logseq/pages/hls__SOSYM-23-00004493_Proof_hi_1692636004494_0.md

file:: SOSYM-23-00004493_Proof_hi_1692636004494_0.pdf file-path:: ../assets/SOSYM-23-00004493_Proof_hi_1692636004494_0.pdf

• evolutionary algorithm (EA) approach to automatically repair transformations containing many semantic errors. ls-type:: annotation hl-page:: 2 hl-color:: purple id:: 64ee16be-1627-4a5a-9d9f-1ddadf7d0ad3
• To prevent the fitness plateaus and the single fitness peak limitations from our previous work, we include the notion of social diversity as an objective for our EA to promote repair patches tackling errors that are less covered by the other patches of the population. ls-type:: annotation hl-page:: 2 hl-color:: purple id:: 64ee16d2-ad2f-48ea-810a-9dba3b9e4a4d
• n this context, MDE sees models as first-class artifacts where domain-specific modeling languages to capture specific aspects of the solution. ls-type:: annotation hl-page:: 2 hl-color:: red id:: 64ee1700-8041-446e-b39d-bfbf97fc6e92 hl-stamp:: 1693325058884
• l. ls-type:: annotation hl-page:: 2 hl-color:: red id:: 64ee1761-2a83-4564-9002-a1144dd14f1b
• when the transformation compiles but the implemented behavior is not the one that was intended by the developers, we say that it contains semantic errors. ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 64ee17ae-8038-42f9-8cb7-4e2317e9a6e4 hl-stamp:: 1693325233231
• Because semantic errors pertain to the transformation’s behavior and each faulty transformation needs tailored patches, predefined patches are not well-suited for semantic errors. ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 64ee17c5-8cf4-4236-953b-bc7fbd65c099
• o fix errors related to a transformation’s behavior, automated approaches usually rely on a specification of the expected behavior (e.g., test cases or examples) to assess the fitness of a patch, and thus efficiently guide the search strategy. ls-type:: annotation hl-page:: 3 hl-color:: purple id:: 64ee1ae5-2d5a-4958-ba91-de31fb6b2039 hl-stamp:: 1693326055885
• utomated patch generation for fixing semantic errors in ATL transformation rules. ls-type:: annotation hl-page:: 25 hl-color:: green id:: 64ee1b0a-3320-47f4-8e0e-559ef6024ffc hl-stamp:: 1693326092268
• This approach usually finds patches to correct transformations having fewer errors, but in the presence of more errors, the approach cannot find a solution or will take too long to converge toward suitable patches ls-type:: annotation hl-page:: 3 hl-color:: yellow id:: 64eeefb2-14e6-4759-8d50-a309b2b7e88d hl-stamp:: 1693380531967
• EAs are known to give more power to good solutions, which can cause converging issues due to loss of diversity, a problem known as single fitness peak. ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 64eef00f-1c9d-4a1d-ad79-e9969b7388bc hl-stamp:: 1693380625482
• sing behavior specifications such as test cases to guide the search in EAs can exacerbate these limitations ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 64eef01c-b908-4377-944c-45ddfad615fe
• o improve the efficiency and effectiveness of EAs using test cases, our improved approach leverages the notion of social diversity [4]. ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 64eef044-1fcd-4094-a953-f832ba08adb1
• Our hypothesis is that including this measure in the process will maintain or improve the diversity of the patches, thereby reducing the negative impact on convergence of single fitness peak and fitness plateau ls-type:: annotation hl-page:: 3 hl-color:: yellow id:: 64eef068-367d-4ffe-b559-70822646ff3b hl-stamp:: 1693380715118
• e formulate the transformation repair as a multi-objective optimization problem, where solutions must optimize several objectives including social diversity. Our approach is implemented using the NSGA-II algorithm, a fast multi-objective EA [10] ls-type:: annotation hl-page:: 3 hl-color:: purple id:: 64eef09c-7959-4d17-9aaf-d372f7b1d44c hl-stamp:: 1693380766522
