We introduce a novel framework for studying small-scale primordial perturbations and their cosmological implications. The framework uses a deep reinforcement learning to generate scalar power spectrum profiles that are consistent with current observational constraints. The framework is shown to predict the abundance of primordial black holes and the production of secondary induced gravitational waves. We demonstrate that the set up under consideration is capable of generating predictions that are beyond the traditional model-based approaches.

中文翻译：

---

原初扰动的一般预测及其影响


我们引入了一种新的框架来研究小规模原初扰动及其宇宙学含义。该框架使用深度强化学习来生成与当前观测约束一致的标量功率谱剖面。该框架被证明可以预测原初黑洞的丰度和次级诱导引力波的产生。我们证明，所考虑的设置能够生成超出传统基于模型的方法的预测。