The progress of deep learning has been remarkable, not only in theory, but also in combination with advances in both computer software and hardware. The outcomes in the areas of language, images, and video will be implemented in various fields of our society in the future. The STAIR Lab is actively engaged in social implementation, and at the same time, we will continue to tackle the next big challenges such as higher-order logical thinking, explainability, long-term memory mechanisms, fact/non-fact distinction, and multimodal memory mechanism. This is the beginning of a new chapter.

The size of deep learning models has increased by orders of magnitude over the past decade. Model training time has correspondingly increased exponentially, but parallel and distributed software technology has made it possible to train on the scale of large computational infrastructures. Training deep learning models is now one of the very large scale applications of computers. Software technology is now required to maximize the performance of increasingly large computational resources.