of these drawbacks have been lessened considerably, mak-ing it worthwhile to replace parts (or all) of the proposed method with deep neural network-based solutions. Considerable research has been done on reducing overfit-ting via pre-training networks on simulated data [10, 11], as well as high-efficiency inference in embedded systems [10]. Moreover, advances in self-supervised learn-ing [12, 1] can also alleviate the need for labeled data.
The easiest way to create a fully neural pipeline would be to replace the segmen-tation and individual node classification steps with a 3D semantic segmensegmen-tation or object detection network. A few advances have been made in recent years to reduce overfitting in case of small datasets [14, 13]. There are several worthy research goals that could be pursued here, including the robust, yet efficient application of convo-lutional networks for 3D data [124], or the application of sample-efficient learning in this context, which would be required if we wish to allow developers to introduce new virtual object categories.
Lastly, the most challenging research goal would be the replacement of the final global optimization step with a fully neural alternative. Using neural networks for global parameter optimization is a novel and challenging field of research, which has been mostly applied to hyperparameter search. Applying or extending modern meta-learning techniques [125] for this problem would be a major contribution.
Márton Szemenyei 90/130 CONCLUSION
List of Figures
1.1 The HTC Vive VR glasses and controllers (Source: vive.com ). . . 9 1.2 The Tiles (top) [31] and MagicCup (bottom) [32] systems. . . 10 2.1 The attention mechanism. SS and ST are the lengths of the source
and target sequences with F features. FQK is the number of key and query vector features, andFV is the depth of the output. In the case of self-attention, the source and target sequences are the same. . . 22 2.2 The graph constructed from primitive shapes (only close edges are
drawn). A few example node vectors are displayed. . . 24 2.3 Bayesian t-test comparing the raw and explicit embeddings. The
figure shows the posterior distributions of the difference between the performances of the two method, with the black line displaying the 95% Credible Interval (CI). In the top left corner the probabilities of the difference being less than and greater than zero are displayed in green. . . 34 2.4 Bayesian t-test comparing the raw and RWK embeddings. . . 34 2.5 Bayesian t-test comparing the explicit and RWK embeddings. . . 35 2.6 Bayesian t-test comparing the RWK and explicit embeddings’
ten-dency to overfit. . . 36 2.7 Bayesian t-test comparing the raw and explicit embedding on the
scene versions datasets. . . 37 2.8 Bayesian t-test comparing the neural baseline and explicit embedding
on the scene versions datasets. . . 37 3.1 Typical cases: separable nodes but indistinguishable means (a),
where LDA fails to separate between nodes (b). The solid line is the dimension selected by LDA, while the dashed line shows a dimension that separates all subclasses. . . 47
3.2 The singular values of the between class and the within instance scatter matrices. With the between class scatter matrix, a clear knee point is visible in the singular values, while the knee is much less pronounced in the within instance scatter matrix. . . 49 3.3 A simple case where the vertical dimension, which is only required to
separate between nodes of the same class is included in the between subclass scatter. . . 53 3.4 Average ranks of the different methods. . . 59 3.5 Bayesian t-test between the SSCDA and SCDA methods forawi(top)
and ac (bottom). . . 60 3.6 Bayesian t-test between the SSCDA and SURF methods forawi(top)
and ac (bottom). . . 61 4.1 The mutation operator: (a) is the original setup, (b) is the traditional
flip mutation. (c) is flip mutation using drag, and (d) shows label permutation. . . 73 4.2 The crossover operator: (a) and (b) are the parents, (c) is the offspring. 74 4.3 Example results from the real image localization set. The box/book