• The evaluation shows that social diversity is able to improve both the efficiency and the efficacy of EAs to fix faulty transformations, even when they contain up to five semantic errors ls-type:: annotation hl-page:: 4 hl-color:: purple id:: 64ef04a0-20d3-4582-9d86-9113d8869909
• We then present the types of errors that can be found in such transformations, including semantic errors, which are the target of this work ls-type:: annotation hl-page:: 4 hl-color:: green id:: 64ef04de-145d-4ba9-a13e-02f48b796c6d hl-stamp:: 1693385953481
• hus, a given transformation is defined for a pair of meta-models, ls-type:: annotation hl-page:: 4 hl-color:: yellow id:: 64ef075b-0807-4ba9-9d00-eca724425600 hl-stamp:: 1693386589091
• Syntactic errors can be due to type misuse such as referring to elements that are not in the meta-models or setting properties with values of the wrong type. ls-type:: annotation hl-page:: 6 hl-color:: green id:: 64ef0b14-6598-45a6-9d77-8a35c8ca2332 hl-stamp:: 1693387542184
• Semantic errors make a transformation behave in a way that differs from what is expected, i.e., the transformation is semantically incorrect with respect to a specification of the expected behavior ls-type:: annotation hl-page:: 6 hl-color:: blue id:: 64ef0b27-cedf-482a-805d-de86d6469091 hl-stamp:: 1693387561631
• Program repair can be defined by the transformation of an unacceptable behavior of a program into an acceptable one according to a specification [25] ls-type:: annotation hl-page:: 6 hl-color:: blue id:: 64ef0bdc-9a0f-48f4-ab00-c2e327afa47d hl-stamp:: 1693387747203
• patch a sequence of edit operations which modifies a transformation’s source code. ls-type:: annotation hl-page:: 6 hl-color:: green id:: 64ef0c0d-5446-4321-96e8-98c3125528b3 hl-stamp:: 1693387791444
• [:span] ls-type:: annotation hl-page:: 7 hl-color:: green id:: 64ef0c46-e3d4-4971-b3db-3a7f0c98d933 hl-type:: area hl-stamp:: 1693387846594
• We thus use these edit operations to compose the patches to repair faulty ATL transformations. W ls-type:: annotation hl-page:: 6 hl-color:: green id:: 64ef0c76-0862-40a8-9f32-a48abd4eb1b1
• resents related work and Section 8 concludes the pape ls-type:: annotation hl-page:: 4 hl-color:: yellow id:: 64ef0c9e-53be-4e0e-a82a-80b43b9ac49f hl-stamp:: 1693387937866
• This patch therefore modifies the faulty transformation behavior, and the patched transformation produces the expected output model. This three-edit patch is thus considered optimal to repair the transformation with regards to the provided test case ls-type:: annotation hl-page:: 7 hl-color:: yellow id:: 64ef37c8-3018-4922-853d-7b0c16b6e2e9
• Input/output in test cases may reveal the presence of semantic errors, but do not provide a clear indication of what is causing the errors, nor the rules in which they may occur. ls-type:: annotation hl-page:: 7 hl-color:: yellow id:: 64ef385d-2265-4d40-b96b-d14d95818a44
• In such situations, an alternative is to formulate the task as an optimization problem, where the goal is to automatically find optimal solutions in the space of all possible solutions ls-type:: annotation hl-page:: 7 hl-color:: yellow id:: 64ef3895-e5e5-4881-b7b8-877d13105706
• edit operations, a ls-type:: annotation hl-page:: 7 hl-color:: yellow id:: 64ef3970-80ba-4e24-ad30-94c074a0dadb
• test suite, where each test case in that suite consists of one input model and its corresponding output model ls-type:: annotation hl-page:: 7 hl-color:: green id:: 64ef3a3d-2e67-478b-9cd1-8078443b0302 hl-stamp:: 1693399617623
• how these produced output models differ from the reference output models in that test suite. ls-type:: annotation hl-page:: 8 hl-color:: green id:: 64ef3a5a-430c-43dc-a6f3-c4deda8e2a2a
• how does this differ from the reference output model outi? ls-type:: annotation hl-page:: 8 hl-color:: green id:: 64ef3a6b-d9c4-4041-bc55-d1dcde751ce1
• Equation 2 which collects this set of differences. This will allow for the measuring of how many and which errors are fixed by patching a transformation. ls-type:: annotation hl-page:: 8 hl-color:: green id:: 64ef3a86-1f88-4ac7-a988-846720effdcf
• edit operations ls-type:: annotation hl-page:: 8 hl-color:: yellow id:: 64ef3ab5-7feb-41a3-9749-d0faeacccb66 hl-stamp:: 1693399737748
