Hello friends, I have a project that I used finetuning

at first, and then the vision transformer method for implementation. The accuracy obtained is 100%, but unfortunately it is slightly overfit. As far as I know, when finetuning is used In your opinion, apart from changing some parameters such as learning rate, train brach size, eval batch size and epoch, or using the grid search method, what can I do to improve the result?

If you used a pre-trained model for fine-tuning, try using a different base model for transfer learning. Different architectures may have different generalization capabilities. If applicable, you may also use techniques like k-fold cross-validation to get a better estimate of your model's performance and reduce overfitting. Otherwise, you can use some of the regularization techniques like Dropout or L1/L2 Regularization. In case of Dropout, apply dropout layers to the model architecture. This randomly sets a fraction of input units to 0 during training, which can help prevent overfitting. For L1/L2 Regularization, you can add L1 or L2 regularization to the model's weights. This adds a penalty term to the loss function, which discourages the model from assigning too much importance to any one feature.

You can try each of these techniques and evaluate its impact on the overfitting issue. You can also use cross-validation to get a more robust estimate of your model's performance.

Thank you for answering my question🙏🏻 Model Transformer is a deep learning model used to process structured data such as linguistic and image data. This model is able to recognize complex patterns in the data and improve the performance of the model in different tasks by using transformer layers that include the main blocks of this model. On the other hand, transformer learning is a learning method that is used to train natural language processing models. In this method, the input data is considered as linguistic sentences, and by using transformer layers, the model automatically improves its performance in recognizing linguistic patterns from data without the need to define specific rules and algorithms. obtains In the field of machine vision, the Vision Transformer model is an example of a model transformer used for image processing. By using transformer layers, this model is able to recognize complex features in images and improve the performance of the model in various tasks.