HARK Side of Deep Learning -- From Grad Student Descent to Automated Machine Learning
Authors:
Oguzhan Gencoglu,
Mark van Gils,
Esin Guldogan,
Chamin Morikawa,
Mehmet Süzen,
Mathias Gruber,
Jussi Leinonen,
Heikki Huttunen
Abstract:
Recent advancements in machine learning research, i.e., deep learning, introduced methods that excel conventional algorithms as well as humans in several complex tasks, ranging from detection of objects in images and speech recognition to playing difficult strategic games. However, the current methodology of machine learning research and consequently, implementations of the real-world applications…
▽ More
Recent advancements in machine learning research, i.e., deep learning, introduced methods that excel conventional algorithms as well as humans in several complex tasks, ranging from detection of objects in images and speech recognition to playing difficult strategic games. However, the current methodology of machine learning research and consequently, implementations of the real-world applications of such algorithms, seems to have a recurring HARKing (Hypothesizing After the Results are Known) issue. In this work, we elaborate on the algorithmic, economic and social reasons and consequences of this phenomenon. We present examples from current common practices of conducting machine learning research (e.g. avoidance of reporting negative results) and failure of generalization ability of the proposed algorithms and datasets in actual real-life usage. Furthermore, a potential future trajectory of machine learning research and development from the perspective of accountable, unbiased, ethical and privacy-aware algorithmic decision making is discussed. We would like to emphasize that with this discussion we neither claim to provide an exhaustive argumentation nor blame any specific institution or individual on the raised issues. This is simply a discussion put forth by us, insiders of the machine learning field, reflecting on us.
△ Less
Submitted 16 April, 2019;
originally announced April 2019.
Design and Evaluation of Vision-based Head and Face Tracking Interfaces for Assistive Input
Authors:
Chamin Morikawa,
Michael J. Lyons
Abstract:
Interaction methods based on computer-vision hold the potential to become the next powerful technology to support breakthroughs in the field of human-computer interaction. Non-invasive vision-based techniques permit unconventional interaction methods to be considered, including use of movements of the face and head for intentional gestural control of computer systems. Facial gesture interfaces ope…
▽ More
Interaction methods based on computer-vision hold the potential to become the next powerful technology to support breakthroughs in the field of human-computer interaction. Non-invasive vision-based techniques permit unconventional interaction methods to be considered, including use of movements of the face and head for intentional gestural control of computer systems. Facial gesture interfaces open new possibilities for assistive input technologies. This chapter gives an overview of research aimed at develo** vision-based head and face-tracking interfaces. This work has important implications for future assistive input devices. To illustrate this concretely we describe work from our own research in which we developed two vision-based facial feature tracking algorithms for human computer interaction and assistive input. Evaluation forms a critical component of this research and we provide examples of new quantitative evaluation tasks as well as the use of model real-world applications for the qualitative evaluation of new interaction styles.
△ Less
Submitted 25 July, 2017; v1 submitted 25 July, 2017;
originally announced July 2017.