• We will combine i) and ii) in our approach to repeatedly patch transformations and determine the fitness of that patch in terms of the errors fixed ls-type:: annotation hl-page:: 8 hl-color:: yellow id:: 64ef3b30-19a3-4139-97a0-ef63ee319beb
• We discuss EAs as applied to transformation repair, in the context of the multi-objective algorithm introduced by our previous work [41] ls-type:: annotation hl-page:: 8 hl-color:: green id:: 64ef3c52-07b9-454a-a2e8-1707ea2208ac hl-stamp:: 1693400148586
• A second objective is also discussed which prevents the patches from growing unnecessarily large during the search, an issue known as bloating ls-type:: annotation hl-page:: 8 hl-color:: green id:: 64ef4349-1534-401d-9d62-05f939afb166 hl-stamp:: 1693401934239
• efficiently exploring the search space ls-type:: annotation hl-page:: 8 hl-color:: green id:: 64ef43ba-e4f5-4732-a79d-d5047fe455b5
• EAs maintain a population of candidate solutions which undergo an evolution process through several generations. At each generation, some solutions are mutated (i.e., we use an existing solution to create a slightly different solution) and other solutions are bred (i.e., several existing solutions are recombined to create new solutions). ls-type:: annotation hl-page:: 8 hl-color:: blue id:: 64ef43d3-9194-4dc2-9dca-0ff2d6704af0 hl-stamp:: 1693402069846
• The considered mutations here are 1) adding an edit operation, 2) removing an edit operation and or3) modifying an edit operation. ls-type:: annotation hl-page:: 9 hl-color:: yellow id:: 64ef4597-b4cf-4ab9-aa15-3babe527a6f9 hl-stamp:: 1693402523566
• union ls-type:: annotation hl-page:: 10 hl-color:: yellow id:: 64ef4606-e94a-42c9-bbc0-892bd318d1c1
• 392393 ls-type:: annotation hl-page:: 9 hl-color:: yellow id:: 64ef4641-5789-4588-9036-10525b7cae5e
• This ranking is then used to select the better half of the population, and discard the solutions with poor fitness. ls-type:: annotation hl-page:: 9 hl-color:: green id:: 64ef4643-be7d-4872-98ce-c7e6452e93d8 hl-stamp:: 1693402693295
• At each generation, the fitness function would thus favor the patches passing the most test cases, until finding one passing them all. ls-type:: annotation hl-page:: 9 hl-color:: yellow id:: 64ef466f-2a72-421b-b8ee-db588e313e49 hl-stamp:: 1693402766349
• Multi-objective optimization problems introduce the idea that the fitness of candidate solutions may be evaluated based on several objectives, which may conflict with each other. ls-type:: annotation hl-page:: 9 hl-color:: blue id:: 64ef475d-fd48-4b41-8113-044bf6cde999 hl-stamp:: 1693402975312
• Thus, non-dominated solutions are not comparable and can be considered equally good. ls-type:: annotation hl-page:: 10 hl-color:: blue id:: 64ef477e-bc2d-40d5-94ae-787955554967
• NSGA-II [10], a well-known fast multi-objective genetic algorithm, that is suitable to the kind of problem we are solving ls-type:: annotation hl-page:: 10 hl-color:: green id:: 64ef47dd-22cb-42f3-a754-1a50cc9dec87 hl-stamp:: 1693403104543
• a number of iterations or one or more objectives greater than a certain threshold. ls-type:: annotation hl-page:: 10 hl-color:: green id:: 64ef648b-1567-4557-8a6a-3891c7abfc3a
• repairing transformations ls-type:: annotation hl-page:: 10 hl-color:: green id:: 64ef64ac-91a5-419d-8d12-44fefb29cb10
• e more test cases pass, the better the patch. ls-type:: annotation hl-page:: 10 hl-color:: green id:: 64ef64b7-54a7-4bcb-a6a3-a7caff1d00e6
• n the case of ATL transformations, test cases are pairs of input/output models: provided with the input models, a correct transformation should output the expected models. ls-type:: annotation hl-page:: 10 hl-color:: green id:: 64ef64c1-c763-443c-b1c1-695ea5535745
• equivalent to the expected output, then the test case passes ls-type:: annotation hl-page:: 10 hl-color:: yellow id:: 64ef64de-a45a-4a79-b0ef-eb6a379c1e01 hl-stamp:: 1693410527962
• efine the fitness score by considering the differences between the two output models ls-type:: annotation hl-page:: 10 hl-color:: green id:: 64ef6511-65c2-4850-bd99-49f008a47a0b hl-stamp:: 1693410582425
