Currently, the majority of whole-slide image categorization algorithms heavily depend on manually adding annotations at the pixel level. This process is both complicated and laborious, requiring annotators with specific knowledge in the subject matter. In order to overcome this issue, we suggest the utilization of self-supervised learning as well as multiple instances learning techniques for the management of huge whole slide imaging (WSI) datasets, where the only available labels are the reported diagnoses. In this study, we employed various machine learning techniques, namely K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Convolutional Neural Network (CNN) with the specific architecture of Alexnet. Our findings indicate that the accuracy of CNN and Alexnet surpasses that of KNN and SVM. Furthermore, the characteristics learned by CNN and Alexnet demonstrate superior suitability for classification tasks. Presently, the majority of classification models for entire slide images depend on human pixel-level annotations, necessitating the involvement of domain specialists who possess specialized knowledge and expertise. This annotation process is intricate and demands a significant amount of time and effort. In order to address this issue, we suggest integrating self-supervised learning using various instance learning as a means of effectively handling extensive whole slide imaging (WSI) datasets, utilizing solely the provided diagnosis as labels. The classification problem in WSI poses a significant difficulty in acquiring effective picture representation, with self-supervised learning demonstrating considerable promise in this regard. In this research endeavor, we intend to employ a self-supervised learning network known as Bootstrap. The pre-trained network known as Your Own Latent has the capability to be developed with unlabeled input and acquire profound domain-specific properties. The suggested framework was assessed on a large-scale uterine cervical dataset consisting of 3,063 complete slide pictures. The findings of our study indicate that the integration of a self-supervised learning model with a multiple occurrence learning model has demonstrated the ability to achieve and surpass the success of previous methodologies.