Reusing Extracted Knowledge in Genetic Programming to Solve Complex Texture Image Classification Problems

Iqbal, Muhammad; Xue, Bing; Zhang, Mengjie

doi:10.1007/978-3-319-31750-2_10

Cited by 13 publications

(5 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, Iqbal et al [16], [65] introduced transfer learning in GP-criptor [8] to extract blocks of useful knowledge from simple texture image classification problems, and then reused the extracted knowledge to learn complex problems. The obtained results indicate that reusing the extracted knowledge improves classification accuracy in learning various texture image classification tasks in the same domain.…”

Section: Image Analysis and Transfer Learningmentioning

confidence: 99%

“…Regarding "how to transfer", existing methods [15] often transfer an evolved tree from the source domain to the target domain as a whole tree, which may not be promising, but the evolved tree might be very useful if it is used as subtree to form a new tree/solution for the target problem. Recently, Iqbal et al [16] presented a transfer learning method in GP that successfully utilizes the extracted knowledge at the initialization process as well as the mutation process in GP to learn image classification tasks in the same domain.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Cross-Domain Reuse of Extracted Knowledge in Genetic Programming for Image Classification

Iqbal

Xue

Al-Sahaf

et al. 2017

IEEE Trans. Evol. Computat.

Self Cite

View full text Add to dashboard Cite

Genetic programming (GP) is a well-known evolutionary computation technique, which has been successfully used to solve various problems, such as optimisation, image analysis and classification. Transfer learning is a type of machine learning approach that can be used to solve complex tasks. Transfer learning has been introduced to genetic programming to solve complex Boolean and symbolic regression problems with some promise. However, the use of transfer learning with genetic programming has not been investigated to address complex image classification tasks with noise and rotations, where GP cannot achieve satisfactory performance, but GP with transfer learning may improve the performance. In this paper, we propose a novel approach based on transfer learning and genetic programming to solve complex image classification problems by extracting and reusing blocks of knowledge/information, which are automatically discovered from similar as well as different image classification tasks during the evolutionary process. The proposed approach is evaluated on three texture data sets and three office data sets of image classification benchmarks, and achieves better classification performance than the state-of-the-art image classification algorithm. Further analysis on the evolved solutions/trees shows that the proposed approach with transfer learning can successfully discover and reuse knowledge/information extracted from similar or different problems to improve its performance on complex image classification problems.

show abstract

Section: Image Analysis and Transfer Learningmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Cross-Domain Reuse of Extracted Knowledge in Genetic Programming for Image Classification

Iqbal

Xue

Al-Sahaf

et al. 2017

IEEE Trans. Evol. Computat.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Another kind of knowledge is related to the feature weights as in [15]. In [105,103], the transferred knowledge is represented as code fragments. This approach is related to layered learning and population seeding [163].…”

Section: Transfer Learning For Gp and Symbolic Regressionmentioning

confidence: 99%

Genetic Programming for Symbolic Regression on Incomplete Data

Al-Helali¹

View full text Add to dashboard Cite

Symbolic regression is the process of constructing mathematical expressions that best fit given data sets, where a target variable is expressed in terms of input variables. Unlike traditional regression methods, which optimise the parameters of pre-defined models, symbolic regression learns both the model structure and its parameters simultaneously. Genetic programming (GP) is a biologically-inspired evolutionary algorithm, that automatically generates computer programs to solve a given task. The flexible representation of GP along with its ``white box" nature makes it a dominant method for symbolic regression. Moreover, GP has been successfully employed for different learning tasks such as feature selection and transfer learning. Data incompleteness is a pervasive problem in symbolic regression, and machine learning in general, especially when dealing with real-world data sets. One common approach to handling data missingness is data imputation. Data imputation is the process of estimating missing values based on existing data. Another approach to deal with incomplete data is to build learning algorithms that directly work with missing values. Although a number of methods have been proposed to tackle the data missingness issue in machine learning, most studies focus on classification tasks. Little attention has been paid to symbolic regression on incomplete data. The existing symbolic regression methods are only applicable when the given data set is complete. The overall goal of the thesis is to improve the performance of symbolic regression on incomplete data by using GP for data imputation, instance selection, feature selection, and transfer learning. This thesis develops an imputation method to handle missing values for symbolic regression. The method integrates the instance-based similarity of the k-nearest neighbour method with the feature-based predictability of GP to estimate the missing values. The results show that the proposed method outperforms existing popular imputation methods. This thesis develops an instance selection method for improving imputation for symbolic regression on incomplete data. The proposed method has the ability to simultaneously build imputation and symbolic regression models such that the performance is improved. The results show that involving instance selection with imputation advances the performance of using the imputation alone. High-dimensionality is a serious data challenge, which is even more difficult on incomplete data. To address this problem in symbolic regression tasks, this thesis develops a feature selection method that can select a good set of features directly from incomplete data. The method not only improves the regression accuracy, but also enhances the efficiency of symbolic regression on high-dimensional incomplete data. Another challenging problem is data shortage. This issue is even more challenging when the data is incomplete. To handle this situation, this thesis develops transfer learning methods to improve symbolic regression in domains with incomplete and limited data. These methods utilise two powerful abilities of GP: feature construction and feature selection. The results show the ability of these methods to achieve positive transfer learning from domains with complete data to different (but related) domains with incomplete data. In summary, the thesis develops a range of approaches to improving the effectiveness and efficiency of symbolic regression on incomplete data by developing a number of GP-based methods. The methods are evaluated using different types of data sets considering various missingness and learning scenarios.

