تلفیق روش‌های یادگیری شات محدود جهت بهبود عملکرد طبقه‌بندی تصاویر با مجموعه داده‌های کم

ﺑﺎوﺟﻮد ﭘﯿﺸﺮﻓﺖ ﭼﺸﻢﮔﯿﺮ روشﻫﺎی ﻫﻮش ﻣﺼﻨﻮﻋﯽ در ﺳﺎلﻫﺎی اﺧﯿﺮ، ﻫﻤﭽﻨﺎن ﻧﯿﺎز ﺑﻪ دادهﻫﺎی ﻓﺮاوان ﺑﺮای ﯾﺎدﮔﯿﺮی اﯾﻦ روشﻫﺎ ﻣﺸﺎﻫﺪه ﻣﯽﺷﻮد. ﺑﻪﻣﻨﻈﻮر رﻓﻊ اﯾﻦ ﻧﯿﺎز، اﻟﮕﻮی ﺟﺪﯾﺪ ﯾﺎدﮔﯿﺮی ﻣﺎﺷﯿﻦ ﺑﻪ ﻧﺎم ﯾﺎدﮔﯿﺮی ﺷﺎت ﻣﺤﺪود ﭘﯿﺸﻨﻬﺎدﺷﺪه اﺳﺖ. ﯾﮑﯽ ازروشﻫﺎی ﻣﻄﺮح در اﯾﻦ ﺣﻮزه روﯾﮑﺮد ﺷﺒﮑﻪﻫﺎی ﻧﻤﻮﻧﻪ اوﻟﯿﻪ اﺳﺖ ﮐﻪ درواﻗﻊ ﺗﺮﮐﯿﺒﯽ از روشﻫﺎی ﯾﺎدﮔﯿﺮی ﻣﺘﺮﯾﮏ و ﻓﺮا ﯾﺎدﮔﯿﺮی اﺳﺖ. در اﯾﻦ ﺷﺒﮑﻪﻫﺎ، ﻃﺒﻘﻪﺑﻨﺪ ﺳﻌﯽ ﻣﯽﮐﻨﺪ ﺗﺎ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﺗﻨﻬﺎ ﺗﻌﺪاد ﮐﻤﯽ از ﻧﻤﻮﻧﻪﻫﺎی ﻫﺮ ﮐﻠﺎس ﺟﺪﯾﺪ، ﻧﺴﺒﺖ ﺑﻪ اﯾﻦ ﮐﻠﺎسﻫﺎ ﺗﻌﻤﯿﻢ ﯾﺎﺑﺪ. در اﯾﻦ ﭘﮋوﻫﺶ ﺳﻌﯽ ﺑﺮ آن ﺷﺪ ﺗﺎ ﺷﮑﻞ ﺗﻐﯿﯿﺮﯾﺎﻓﺘﻪای از ﺷﺒﮑﻪﻫﺎی ﻧﻤﻮﻧﻪ اوﻟﯿﻪ ﺑﺮای ﺣﻞ ﻣﺴﺎﻟﻪ ﻃﺒﻘﻪﺑﻨﺪی ﺷﺎت ﻣﺤﺪود ﭘﯿﺸﻨﻬﺎد ﺷﻮد. در اﺑﺘﺪا ﺑﻪﻣﻨﻈﻮر ﺑﻬﺒﻮد ﻋﻤﻠﮑﺮد در ﺷﺒﮑﻪﻫﺎی ﻧﻤﻮﻧﻪ اوﻟﯿﻪ، ﺑﻪﺟﺎی ﻓﺎﺻﻠﻪ اﻗﻠﯿﺪﺳﯽ از ﻓﺎﺻﻠﻪ ﻣﺎﻫﺎﻟﺎﻧﻮﺑﯿﺲ ﺑﺮای اﻧﺪازهﮔﯿﺮی ﻓﺎﺻﻠﻪ ﺑﯿﻦ ﻧﻤﻮﻧﻪﻫﺎ اﺳﺘﻔﺎده ﺷﺪ. اﯾﻦ ﮐﺎر ﻣﻮﺟﺐ ﺑﻬﺒﻮد ﻋﻤﻠﮑﺮد اﯾﻦ ﺷﺒﮑﻪﻫﺎ در ﻃﺒﻘﻪﺑﻨﺪی ﺗﺼﺎوﯾﺮ omniglot و miniimagenet ﺷﺪ ﺑﻪﻃﻮریﮐﻪ ﺷﺒﮑﻪ ﭘﯿﺸﻨﻬﺎدی ﺗﻮاﻧﺴﺖ ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﻪ دﻗﺖﻫﺎی 99/1% و 68/5% ﺑﺮ روی اﯾﻦ دو ﻣﺠﻤﻮﻋﻪ داده دﺳﺖ ﯾﺎﺑﺪ. در ﺑﺨﺶ ﺑﻌﺪی روﯾﮑﺮدی ﮐﻠﯽ ﻣﻌﺮﻓﯽ ﺷﺪ ﮐﻪ ﻣﯽﺗﻮاﻧﺪ ﻣﻌﻤﺎری ﺷﺒﮑﻪﻫﺎی ﻋﺼﺒﯽ ﮐﺎﻧﻮﻟﻮﺷﻨﯽ را ﺑﺎ اﺳﺘﻔﺎده از اﻟﮕﻮرﯾﺘﻢ ژﻧﺘﯿﮏ ﺑﻪﻃﻮر ﺧﻮدﮐﺎر ﺑﻬﺒﻮد ﺑﺒﺨﺸﺪ. در اﯾﻦ ﭘﮋوﻫﺶ از اﯾﻦ روﯾﮑﺮد ﺑﻪﻃﻮر ﺧﺎص ﺑﺮ روی ﻣﺠﻤﻮﻋﻪ دادهﻫﺎی omniglot ﺑﺎ ﻣﻌﻤﺎری اوﻟﯿﻪ ﭘﯿﺸﻨﻬﺎدی در ﺷﺒﮑﻪﻫﺎی ﻧﻤﻮﻧﻪ اوﻟﯿﻪ اﺳﺘﻔﺎدهﺷﺪه اﺳﺖ. درﻧﻬﺎﯾﺖ ﺑﺎ اﺳﺘﻔﺎده از اﯾﻦ روﯾﮑﺮد و ﺟﺎﯾﮕﺰﯾﻨﯽ ﻣﻌﻤﺎری ﭘﯿﺸﻨﻬﺎدی آن ﺑﺎ ﻣﻌﻤﺎری اﺻﻠﯽ ﺷﺒﮑﻪ ﻧﻤﻮﻧﻪ اوﻟﯿﻪ دﻗﺖ ﺷﺒﮑﻪ ﺑﻬﺒﻮدﯾﺎﻓﺘﻪ و ﺗﻮاﻧﺴﺘﻪ ﺑﻪدﻗﺖ 99/5% دﺳﺖ ﯾﺎﺑﺪ.

integration of few-shot learning methods to improve image classification performance with small data sets

despite the significant advancement of artificial intelligence methods in recent years, there is still a need for a lot of data to learn these methods. to meet this need, a new machine learning model called few-shot learning has been proposed. one of the methods in this field is the prototypical networks approach, which is actually a combination of metric learning and meta-learning methods. in these networks, the classifier tries to generalize to these classes according to only a small number of samples of each new class. in this study, an attempt was made to propose a modified form of the prototype networks to solve the finite shot classification problem. initially, in order to improve the performance of the prototype networks, instead of the euclidean distance, the mahalanobis distance was used to measure the distance between the samples. this improved the performance of these networks in classifying omniglot and miniimagenet images so that the proposed network was able to achieve 99.1% and 68.5% accuracy on these two datasets, respectively. the next section introduces a general approach that can automatically improve the architecture of convolutional neural networks using a genetic algorithm. in this research, this approach has been used specifically on omniglot datasets with the proposed primary architecture in prototype networks. finally, by using this approach and replacing the proposed architecture with the main architecture of the prototype network, the network accuracy was improved and was able to achieve 99.5% accuracy.