category is highlighted as green, mug/can is red, ball is blue and pointer is yellow. The second figure shows a typical failure case: a distant, small object blends into the ground/desk shape. . . 77 4.4 Bayesian t-test between the Greedy and the GA methods for eopt
(top) and ec (bottom). . . 79 4.5 Bayesian t-test between the SA and the GA methods for eopt (top)
and ec (bottom). . . 80 4.6 Bayesian t-test between esvm and ec using GA and explicit embedding. 81 4.7 Bayesian t-test between no embedding and using explicit embedding
for ecost (top) and ec (bottom). . . 83 4.8 Bayesian t-test comparing the raw and explicit embedding on the
scene versions datasets. . . 84 4.9 Bayesian t-test between vanilla GA and using all custom operators
for eopt (top) and ec (bottom). . . 85 A.1 A few example images from the synthetic image dataset and the
corresponding reconstructions in VisualSFM. . . 111
Márton Szemenyei 92/130 LIST OF FIGURES
A.2 A few example images from the real image dataset and the corre-sponding reconstructions in VisualSFM. . . 112 A.3 Example images from the synthetic image (top) and the real image
(bottom) scene sets. . . 113 A.4 Bayesian t-test between SCDA and LDA forawi. . . 116 A.5 Bayesian t-test between SCDA and LDA forac. . . 116 A.6 Bayesian t-test between SCDA and SDA forawi. . . 117 A.7 Bayesian t-test between SCDA and SDA forac. . . 117 A.8 Bayesian t-test between SSCDA and LDA forawi. . . 117 A.9 Bayesian t-test between SSCDA and LDA forac. . . 118 A.10 Bayesian t-test between SSCDA and SDA for awi. . . 118 A.11 Bayesian t-test between SSCDA and SDA for ac. . . 118 A.12 Bayesian t-test between SDA-IC and LDA for awi. . . 119 A.13 Bayesian t-test between SDA-IC and LDA for ac. . . 119 A.14 Bayesian t-test between SDA-IC and SDA forawi. . . 119 A.15 Bayesian t-test between SDA-IC and SDA forac. . . 120 A.16 Bayesian t-test between Wilks’ and LDA for awi. . . 120 A.17 Bayesian t-test between Wilks’ and LDA for ac. . . 120 A.18 Bayesian t-test between Wilks’ and SDA for awi. . . 121 A.19 Bayesian t-test between Wilks’ and SDA for ac. . . 121 A.20 Bayesian t-test between the breakpoint method and no rank
adjust-ment for ac. . . 121 A.21 Bayesian t-test between the breakpoint method and no rank
adjust-ment for the number of dimensions. . . 122 A.22 Bayesian t-test between the information retained method and no rank
adjustment for ac. . . 122 A.23 Bayesian t-test between the information retained method and no rank
adjustment for the number of dimensions. . . 122 A.24 Bayesian t-test between the class-number based method and no rank
adjustment for ac. . . 123
A.25 Bayesian t-test between the class-number based method and no rank adjustment for the number of dimensions. . . 123 A.26 Bayesian t-test between the iterative method and no rank adjustment
for ac. . . 123 A.27 Bayesian t-test between the iterative method and no rank adjustment
for the number of dimensions. . . 124 A.28 Bayesian t-test between breakpoint and iterative rank selection for ac. 124 A.29 Bayesian t-test between breakpoint and iterative rank selection for
the number of dimensions. . . 124 A.30 Bayesian t-test between PCA and iterative rank selection for ac. . . . 125 A.31 Bayesian t-test between PCA and iterative rank selection for the
number of dimensions. . . 125 A.32 Bayesian t-test between class-number based and iterative rank
selec-tion for ac. . . 125 A.33 Bayesian t-test between class-number based and iterative rank
selec-tion for the number of dimensions. . . 126 A.34 Bayesian t-test between vanilla GA and using the custom
initializa-tion operator for eopt. . . 126 A.35 Bayesian t-test between vanilla GA and using the custom
initializa-tion operator for ec. . . 126 A.36 Bayesian t-test between vanilla GA and using custom mutation
op-erator for eopt. . . 127 A.37 Bayesian t-test between vanilla GA and using custom mutation
op-erator for ec. . . 128 A.38 Bayesian t-test between vanilla GA and using custom crossover
op-erator for eopt. . . 129 A.39 Bayesian t-test between vanilla GA and using custom crossover
op-erator for ec. . . 130
Márton Szemenyei 94/130 LIST OF FIGURES
List of Tables
