#1. The complete title of one (or more) paper(s) published in the open literature describing the work that the author claims describes a human-competitive result: Automatic Design of Artificial Neural Networks for Gamma-Ray Detection #2. The name, complete physical mailing address, e-mail address, and phone number of EACH author of EACH paper(s): Filipe Assunção Department of Informatics Engineering Faculty of Sciences and Technology, University of Coimbra Pólo II - Pinhal de Marrocos 3030-290, Coimbra, Portugal fga@dei.uc.pt +351 239790016 João Correia Department of Informatics Engineering Faculty of Sciences and Technology, University of Coimbra Pólo II - Pinhal de Marrocos 3030-290, Coimbra, Portugal jncor@dei.uc.pt +351 239790016 Rúben Conceição Laboratório de Instrumentação e Física Experimental de Partículas Av. Prof. Gama Pinto, n.2 Complexo Interdisciplinar (3is) 1649-003 Lisboa ruben@lip.pt +351 210493600 Mário Pimenta Laboratório de Instrumentação e Física Experimental de Partículas Av. Prof. Gama Pinto, n.2 Complexo Interdisciplinar (3is) 1649-003 Lisboa pimenta@lip.pt +351 210493600 Bernardo Tomé Laboratório de Instrumentação e Física Experimental de Partículas Av. Prof. Gama Pinto, n.2 Complexo Interdisciplinar (3is) 1649-003 Lisboa bernardo@lip.pt +351 210493600 Nuno Lourenço Department of Informatics Engineering Faculty of Sciences and Technology, University of Coimbra Pólo II - Pinhal de Marrocos 3030-290, Coimbra, Portugal naml@dei.uc.pt +351 239790016 Penousal Machado Department of Informatics Engineering Faculty of Sciences and Technology, University of Coimbra Pólo II - Pinhal de Marrocos 3030-290, Coimbra, Portugal machado@dei.uc.pt +351 239790052 #3. The name of the corresponding author: Filipe Assunção #4. The abstract of the paper(s): The goal of this work is to investigate the possibility of improving current gamma/hadron discrimination based on their shower patterns recorded on the ground. To this end, we propose the use of Convolutional Neural Networks (CNNs) for their ability to distinguish patterns based on automatically designed features. In order to promote the creation of CNNs that properly uncover the hidden patterns in the data, and at the same time avoid the burden of hand-crafting the topology and learning hyper-parameters we resort to NeuroEvolution; in particular, we use Fast-DENSER++, a variant of Deep Evolutionary Network Structured Representation. The results show that the best CNN generated by Fast-DENSER++ improves by a factor of 2 when compared with the results reported by classical statistical approaches. Additionally, we experiment ensembling the 10 best generated CNNs, one from each of the evolutionary runs; the ensemble leads to an improvement by a factor of 2.3. These results show that it is possible to improve the gamma/hadron discrimination based on CNNs that are automatically generated and are trained with instances of the ground impact patterns. #5. A list containing one or more of the eight letters (A, B, C, D, E, F, G, or H) that correspond to the criteria (see above) that the author claims that the work satisfies: (B) The result is equal to or better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal. (D) The result is publishable in its own right as a new scientific result independent of the fact that the result was mechanically created. (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. (F) The result is equal to or better than a result that was considered an achievement in its field at the time it was first discovered. (G) The result solves a problem of indisputable difficulty in its field. #6. A statement stating why the result satisfies the criteria that the contestant claims (see examples of statements of human-competitiveness as a guide to aid in constructing this part of the submission): (B) The result is equal to or better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal. The paper describes the application of Artificial Neural Networks (ANNs) to gamma-ray detection; more precisely, our objective is to distinguish between gamma and proton radiations based on their ground impact patterns. We use Fast-DENSER++: a NeuroEvolution method that can effectively search for fully-trained deep networks to search for Convolutional Neural Networks (CNNs). We compare the results of F-DENSER++ to those reported by Assis et al. in [1]: the authors apply classic statistics to the same data; the results of F-DENSER++ surpass the reported by Assis et al. [1] by a factor of 2.3. The fitness is the highest value of the true positive rate of being a proton divided by the false positive rate. This is related to the fact that the observation of astrophysical gamma-ray sources relies on the identification of gamma-rays which are immersed in a huge cosmic ray (hadronic) background. The higher the fitness value is better. The fittest CNN generated by F-DENSER++, and the ensemble formed by the 10 best CNNs (one from each run) report respectively a performance of 8.34, and 9.89; the result reported by the best performing classic statistic of [1] is of 4.22. (D) The result is publishable in its own right as a new scientific result independent of the fact that the result was mechanically created. To the best of our knowledge this is the first time that the gamma/proton detection problem is addressed recurring to ANNs. The results stated above, i.e., a CNN with a performance of 8.34 prove that not only it is possible to solve the gamma/proton detection problem based on the ground pattern with ANNs, but they establish the new state-of-the-art result. Prior to applying F-DENSER++ to automatically search for the most effective CNN topology and learning strategy we have experimented with hand-designed CNN topologies; the attained performances are significantly lower than the automatically generated CNN; nonetheless, they are superior to the result of the classic methods of [1]. The results of F-DENSER++ improve the performance reported by the best classic approach by a factor of 2.3. This implies that with the same grid of sensors we perform twice better than other methods; or, that if the previous performance is deemed as good enough we require less sensors, which translates into a considerably lower investment, in the order of millions of euros. (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. The best results on the classification based on the ground impact of gamma/proton particles are performed by the application of classic methods that require the human-design