• The idea is that even if a patch does not correct all the errors and does not pass all the tests, a partial solution should lead to less discrepancies between the output models and the expected ones compared to a random solution ls-type:: annotation hl-page:: 10 hl-color:: yellow id:: 64ef6524-42b7-4a96-a0ae-de2f2fe122bc hl-stamp:: 1693410598827
• When such a patch is found, the process stops ls-type:: annotation hl-page:: 11 hl-color:: green id:: 64ef65e2-d5e7-49b0-944f-705ac9d9f34b hl-stamp:: 1693410787928
• ompute the number of differences between these models ls-type:: annotation hl-page:: 11 hl-color:: green id:: 64f03cff-821c-483e-9f6e-9d155f8c1719
• loating is a known issue in EAs where the solutions considered during a run grow in size and become larger than necessary to represent good solutions. ls-type:: annotation hl-page:: 11 hl-color:: green id:: 64f03d8d-77e4-4449-84d7-35cdaf993230
• number of operations in the patch ls-type:: annotation hl-page:: 11 hl-color:: green id:: 64f03db8-3713-4aac-bb23-206437bfc12b
• his objective thus favors patches of small size to avoid generating candidate patches using too many edit operations ls-type:: annotation hl-page:: 11 hl-color:: blue id:: 64f03dc3-14d3-45fa-822a-d8ae2c8d81db
• We then present the third objective in our algorithm, which is designed to preserve diversity in the population. ls-type:: annotation hl-page:: 12 hl-color:: blue id:: 64f03e01-7fcf-40de-a73a-a259054ac1df
• Carefully designing the fitness function is essential and may impact both the approach’s efficiency (time to converge toward an optimal solution) and efficacy(whether it converges towards optimal solution or not). ls-type:: annotation hl-page:: 12 hl-color:: yellow id:: 64f0458e-5577-4dda-abc4-87357be6e9dc
• nd the necessary material to cover all errors and pass all tests is lost to their profits. ls-type:: annotation hl-page:: 12 hl-color:: yellow id:: 64f0489e-d6ed-4822-989f-b34b62f694fe
• ustaining a certain level of diversity within the population, i.e., ensuring that individuals are scattered in different regions of the search space, increases the chances to find good solutions efficiently ls-type:: annotation hl-page:: 12 hl-color:: blue id:: 64f048d6-e725-40a1-954d-4e0ecf041c70
• s a consequence, a lot of candidate solutions (partial or bad) have the same fitness score, thus creating fitness plateaus, i.e., large parts of the fitness landscape where all solutions have the same fitness score even though they are different from one another, and even though some of them are partial solutions [11, 36] ls-type:: annotation hl-page:: 12 hl-color:: blue id:: 64f04926-0b3a-43de-8d7c-098ff286e39b
• We found that partial patches (partial solutions) were quickly discarded in the process due to fitness plateaus. The more errors to correct, the larger the size of the plateaus and the less effective the search for an optimal patch ls-type:: annotation hl-page:: 13 hl-color:: blue id:: 64f04958-b8ce-4e75-8a9b-dfc3500ed164
• Semantic Diversity ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04974-0412-4159-8542-6725fc5362c7
• genotypic ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04ce5-2234-44d1-9b05-b7a2a1c3c0b4
• syntactic ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04ce9-1989-42b2-8f1e-73a0e939df31
• diversity ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04cec-84d8-40af-8519-dcf5d9b173f0
• syntactic diversity would promote patches of variable size and using dissimilar edit operations ls-type:: annotation hl-page:: 13 hl-color:: blue id:: 64f04cfb-fa92-4a31-a784-4529f72fceda
• semantic. This time, it distinguishes individuals based on their behaviors without considering their structure. ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04d24-7758-4430-81cb-cddcc674ff74
• When targeting semantic errors in transformations, maintaining diversity in transformations’ behaviors is highly relevant ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04d40-74b6-442a-86c8-b201a7df370a