2.1 Features and reference frames for every primitive shape type . . . 24 2.2 Node-by-node classification errors on the two synthetic datasets . . . 33 2.3 Node-by-node classification errors on the two image-based datasets . 33 2.4 Summary of graph node embedding method capabilities. . . 39 3.1 Methods used for comparison . . . 54 3.2 Results of the algorithms for the four synthetic datasets. Our
meth-ods are in bold. . . 57 3.3 Results of the algorithms for the special synthetic datasets. . . 57 3.4 The algorithms’ performance on 3D shape graph datasets. . . 58 3.5 The methods’ accuracy on image datasets with few (<5) classes. . . 58 3.6 The methods’ accuracy on image datasets with many (>5) classes. . 58 3.7 The comparison of our methods to the SURF feature extractor . . . . 59 3.8 The 95% credible intervals of the Bayesian t-tests . . . 59 3.9 Comparison of methods for selecting the rank of the within instance
scatter matrix. Note that the number of dimensions is also reported as a percent of the maximum . . . 62 3.10 The 95% credible intervals of the Bayesian t-tests . . . 62 4.1 Result of the scene optimization . . . 79 4.2 Change in errors caused by the special genetic operators. . . 81 4.3 Results before and after the context optimization . . . 84 A.1 Node-by-node classification errors on all datasets . . . 114 A.2 Node-by-node classification errors on all scene datasets . . . 115
A.3 Methods ranked on the datasets according toac . . . 115 A.4 Methods ranked on the datasets according toawi . . . 116 A.5 Localization results with different optimization methods. . . 127 A.6 Localization results with different genetic operators. . . 128 A.7 Localization results with different embeddings. . . 129 A.8 Node-by-node classification accuracy achieved by the Graph
Self-Attention baseline . . . 130
Márton Szemenyei 96/130 LIST OF TABLES
References
Thesis Publications
[1] M. Szemenyei and P. Reizinger, “Attention-based curiosity in multi-agent reinforcement learning environments,” in 2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO), IEEE, May 2019.doi: 10.1109/iccairo47923.2019.00035.
[2] M. Szemenyei, “Neural graph node classification vie self-attention,” in Work-shop on the Advances of Information Technology, B. Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2020, pp. 33–38.
[3] M. Szemenyei and F. Vajda, “3d object detection and scene optimization for tangible augmented reality,” Periodica Polytechnica Electrical Engineering and Computer Science, vol. 62, no. 2, pp. 25–37, 2018. doi:10.3311/ppee.
10482.
[4] M. Szemenyei and F. Vajda, “Learning 3d object recognition using graphs based on primitive shapes,” in Workshop on the Advances of Information Technology, B. Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2015, pp. 67–71.
[5] M. Szemenyei and F. Vajda, “3d object detection using vectorial graph node embedding,” in Workshop on the Advances of Information Technology, B.
Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2017, pp. 45–53.
[6] M. Szemenyei and F. Vajda, “Dimension reduction for structured composite classes in multi-object environments,” inProceedings of the 15th International Conference on Artificial Intelligence, Knowledge Engineering and Data Bases (AIKED), V. Mladenov, Ed., ser. Recent Advances in Electrical Engineering, WSEAS, vol. 58, WSEAS Press, 2016, pp. 134–141.
[7] M. Szemenyei and F. Vajda, “Dimension reduction for objects composed of vector sets,” International Journal of Applied Mathematics and Computer Science, vol. 27, no. 1, pp. 169–180, 2017. doi:10.1515/amcs-2017-0012.
[8] M. Szemenyei and F. Vajda, “Optimal feature selection for objects composed of vector sets,” inHungarian Conference on Computer Graphics and Geom-etry, L. Szirmay-Kalos and G. Renner, Eds., NJSZT, 2016, pp. 7–14.
[9] M. Szemenyei, “Evolutionary scene arrangement for adaptive augmented re-ality systems,” in Workshop on the Advances of Information Technology, B.
Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2018, pp. 10–17.
Unrelated Publications
[10] M. Szemenyei and V. Estivill-Castro, “Real-time scene understanding using deep neural networks for RoboCup SPL,” inRoboCup 2018: Robot World Cup XXII, Springer International Publishing, 2019, pp. 96–108. doi: 10.1007/
978-3-030-27544-0_8.
