
AI Blueprints
By :

In conclusion, we now look ahead at what might be coming in the next few years. It is clear DL is here to stay, given its dramatic and broad successes in many application domains. In the near future, expect DL to be applied to even more applications, particularly in healthcare and medicine. There is also significant research interest in connecting data of different modalities together with DL. For example, building models that can create text descriptions of images, or creating images from text descriptions. This kind of research aims to put more logic and structure in DL architectures, so it's more sophisticated than simple "input/output" pairs (for example, input = image, output = "cat"). For example, Zhu and Jiang recently reported success in training a system to understand relations like the person is next to the horse just by looking at a photo (Deep Structured Learning for Visual Relationship Detection, Zhu, Yaohui, and Shuqiang Jiang, Proceedings of the Thirty-Second...