show abstract

“…After transferred knowledge has been discovered, a learning algorithm needs to be devised to extract [117] and then transfer the knowledge to the target problem, which corresponds to the issue "How to transfer" [62]. It is also in this step that should be decided how the knowledge should be used [117].…”

Section: Types Of Transfer Learningmentioning

confidence: 99%

“…In initialisation phase, each child of the root node is either generated randomly or selected from code fragments with a probability of 0.5. In the mutation phase, a subtree of an individual is selected and replaced with either a randomly generated tree or a code fragment selected with the same probability [117]. In a later work, Igbal used the same technique but with probabilities µ I and µ M for initialisation and mutation respectively for image classification [116].…”

Section: Haslam Et Al Proposed An Update To Dinh's Work Inmentioning

confidence: 99%

Transfer Optimisation in Genetic Programming for Solving Uncertain Capacitated Arc Routing Problem

Ardeh¹

View full text Add to dashboard Cite

Uncertain Capacitated Arc Routing Problem (UCARP) is a combinatorial optimisation problem with many important real-world applications. Genetic programming (GP) is a powerful machine learning technique that has been successfully used to automatically evolve routing policies for UCARP. Reusability is an open issue in the field of UCARP and in this direction, an open challenge is the case of scenario changes, e.g. change in the number of vehicles and probability distributions of random demands, which typically requires new training procedures to be initiated. Considering the expensive training cost of evolving routing policies for UCARP, a promising strategy is to learn and reuse knowledge from a previous problem-solving process to improve the effectiveness and efficiency of solving a new related problem, i.e. transfer learning. The overall goal of this thesis is to develop novel knowledge transfer algorithms for GP for solving UCARP to handle environment changes more effectively and efficiently. To fulfil this goal, a plethora of machine learning techniques, i.e. surrogate models, feature selection, searching and specialised genetic operators, are utilised in this thesis. First, this thesis explores the effectiveness of the existing transfer optimisation methods for solving UCARP. Accordingly, one of the main directions of this thesis is towards identifying the nature of transferable knowledge, which can impact the quality of knowledge transfer for GP to solve UCARP. For this purpose, a collection of the state-of-the-art transfer optimisation GP algorithms are evaluated for UCARP. After identifying some potential gaps in the literature, a number of preliminary transfer optimisation algorithms are proposed that supplement the literature. To evaluate the algorithms, a large set of knowledge transfer scenarios with various source and target problems were designed based on real-world datasets. According to the results, none of the methods showed significant improvement in the effectiveness of the trained UCARP routing policies. These results revealed the need for more effective transfer optimisation methods specifically designed for UCARP. Furthermore, our investigations revealed that the presence of duplicates in knowledge sources is one of the main challenges for effective transfer optimisation in solving UCARP. Second, we propose approaches to handling the presence of duplicates in the transferred knowledge. The first approach increases population diversity after knowledge transfer to counteract the loss of diversity that is introduced by the presence of duplicates in the transferred knowledge. In the second approach, the duplicates are removed from the transferred knowledge. Then, the transferred knowledge is utilised to create a diverse initial GP population of high-quality individuals. Both approaches are investigated through detailed experimental studies. The results indicate that, while the first approach did not perform better than GP with knowledge transfer, the second can improve the effectiveness of training routing policies with GP significantly. Third, this thesis proposes a novel algorithm that transfers the phenotypic characteristics of the routing policies for solving the source problem. In the new algorithm, the most fit and unique source routing policies are utilised for initialising GP for solving the target problem. Then, a tabu list is placed on the source routing policies and the GP process is prohibited from recreating any of the source routing policies. The motivation for this approach is that, due to the existence of similarity between the source and target problems, source routing policies are unlikely to have a good performance for the target problem. Our experimental studies confirmed that by prohibiting GP from recreating source policies, and the computational resources will be spent on searching and evaluating new regions of the search space, which can lead to discovering better solutions. Fourth, this thesis proposes a novel knowledge transfer algorithm based on the idea of maintaining the transferred knowledge as an auxiliary population. In this approach, first, the best individuals of the duplicate-free knowledge source are used to initialise GP. Additionally, these transferred individuals are also maintained as an auxiliary population and are evolved alongside the main population. To save the computational cost, the auxiliary population is evolved with a surrogate method. Additionally, an elaborate knowledge exchange mechanism between the two populations is devised that emphasises transferring high-quality and unique individuals, the transfer of which can improve the diversity of the receiving population. This allows GP to overcome the problem of losing its population diversity during the evolutionary process. Our detailed experimental results confirmed the superior performance of the proposed algorithm and confirmed that the proposed method improved the phenotypic diversity of GP population.

show abstract

Reusing Extracted Knowledge in Genetic Programming to Solve Complex Texture Image Classification Problems

Cited by 13 publications

References 15 publications

Cross-Domain Reuse of Extracted Knowledge in Genetic Programming for Image Classification

Cross-Domain Reuse of Extracted Knowledge in Genetic Programming for Image Classification

Genetic Programming for Symbolic Regression on Incomplete Data

Transfer Optimisation in Genetic Programming for Solving Uncertain Capacitated Arc Routing Problem

Contact Info

Product

Resources

About