[11] M. Szemenyei and V. Estivill-Castro, “ROBO: Robust, fully neural object de-tection for robot soccer,” inRoboCup 2019: Robot World Cup XXIII, Springer International Publishing, 2019, pp. 309–322. doi: 10 . 1007 / 978 3 030 -35699-6_24.
[12] P. Reizinger and M. Szemenyei, “Attention-based curiosity-driven exploration in deep reinforcement learning,” inICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, May 2020.doi: 10.1109/icassp40776.2020.9054546.
[13] C. Guo, M. Szemenyei, Y. Yi, Y. Xue, W. Zhou, and Y. Li, “Dense residual network for retinal vessel segmentation,” inICASSP 2020 - 2020 IEEE Inter-national Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, May 2020. doi: 10.1109/icassp40776.2020.9054290.
[14] C. Guo, M. Szemenyei, Y. Pei, Y. Yi, and W. Zhou, “SD-unet: A structured dropout u-net for retinal vessel segmentation,” in2019 IEEE 19th Internati-onal Conference on Bioinformatics and Bioengineering (BIBE), IEEE, Oct.
2019. doi:10.1109/bibe.2019.00085.
[15] M. Szemenyei and F. Vajda, “Learning shape matching for augmented re-ality,” in Hungarian Conference on Computer Graphics and Geometry, L.
Szirmay-Kalos and G. Renner, Eds., NJSZT, 2014, pp. 126–133.
[16] M. Szemenyei and V. Estivill-Castro, “Deep semantic segmentation for robot soccer,” inWorkshop on the Advances of Information Technology, L. Szirmay-Kalos and G. Renner, Eds., NJSZT, 2018, pp. 109–117.
[17] M. Szemenyei, “Battling transaction costs: Establishing an e-exchange system for coaseian bargaining,” E-conom, vol. V, no. 1, 2016. doi: 10.17836/ec.
2016.1.124.
Márton Szemenyei 98/130 REFERENCES
[18] M. Szemenyei and T. Princz-Jakovics, “Informatika a fenntarthatóságért – online platform a magánalkuk segítésére,” in Fenntarthatóság - utópia vagy realitás?, N. N. Csigéné and T. Princz-Jakovics, Eds., BME, 2016, pp. 69–83.
[19] M. Szemenyei and T. Princz-Jakovics, “Informatikai megoldások környezeti problémák kezelésére,” Magyar Tudomány, vol. 175, no. 4, pp. 1086–1986, 2014.
[20] M. Szemenyei, Számítógépes látás a közigazgatásban - Alapfeladatok és al-kalmazások, Z. Dalloul, Ed., ser. Okosváros technológiák - A technológia fe-jlődésének iránya és hatása. Dialóg Campus, 2019, isbn: 978-615-5945-36-6.
[Online]. Available: https://akfi-dl.uni-nke.hu/pdf_kiadvanyok/Web_
PDF_Szamitogepes_latas_a_kozigazgatasban.pdf.
[21] P. Reizinger and M. Szemenyei, “Simultaneous cooperation and competi-tion in reinforcement learning,” inWorkshop on the Advances of Information Technology, B. Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2020, pp. 56–61.
[22] M. Szemenyei, “Efficient object detection in robot soccer environments using deep learning,” in Workshop on the Advances of Information Technology, B.
Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2019, pp. 187–195.
[23] K. Póka, S. Hornyák, and M. Szemenyei, “Controlled human face synthesis with generative adversarial networks,” in Workshop on the Advances of In-formation Technology, B. Kiss and L. Szirmay-Kalos, Eds., BME-IIT, 2020, pp. 47–55.
Other References
[24] H. Papagiannis, Augmented Human. O’Reilly UK Ltd., Oct. 1, 2017, isbn:
1491928328. [Online]. Available: https : / / www . ebook . de / de / product / 24777492/helen_papagiannis_augmented_human.html.