• understanding the impact of syntactic diversity on the behavior of a program is quite complex [23]. ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04d62-4666-4847-bfc1-e345f5a7bf9f
• semantic diversity, which is also known to be more efficient to prevent single fitness peak [4, 40] ls-type:: annotation hl-page:: 13 hl-color:: green id:: 64f04d71-b9c1-4608-b91d-6217934b4bb6
• ocial diversity measure is to assess a candidate solution not only by examining the solution alone, but also by considering the solution as a part of the population. ls-type:: annotation hl-page:: 13 hl-color:: blue id:: 64f04d8b-3d69-43d3-804e-0d10f236bc8f
• t aims at assessing the value a candidate patch brings to the entire population ls-type:: annotation hl-page:: 13 hl-color:: yellow id:: 64f04db0-248f-4879-a3dd-46bb07de8af1
• we propose a social diversity measure relying, not on the number of passing test cases, but on the differences between the obtained output models and the expected ones. ls-type:: annotation hl-page:: 13 hl-color:: blue id:: 64f0535a-500e-4f8d-be63-87bdf5c3e879
• . We think that bringing social diversity in our fitness function will help maintain a population of patches addressing different parts of the output models, thus increasing the chances to escape fitness plateaus caused by errors interactions. ls-type:: annotation hl-page:: 14 hl-color:: green id:: 64f053dc-88bf-4325-a08e-76dc2d0d0aa7
• Social diversity is calculated for patches by determining the uniqueness of the errors they address, compared to the rest of the population ls-type:: annotation hl-page:: 14 hl-color:: green id:: 64f05712-d826-4245-bd13-27c03d26239c
• The patches which address a unique set of errors are then assigned a better score. ls-type:: annotation hl-page:: 14 hl-color:: purple id:: 64f0576d-d50e-4035-baa7-64c5fd663a38
• A matrix D is constructed to record which of the errors are addressed by each patch ls-type:: annotation hl-page:: 14 hl-color:: green id:: 64f057ac-2a20-4783-9ef3-54095e6c358d
• Fig. 10 An overview of our model transformation repair approach ls-type:: annotation hl-page:: 15 hl-color:: yellow id:: 64f058c4-5a31-457e-a839-59c169abc72f
• Automatix ls-type:: annotation hl-page:: 15 hl-color:: purple id:: 64f0591c-a0db-4817-a805-331fc32fb094
• RQ1: What is the impact of social diversity on the effectiveness of the approach (i.e., finding a patch correcting all the errors)? ls-type:: annotation hl-page:: 15 hl-color:: green id:: 64f05a62-4c36-4dcd-b23d-09a8077563a5
• RQ2: What is the impact of social diversity on the efficiency of the approach (i.e., the convergence time)? ls-type:: annotation hl-page:: 15 hl-color:: green id:: 64f05a76-dc61-4427-b0e0-93ea56f9484e
• RQ3: What is the impact of social diversity on the type of errors which are corrected? ls-type:: annotation hl-page:: 15 hl-color:: green id:: 64f05a7f-2555-4be5-bfdc-dc312cbfc5e3
• We tested each mutant with the AnAtlyser tool [8], which finds a wide range of syntactic errors(including type errors) in ATL transformations using static analysis. ls-type:: annotation hl-page:: 15 hl-color:: green id:: 64f05ae0-e045-40b6-8a54-1624abd53601
• We reused the approach presented in [42] to merge several mutants with one error to obtain mutants with several errors. ls-type:: annotation hl-page:: 16 hl-color:: green id:: 64f05b07-cfc8-4ee8-8d50-d3dc89dd4795
• the properties of input/output object, ls-type:: annotation hl-page:: 16 hl-color:: green id:: 64f05b58-43d9-4368-aab8-df08fba7170c
• arguments ls-type:: annotation hl-page:: 16 hl-color:: yellow id:: 64f05ba6-5bc4-487d-a894-de0c25999c3f
• We identified nine kinds of elements that could be modified by an atomic edit operation: ls-type:: annotation hl-page:: 16 hl-color:: yellow id:: 64f05bce-71f8-4344-8106-a75fe6e81f97
• four sets with respectively two to five mutants and then merged them in each set to form four faulty transformation mutants with two to five semantic errors. ls-type:: annotation hl-page:: 16 hl-color:: green id:: 64f05c76-57c4-4f49-a8e1-8b6632ba78fe
