Deep Learning is an emerging topic in data analysis. Deep Networks are special kinds of neural networks that try to mimic biological neural networks in hardware or software. Though neural networks is a very old topic in computer science, in recent years surprising advances have been made in this area. This is evidenced by several awards in very diverse learning competitions including speech, handwriting recognition, speech translation, vision, and game playing. Some of these advances are due to better learning algorithms and/or very clever network architectures, others are simply due to much better, highly parallel hardware. More recently, companies like Google and Facebook have started making major investments in the area of deep learning, e.g. in 2014, Google acquired Deep Mind for approximately 500 million pounds; and Facebook founded a deep learning lab in Paris. In addition, some of the fascinating results of deep learning created big waves in the press, e.g. Google Inceptionism and Google Deep Dream.
In this seminar we will investigate what is behind deep learning. We will loook at fundamental techniques, network architectures, learning algorithms, frameworks, and applications.
The seminar will be classical in the sense that students are expected to give presentations. In addition, we expect students to prepare demos.