[25] J. Vince, Introduction to Virtual Reality. Springer London, Mar. 30, 2004, 188 pp., isbn: 1852337397. [Online]. Available: https: / / www . ebook . de / de/product/2630353/john_vince_introduction_to_virtual_reality.
html.
[26] D. Chatzopoulos, C. Bermejo, Z. Huang, and P. Hui, “Mobile augmented reality survey: From where we are to where we go,” IEEE Access, vol. 5, pp. 6917–6950, 2017. doi: 10.1109/access.2017.2698164.
[27] H. Ishii and B. Ullmer, “Tangible bits,” inProceedings of the SIGCHI confer-ence on Human factors in computing systems - CHI ’97, ACM Press, 1997.
doi: 10.1145/258549.258715.
[28] H. Liang, J. Yuan, D. Thalmann, and N. M. Thalmann, “AR in hand,” in Proceedings of the 23rd ACM international conference on Multimedia - MM
’15, ACM Press, 2015. doi: 10.1145/2733373.2807972.
[29] D. Bandyopadhyay, R. Raskar, and H. Fuchs, “Dynamic shader lamps: Paint-ing on movable objects,” inProceedings of the IEEE and ACM International Symposium on Augmented Reality (ISAR’01), ser. ISAR ’01, Washington, DC, USA: IEEE Computer Society, 2001, pp. 207–,isbn: 0-7695-1375-1. [On-line]. Available: http://dl.acm.org/citation.cfm?id=582828.881324.
[30] K. Matkoviv, T. Psik, I. Wagner, and D. Gracanin, “Dynamic texturing of real objects in an augmented reality system,” in IEEE Proceedings Virtual Reality, IEEE, 2005. doi:10.1109/vr.2005.1492781.
[31] M. Billinghurst, H. Kato, and I. Poupyrev, “Tangible augmented reality,”
in ACM SIGGRAPH ASIA 2008 courses on - SIGGRAPH Asia ’08, ACM Press, 2008. doi: 10.1145/1508044.1508051.
[32] M. Billinghurst, H. Kato, and S. Myojin, “Advanced interaction techniques for augmented reality applications,” in Lecture Notes in Computer Science, Springer Berlin Heidelberg, 2009, pp. 13–22. doi: 10 . 1007 / 978 3 642 -02771-0_2.
[33] G. A. Lee, M. Billinghurst, and G. J. Kim, “Occlusion based interaction methods for tangible augmented reality environments,” inProceedings of the 2004 ACM SIGGRAPH international conference on Virtual Reality contin-uum and its applications in industry - VRCAI ’04, Los Angeles, CA USA:
ACM, 2004, pp. 419–426.doi: 10.1145/1044588.1044680.
[34] W. Broll, E. Meier, and T. Schardt, “The virtual round table - a collaborative augmented multi-user environment,” inProceedings of the third international conference on Collaborative virtual environments - CVE ’00, San Fransisco, CA USA: ACM, 2000, pp. 39–46.doi: 10.1145/351006.351011.
[35] A. Golonivskiy, V. G. Kim, and T. Funkhouser, “Shape-based recognition of 3d point clouds in urban environments,” in 2009 IEEE 12th International Conference on Computer Vision, Kyoto, Japan: IEEE, 2009, pp. 2154–2161.
doi: 10.1109/iccv.2009.5459471.
Márton Szemenyei 100/130 REFERENCES
[36] J. K. Kruschke, “Bayesian estimation supersedes the t test,” Journal of Ex-perimental Psychology: General, vol. 142, no. 2, pp. 573–603, 2013. doi:
10.1037/a0029146.
[37] M. Szemenyei,Phd research code. [Online]. Available:https://github.com/
szemenyeim/PhD-Research.
[38] R. Osada, T. Funkhouser, B. Chazelle, and D. David, “Matching 3d models with shape distributions,” in Proceedings International Conference on Shape Modeling and Applications, Genova: IEEE, 2001, pp. 154–166.doi:10.1109/
sma.2001.923386.
