Deep neural networks learning (DNNs) becomes extremely challenging in high-noise environments. Currently, popular learning strategies under noisy label conditions can be roughly classified into four categories: Robust Architecture, Robust Regularization, Robust Loss Design, and Sample Selection. Building upon our theoretical framework, we integrate two major strategies: sample selection and robust loss functions. The robust loss function can mitigate the negative impact of noise on DNNs, while the sample selection strategy removes some noisy samples. These two strategies complement each other. We integrate them into a new framework to enhance the performance of DNNs in high-noise environments. Additionally, we propose two new strategies: 1) Limitation of forget rate, and 2) Replacement of sample selection loss functions, effectively addressing the shortcomings of sample selection and robust loss functions. We conducted experiments on datasets such as CIFAR-10, achieving significant accuracy improvements under high-noise conditions (0.8 noise rate). We have published the code at: https://github.com/fang-sen-wei/Collaborative-Learning-for-Noisy-Labels .