Evolutionary Simulated Annealing for Transfer Learning Optimization in Plant-Species Identification
Gusti Ahmad Fanshuri Alfarisy presents his conference paper
On September 7, 2022 2:00 PM, the School of Digital Science organised an online seminar titled “Evolutionary Simulated Annealing for Transfer Learning Optimization in Plant-Species Identification“, given by PhD student Gusti Ahmad Fanshuri Alfarisy. His paper has been published at the 2022 IEEE 9th International Conference on Computational Intelligence and Virtual Environments for Measurement Systems and Applications (CIVEMSA), available at IEEE Xplore Library here:
Paper: https://ieeexplore.ieee.org/abstract/document/9853679
Code: https://github.com/gusti-alfarisy/evosa
Gusti is supervised by Dr Owais Malik and Dr Wee Hong Ong.
Please find talk abstract and speaker’s biodata below:
Abstract:
The reuse of the pre-trained deep neural network models has been found successful in improving the classification accuracy for the plant species identification task. However, most of these models have a large number of parameters, and layers and take more storage space which makes them difficult to deploy on embedded or mobile devices for real-time classification. Optimization techniques, such as Simulated Annealing (SA), can help to reduce the number of parameters and the size of these models. However, SA can easily get trapped into local optima when dealing with such complex problems. To solve this problem, we propose a new technique, namely Evolutionary Simulated Annealing (EvoSA), which optimizes the process of transfer learning for the plant-species identification task. We incorporate the genetic operators (e.g., mutation and recombination) on SA to avoid the local optima problem. The technique was tested using the MNetV3-Small as a pre-trained model due to its efficiency on mobile for two plant species data sets (MalayaKew and UBD botanical garden). As compared to the standard SA and Bayesian Optimization techniques, the EvoSA provides the least cost value with a similar number of objective evaluations. Moreover, the EvoSA produces approximately 14x and 6x less cost compared to SA for MalayaKew and UBD botanical data sets, respectively. The results show that the EvoSA can generate solutions with higher test accuracy than typical transfer learning with a competitive number of parameters.
Speaker:
Gusti Ahmad Fanshuri Alfarisy is currently pursuing his Ph.D. in the department of computer science, Universiti Brunei Darussalam. He received his B.C.S. and M.C.S. from Brawijaya University in 2014 and 2017. He has been a government lecturer in the informatics department at Kalimantan Institute of Technology, Balikpapan, Indonesia, since 2018. His research interests focus on web mining, deep learning, automated machine learning, and lifelong machine learning
Universiti Brunei Darussalam
Jalan Tungku Link, BE1410
Negara Brunei Darussalam
office.sds@ubd.edu.bn