[39] F. Tombari and L. Di Stefano, “Object recognition in 3d scenes with occlu-sions and clutter by hough voting,” in 2010 Fourth Pacific-Rim Symposium on Image and Video Technology, Singapore: IEEE, 2010, pp. 349–355. doi:
10.1109/psivt.2010.65.
[40] M. A. Fishler and R. C. Bolles, “Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartogra-phy,” Magazine Communications of the ACM, vol. 24, pp. 381–395, 6 1981.
doi: 10.1145/358669.358692.
[41] S. Lazebnik, C. Schmid, and J. Ponce, “A sparse texture representation using local affine regions,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 27, pp. 1265–1278, 2005. doi: 10.1109/TPAMI.2005.151.
[42] R. Rusu, N. Blodow, and M. Beetz, “Fast point feature histograms (fpfh) for 3d registration,” in2009 IEEE International Conference on Robotics and Automation, Kobe: IEEE, 2009. doi:10.1109/ROBOT.2009.5152473.
[43] Z. Wu, S. Song, A. Khosla, L. Z. F. Yu, X. Tang, and J. Xiao, “3d shapenets:
A deep representation for volumetric shape modeling,” in 2015 IEEE Con-ference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2015.
doi: 10.1109/CVPR.2015.7298801.
[44] C. Szegedy, W. Liu, and Y. Jia, “Going deeper with convolutions,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2015. doi:10.1109/CVPR.2015.7298594.
[45] C. Wu, I. Lenz, and A. Saxena, “Hierarchical semantic labeling for task-relevant rgb-d perception,” in Robotics: Science and Systems X, Robotics:
Science and Systems Foundation, 2014. doi: 10.15607/rss.2014.x.006.
[46] M. Schwarz, H. Schulz, and S. Behnke, “Rgb-d object recognition and pose estimation based on pre-trained convolutional neural network features,” in 2015 IEEE International Conference on Robotics and Automation (ICRA), Seattle: IEEE, 2015. doi: 10.1109/icra.2015.7139363.
[47] S. Bai, X. Bai, Z. Zhou, Z. Zhang, and L. J. Latecki, “Gift: A real-time and scalable 3d shape search engine.,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2016. doi: 10.1109/CVPR.
2016.543.
[48] S. Gupta, R. Girshick, P. Arbel�aez, and J. Malik, “Learning rich features from rgb-d images for object detection and segmentation,” in Computer Vi-sion – ECCV 2014, Springer International Publishing, 2014, pp. 345–360.
doi: 10.1007/978-3-319-10584-0_23.
[49] Y. Bengio, “Practical recommendations for gradient-based training of deep architectures,” in Neural Networks: Tricks of the Trade. Springer, 2012, pp. 437–478.doi: 10.1007/978-3-642-35289-8\_26.
[50] L. Bottou, “Stochastic gradient descent tricks,” in Neural Networks, Tricks of the Trade, Reloaded. Springer, 2012, pp. 430–445. doi: 10.1007/978-3-642-35289-8\_25.
[51] S. Sabour, N. Frosst, and G. E. Hinton, “Dynamic routing between capsules,”
Oct. 26, 2017. arXiv: 1710.09829v2 [cs.CV].
[52] R. Schnabel, R. Wahl, and R. Klein, “Efficient ransac for point-cloud shape detection,”Computer Graphics Forum, vol. 26, no. 2, pp. 214–226, 2007.doi:
10.1111/j.1467-8659.2007.01016.x.
[53] R. Schnabel, R. Wessel, R. Wahl, and R. Klein, “Shape recognition in 3d point-clouds,” in The 16-th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision’2008, V. Skala, Ed., UNION Agency-Science Press, 2008, isbn: 978-80-86943-15-2.
[54] R. Sharan and T. Ideke, “Modeling cellular machinery through biological network comparison,” Nature Biotechnology, vol. 24, pp. 427–433, 4 2006.
doi: 10.1038/nbt1196.
[55] R. Kumar, J. Novak, and A. Tomkins, “Structure and evolution of online social networks,” in Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD ’06, ACM Press, 2006. doi:10.1145/1150402.1150476.
