This paper introduces SimTrip, a novel self-supervised learning method that efficiently learns meaningful representations from unlabeled data with small batch sizes and reduced computational power. SimTrip is inspired by the conventional triplet network, which typically includes three inputs: an anchor, a positive, and a negative input. However, unlike the triplet network, SimTrip uses three distinct augmented views derived from a single image. The architecture of SimTrip is illustrated in Fig. 1, which includes an encoder, projection, and prediction multilayer perceptrons (MLPs). The model employs a novel loss function, TriLoss, to compute the loss between different branches of SimTrip. The evaluation of SimTrip across two medical image datasets with two proxy tasks: linear evaluation and fine-tuning, along with an ablation study, demonstrates that the model can efficiently extract underlying representations with small batch size and reduced computational power. SimTrip contributes to the field of unsupervised representation learning by presenting a novel paradigm that efficiently circumvents the model collapse problem with the triple-view loss function TriLoss. The model enables the learning of meaningful inherent knowledge with a triple-view architecture, directly maximizing the similarity of one image's triple views instead of relying on negative sample pairs, large batch sizes, or momentum encoders. The model can be combined with other models to implement different proxy tasks, including linear classification, fine-tuning, and transfer learning. The results of the evaluation on linear evaluation and fine-tuning tasks outperform other state-of-the-art methods. The model can efficiently extract underlying representations with a small batch size on lower computational power. The paper also discusses related work in self-supervised learning, including contrastive learning, which aims to draw similar data points closer together while distancing dissimilar ones. The study compares SimTrip with other self-supervised learning methods, such as SimCLR, BYOL, and SimSiam, and demonstrates that SimTrip achieves superior performance in terms of accuracy, precision, recall, and F1-score on both the ALL and LC25000 datasets. The results of the experiments show that SimTrip outperforms other state-of-the-art methods, particularly in scenarios with limited labeled data. The model's efficiency with smaller batch sizes and reduced computational power makes it suitable for practical applications in medical image analysis and other domains where computational resources are limited. The paper concludes that SimTrip is a promising method for unsupervised representation learning, offering a novel approach that can be applied to a wide range of computer vision tasks.This paper introduces SimTrip, a novel self-supervised learning method that efficiently learns meaningful representations from unlabeled data with small batch sizes and reduced computational power. SimTrip is inspired by the conventional triplet network, which typically includes three inputs: an anchor, a positive, and a negative input. However, unlike the triplet network, SimTrip uses three distinct augmented views derived from a single image. The architecture of SimTrip is illustrated in Fig. 1, which includes an encoder, projection, and prediction multilayer perceptrons (MLPs). The model employs a novel loss function, TriLoss, to compute the loss between different branches of SimTrip. The evaluation of SimTrip across two medical image datasets with two proxy tasks: linear evaluation and fine-tuning, along with an ablation study, demonstrates that the model can efficiently extract underlying representations with small batch size and reduced computational power. SimTrip contributes to the field of unsupervised representation learning by presenting a novel paradigm that efficiently circumvents the model collapse problem with the triple-view loss function TriLoss. The model enables the learning of meaningful inherent knowledge with a triple-view architecture, directly maximizing the similarity of one image's triple views instead of relying on negative sample pairs, large batch sizes, or momentum encoders. The model can be combined with other models to implement different proxy tasks, including linear classification, fine-tuning, and transfer learning. The results of the evaluation on linear evaluation and fine-tuning tasks outperform other state-of-the-art methods. The model can efficiently extract underlying representations with a small batch size on lower computational power. The paper also discusses related work in self-supervised learning, including contrastive learning, which aims to draw similar data points closer together while distancing dissimilar ones. The study compares SimTrip with other self-supervised learning methods, such as SimCLR, BYOL, and SimSiam, and demonstrates that SimTrip achieves superior performance in terms of accuracy, precision, recall, and F1-score on both the ALL and LC25000 datasets. The results of the experiments show that SimTrip outperforms other state-of-the-art methods, particularly in scenarios with limited labeled data. The model's efficiency with smaller batch sizes and reduced computational power makes it suitable for practical applications in medical image analysis and other domains where computational resources are limited. The paper concludes that SimTrip is a promising method for unsupervised representation learning, offering a novel approach that can be applied to a wide range of computer vision tasks.