【Arithmer Seminar #02】An invited lecture by Dr. Masato Taki, Senior Research Scientist, RIKEN Interdisciplinary Theoretical and Mathematical Sciences Program, was held at our Roppongi headquarters office on Thursday, January 17.
“Recent developments in application of deep learning and its mathematical understanding.”
■Outline of the lecture：
This lecture aimed to review the recent progress in theoretical and technical studies on deep learning.
Generalization in deep learning, the ability to adapt to unseen circumstances or new data, is yet to be explored compared to other machine learning models. Some interesting approaches from recent studies were examined in the lecture to understand the specificity of generalization in deep learning, and how useful these studies are for practical researches. These studies also show that existing understandings on pre-training approaches or basic technologies like Batch normalization have been wrongly interpreted.
In the area of application, progress on architecture design that has been made in the past couple of years in relation to image-related tasks was also mentioned in the lecture as well as hands-on technologies of learning tricks including data augmentation.
Dr. Masato Taki
Senior Research Scientist, RIKEN Interdisciplinary Theoretical and Mathematical Sciences Program