Márton Szemenyei 102/130 REFERENCES
[56] M. F. Demirci, Y. Osmanlioglu, A. Shokoufandeh, and S. Dickinson, “Effi-cient many-to-many feature matching under the l1 norm,” Journal of Com-puter Vision and Image Understanding, vol. 115, pp. 976–983, 7 2011. doi:
10.1016/j.cviu.2010.12.012.
[57] R. C. Wilson, E. R. Hancock, and B. Luo, “Pattern vectors from algebraic graph theory,” IEEE Transactions on Pattern Analysis and Machine Intelli-gence, vol. 27, pp. 1112–1124, 7 2005. doi: 10.1109/tpami.2005.145.
[58] F. Chung,Spectral graph theory. American Mathematical Society, 1997. doi:
10.1090/cbms/092.
[59] M. Ferrer, F. Serratosa, and A. Sanfeliu, “Synthesis of median spectral graph,” in Lecture Notes in Computer Science. Springer, 2005, vol. 3523, pp. 139–146.doi: 10.1007/11492542\_18.
[60] D. White and R. C. Wilson, “Mixing spectral representations of graphs,”
in 18th International Conference on Pattern Recognition (ICPR’06), IEEE, 2006. doi:10.1109/icpr.2006.803.
[61] P. Zhu and R. C. Wilson, “Stability of the eigenvalues of graphs,” inComputer Analysis of Images and Patterns, Versailles: Springer, 2005, pp. 371–378.doi:
10.1007/11556121_46.
[62] P. Riba, J. Lladós, A. Fornés, and A. Dutta, “Large-scale graph indexing using binary embeddings of node contexts for information spotting in doc-ument image databases,” Pattern Recognition Letters, vol. 87, pp. 203–211, 2017. doi:10.1016/j.patrec.2016.06.015.
[63] S. V. N. Vishwanathan, N. N. Schraudolph, R. Kondor, and K. M. Borgwardt,
“Graph kernels,”Journal of Machine Learning, vol. 11, pp. 1201–1242, 2010.
[64] A. J. Smola and R. Kondor, “Kernels and regularization on graphs,” in Learn-ing Theory and Kernel Machines, SprLearn-inger Berlin Heidelberg, 2003, pp. 144–
158. doi: 10.1007/978-3-540-45167-9_12.
[65] R. Kondor and K. M. Borgwardt, “The skew spectrum of graphs,” in Pro-ceedings of the 25th international conference on Machine learning - ICML
’08, ACM Press, 2008. doi: 10.1145/1390156.1390219.
[66] T. Gartner, P. Flach, and S. Wrobel, “On graph kernels: Hardness results and efficient alternatives,” in Learning Theory and Kernel Machines, Springer Berlin Heidelberg, 2003, pp. 129–143. doi: 10.1007/978- 3- 540- 45167-9_11.
[67] F. Scarselli, M. Gori, A. C. Tsoi, M. Hagenbuchner, and G. Monfardini,
“The graph neural network model,” IEEE Transactions on Neural Networks, vol. 20, no. 1, pp. 61–80, Jan. 2009. doi: 10.1109/tnn.2008.2005605.
[68] J. Zhou, G. Cui, Z. Zhang, C. Yang, Z. Liu, L. Wang, C. Li, and M. Sun,
“Graph neural networks: A review of methods and applications,” Dec. 20, 2018. arXiv:1812.08434v4 [cs.LG].
[69] B. Perozzi, R. Al-Rfou, and S. Skiena, “DeepWalk,” in Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD ’14, ACM Press, 2014.doi:10.1145/2623330.2623732.
[70] P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y. Bengio,
“Graph attention networks,” Oct. 30, 2017. arXiv:1710.10903v3 [stat.ML].
[71] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł.
Kaiser, and I. Polosukhin, “Attention is all you need,” inAdvances in Neural Information Processing Systems 30, I. Guyon, U. V. Luxburg, S. Bengio, H.
Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, Eds., Curran Asso-ciates, Inc., 2017, pp. 5998–6008. [Online]. Available:http://papers.nips.
cc/paper/7181-attention-is-all-you-need.pdf.