• n this experiment, we aim at testing social diversity with two configurations separately: as a crowding distance and as an objective ls-type:: annotation hl-page:: 16 hl-color:: red id:: 64f05cf4-cc04-4566-8a82-c5f3b5876eba
• a) without social diversity, b) with social diversity as a crowding distance, and c) with social diversity as an objective. ls-type:: annotation hl-page:: 16 hl-color:: green id:: 64f05e2a-f037-4a12-82e2-8f45a40a4421
• Note that in our earlier work [41], we only considered approach a), and applied it to two transformations. ls-type:: annotation hl-page:: 16 hl-color:: red id:: 64f05e48-f850-442f-992a-6f200e1614d9
• f. ls-type:: annotation hl-page:: 16 hl-color:: red id:: 64f05e6f-a8e7-4852-aa42-ad18ef49d8ad
• To answer RQ1, we compare the effectiveness of each configuration, i.e., the number of times a run can find an optimal patch. ls-type:: annotation hl-page:: 17 hl-color:: green id:: 64f05edb-f531-404a-8138-d003ece29350
• o answer RQ2, we compare the efficiency of each configuration, i.e., the number of generations necessary for a run ls-type:: annotation hl-page:: 17 hl-color:: green id:: 64f05eeb-11e9-4bda-94d3-c9f2efe07e80
• manually ls-type:: annotation hl-page:: 17 hl-color:: yellow id:: 64f05ef8-d659-4183-8af8-620a00d37379
• RQ1: What is the impact of social diversity on the effectiveness of the approach (i.e., finding a patch correcting all the errors)? ls-type:: annotation hl-page:: 17 hl-color:: green id:: 64f061ee-adc3-4487-9b6e-519eb81bc31c
• We can conclude that using social diversity both as crowding distance and as objective improves the correction of larger number of errors at the same time ls-type:: annotation hl-page:: 17 hl-color:: blue id:: 64f06315-9c56-4a03-be89-2d3b747b32fb hl-stamp:: 1693475988638
• RQ2: What is the impact of social diversity on the efficiency of the approach(i.e., the convergence time)? ls-type:: annotation hl-page:: 18 hl-color:: green id:: 64f06324-edb2-4881-a9b0-0072faf5ead0
• Fig. 12 shows that social diversity improves its efficiency. Here again, social diversity as an objective give better results than as a crowding distance ls-type:: annotation hl-page:: 18 hl-color:: green id:: 64f06470-20d6-40cd-808e-fb43a142ebe3
• Thus, we conclude that using a social diversity measure helps the approach find the optimal solutions faster ls-type:: annotation hl-page:: 18 hl-color:: blue id:: 64f0647e-5333-4bb0-9121-3fcaeec6c4bf hl-stamp:: 1693475983051
• RQ3: What is the impact of social diversity on the type of errors which are corrected? ls-type:: annotation hl-page:: 18 hl-color:: green id:: 64f0648d-b445-4f3f-a4b2-99ce49efe735
• (semi-) ls-type:: annotation hl-page:: 20 hl-color:: yellow id:: 64f064e9-9ee1-4abb-993a-99039d4a88f8
• Our results show that this leads to finding optimal solutions faster than our earlier work ls-type:: annotation hl-page:: 20 hl-color:: purple id:: 64f06542-38f6-4f83-89e4-20387cd5c12f
• A limitation of our approach is that we are fixing ATL transformation rules, not the helpers ls-type:: annotation hl-page:: 20 hl-color:: purple id:: 64f06559-cbb7-4176-ae76-0601daef2b53
• These transformations may not be fully representative with real-world transformations in terms of size and complexity. ls-type:: annotation hl-page:: 20 hl-color:: green id:: 64f0658e-2aea-436d-9b44-80f507ced42a
• we believe that the set of four model transformations used in our experiments is sufficiently representative to demonstrate the benefits of our approach. ls-type:: annotation hl-page:: 20 hl-color:: green id:: 64f06593-1517-4feb-8e1e-2c9da033f20a
• We dedicate an objective which gives a score based on the notion of social diversity that we defined on model difference ls-type:: annotation hl-page:: 22 hl-color:: blue id:: 64f065cd-1991-4010-b37b-d9a80ab48fa3
• Our results showed that injecting our social diversity measure in the search process improves both the effectiveness and the efficiency, and enables to find patches for transformations containing up to five errors ls-type:: annotation hl-page:: 22 hl-color:: blue id:: 64f065e1-99cf-4236-9908-aecd78094196