Deep learning models have already achieved significant results and surpassed other sophisticated models based on features made by experts in various tasks; image processing, sound processing, and natural language processing. They are considered to automatically learn and extract concepts such as “cat face” and “human body” from given big dataset. However, what are indeed called concepts, and how they are extracted? This manuscript provides a rough history of neural networks preceding deep learning, explanation of concepts learned by deep learning models, and then future perspective of deep learning study with (cognitive) neuroscience.