[72] J. B. Lee, R. Rossi, and X. Kong, “Graph classification using structural at-tention,” inProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, ACM, Jul. 2018. doi: 10.1145/
3219819.3219980.
[73] D. Q. Nguyen, T. D. Nguyen, and D. Phung, “Universal self-attention net-work for graph classification,” Sep. 26, 2019. arXiv:1909.11855v8 [cs.LG].
[74] C. Wu, “Towards linear-time incremental structure from motion,” in Inter-national Conference on 3D Vision, IEEE, 2013, pp. 127–134.doi: 10.1109/
3DV.2013.25.
[75] J. Mockus,Bayesian Approach to Global Optimization. Springer Netherlands, 1989. doi:10.1007/978-94-009-0909-0.
[76] X. Liu, Z. Wang, Jun, and Z. Feng, “Face recognition with locality sensi-tive discriminant analysis based on matrix representation,” in 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence), IEEE, 2008, pp. 4052–4058. doi: 10.1109/
IJCNN.2008.4634380.
[77] B. Surendiran and A.Vadivel, “Feature selection using stepwise anova dis-criminant analysis for mammogram mass classification,” ACEEE Internati-onal Journal on Signal and Image Processing, vol. 2, no. 1, pp. 17–19, 2011.
Márton Szemenyei 104/130 REFERENCES
[78] S. Yasuoka, Y. Kang, K. Morooka, and H. Nagahashi, “Texture classification using hierarchical discriminant analysis,” in IEEE Conference on Systems, Man and Cybernetics, IEEE, 2004, pp. 6395–6400. doi: 10.1109/ICSMC.
2004.1401405.
[79] C. Boutsidis, A. Zouzias, M. W. Mahoney, and P. Drineas, “Stochastic di-mensionality reduction for k-means clustering,” ArXiv, vol. abs/1110.2897, 2011. [Online]. Available: http://arxiv.org/abs/1110.2897.
[80] P. Kulczycki and S. Łukasik, “An algorithm for reducing the dimension and size of a sample for data exploration procedures,” International Journal of Applied Mathematics and Computer Science, vol. 24, no. 1, pp. 133–149, 2014.
doi: 10.2478/amcs-2014-0011.
[81] F. Song, Z. Guo, and D. Mei, “Feature selection using principal component analysis,” inInternational Conference on System Science, Engineering Design and Manufacturing Informatization, IEEE, 2010, pp. 27–30. doi: 10.1109/
ICSEM.2010.14.
[82] T.-S. Lin and J. Meador, “Statistical feature extraction and selection for ic test pattern analysis,” in IEEE International Symposium on Circuits and Systems, IEEE, 1992, pp. 391–394. doi: 10.1109/ISCAS.1992.229931.
[83] I. T. Jolliffe, Principal Component Analysis. New York: Springer New York, 2006,isbn: 978-0-387-22440-4. [Online]. Available: https://www.ebook.de/
de/product/25193872/i_t_jolliffe_principal_component_analysis.
html.
[84] A. Hyvarinen, J. Karhunen, and E. Oja, Independent Component Analysis.
New York: John Wiley and Sons, 2001. doi:10.1002/0471221317.
[85] G. J. McLachlan,Discriminant Analysis and Statistical Pattern Recognition.
New York: John Wiley and Sons, 1992. doi:10.1002/0471725293.
[86] C. A. Kumar, “Analysis of unsupervised dimensionality reduction tech-niques,”Computer Science and Information Systems, vol. 6, no. 2, pp. 217–
227, 2009. doi: 10.2298/CSIS0902217K.
[87] D. G. Lowe, “Distinctive image features from scale-invariant keypoints,” In-ternational Journal of Computer Vision, vol. 60, no. 2, pp. 91–110, 2004.doi:
10.1023/B:VISI.0000029664.99615.94.
[88] F.-F. Li and P. Perona, “A bayesian hierarchical model for learning natu-ral scene categories,” in IEEE Conference on Computer Vision and Pattern Recognition, IEEE, 2005, pp. 524–531.doi: 10.1109/CVPR.2005.16.