and extraction of features [1,2,3]. The current work surpasses the results reported in [1] by a factor of 2.3, eliminating the need for hand-crafting the features that allow the model to distinguish between gamma and proton patterns. Further, the work by Assis et al. [1] compares the results to [2] and [3], and conclude that their variables have at least a discrimination power equal to the reported in [2]. In addition to avoid the need for hand-designing the features, the applied methodology also eliminates the trial-and-error process necessary for choosing the best CNN topology and learning strategy to effectively solve the problem; the outcome of evolution is a fully-trained model that requires no further training and can be deployed right-off evolution. (F) The result is equal to or better than a result that was considered an achievement in its field at the time it was first discovered. The literature on the topic from the physics perspective has attempted to deal with the gramma/proton discrimination by engaging in classical and statistical methods, which require the human-design and extraction of features. Following such approaches it was possible to prove that the classification between gamma and proton radiations could be automated using hand-designed features. This was an important achievement in the physics domain considering that the detection of such particles paves the path to investigating extreme phenomena in the Universe, such as gamma-rays arising from fast rotating neutron stars, or supermassive black holes. The results of the current paper surpass the performance of previous methods by a factor of 2.3, and thus are of extreme importance to the field of gamma-ray detection. (G) The result solves a problem of indisputable difficulty in its field. The paper addresses two particularly hard to solve tasks. On the one hand the gamma/proton detection based on the ground impact pattern is a challenging task; if some of the patterns are easy to identify visually because of their clustering characteristics, others are more difficult. The difficulty increases with higher primary energies. On the other hand we are recurring to ANNs to solve the problem, which are known to be hard to design and parameterise. The choice of an adequate model includes numerous stages: (i) pre-processing of the data; (ii) selection of the number, type, parameterisation and connectivity of the layers; and (iii) choice of which learning algorithm to use and its parameters. We overcome the difficulties by recurring to CNNs to distinguish between gamma and proton radiations, for their ability to deal with spatially-correlated data; to automate the design of ready-to-deploy CNNs we resort to F-DENSER++. #7. A full citation of the paper: @article{assunccao2019automatic, title={Automatic Design of Artificial Neural Networks for Gamma-Ray Detection}, author={Assun\c{c}\~ao, Filipe and Correia, Jo\~ao and Concei\c{c}\~ao, R\'uben and Pimenta, M\'ario and Tom\'e, Bernardo and Louren{\c{c}}o, Nuno and Machado, Penousal}, journal={arXiv preprint arXiv:1905.03532}, year={2019} } #8. A statement either that "any prize money, if any, is to be divided equally among the co-authors" OR a specific percentage breakdown as to how the prize money, if any, is to be divided among the co-authors: Any prize money, if any, is to be divided equally among the co-authors. #9. A statement stating why the authors expect that their entry would be the "best,": The current paper explores the application of a novel NeuroEvolution approach – F-DENSER++ – to a challenging, and hard-to-solve real-world problem: the detection of gamma/proton particles based on their ground impact patterns. The problem is of extreme importance in the physics domain, as gamma-rays constitute one of the best probes to investigate extreme phenomena in the Universe (e.g., gamma-rays from fast rotating neutron stars, or supermassive black holes). Our objective is to perform the detection of the gamma-rays on ground arrays of sensors, where we collect the impact energy patterns. To analyse the patterns, previous approaches relied majorly in statistical approaches. However, it is our opinion that ANNs, in particular, CNNs are well suit for extracting the necessary features from the ground patterns and distinguish between gamma and proton radiations. To aid in the selection of an efficient CNN we have used F-DENSER++. The results have demonstrated that our intuition was correct, with the obtained results surpassing the previous classical approaches by a factor of 2.3. The increase of performance is extremely important because it allows us to take full advantage of the grid of sensors. However, we can look to the results from another perspective: motivated by the increase in performance the method may well work with a smaller grid of sensors, which translates into significant budget savings (millions of euros). #10. An indication of the general type of genetic or evolutionary computation used, such as GA (genetic algorithms), GP (genetic programming), ES (evolution strategies), EP (evolutionary programming), LCS (learning classifier systems), GE (grammatical evolution), GEP (gene expression programming), DE (differential evolution), etc.: Grammatical Evolution #11. The date of publication of each paper. If the date of publication is not on or before the deadline for submission, but instead, the paper has been unconditionally accepted for publication and is “in press” by the deadline for this competition, the entry must include a copy of the documentation establishing that the paper meets the "in press" requirement: May, 2019 #References: [1] Assis, P., de Almeida, U.B., Blanco, A., Conceição, R., Piazzoli, B.E., De Angelis, A., Doro, M., Fonte, P., Lopes, L., Matthiae, G. and Pimenta, M., 2018. Design and expected performance of a novel hybrid detector for very-high-energy gamma-ray astrophysics. Astroparticle Physics, 99, pp.34-42. [2] Abeysekara, A.U., Albert, A., Alfaro, R., Alvarez, C., Álvarez, J.D., Arceo, R., Arteaga-Velázquez, J.C., Solares, H.A., Barber, A.S., Bautista-Elivar, N. and Becerril, A., 2017. Observation of the Crab Nebula with the HAWC Gamma-Ray Observatory. The Astrophysical Journal, 843(1), p.39. [3] Bartoli, B., Bernardini, P., Bi, X.J., Bolognino, I., Branchini, P., Budano, A., Melcarne, A.C., Camarri, P., Cao, Z., Cardarelli, R. and Catalanotti, S., 2013. TeV gamma-ray survey of the northern sky using the ARGO-YBJ detector. The Astrophysical Journal, 779(1), p.27.