Deep learning by the stochastic gradient descent with stablizer and its application

The project focuses on the mathematical aspects of gradient descent methods used in machine learning. We start with a stabilizer technique for the classical gradient descent, and develop a stochastic gradient descent method with stabilizers for large networks. In addition, we analyze the convergence property of the proposed method, and compare with existing techniques commonly used in the literature.

Yoshinari Takeishi, Kyushu University and Siyang Wang, Umeå University

AI to predict stress levels in caregivers and patients

Sweden and Japan are facing important societal and public health challenges due to demographic ageing. The projects starting point is the Humanitude project, a new training system based on Augmented Reality (AR) and AI which allows carers to practice their skills to calm down patients with dementia by using eye contact, soft manipulation, and voice. The system consists of VR glasses with AR for the trainee and an artificial human model equipped with sensors on the body simulating the patient with dementia.

Read more here

Roy Kurazume, Kyushu University and Oscar Martinez Mozos, Örebro University
In collaboration with Jönköping University

Creating a general framework for developing ambient multimedia solutions

Currently, ambient multimedia solutions are created ad hoc, and the gained insights are rarely shared publicly. The goal of the research collaboration is to create a general framework for developing ambient multimedia solutions on which designers can build on when creating their ambient multimedia. Within the framework, focus is also put on the effects on users’ perceptions.

Tatsuo Nakajima, Waseda University and Bruce Ferwerda, Jönköping University

Bio-signals and human privacy

Biometrics are physical or behavioral characteristics that are unique to each person, and personal authentication for accessing applications and other network resources. Previous studies in this field have screened for diseasesand assessed autonomic nervous system status. However, to what extent can bio-signals be used for biometrics? This project examine this novel problem by using a machine learning method to classify ECGs of monkeys and humans.

Emi Yuda, Tohoku University and Oscar Martinez Mozos, Örebro University

Towards Swedish-Japanese Collaborations in Applied Artificial Intelligence, Innovation and Entrepreneurship

The project’s objective is to be a first stepping stone towards stronger relationships and collaborations between Linköping University, Kyushu University and the uUniversity of Tokyo, with focus on AI research applied to games, and more broadly also on entrepreneurship/innovation.

Fredrik Heintz, Linköping University and Danilo Vasconcellos Vargas, Kyushu University

Autogeneration of HRV data for subarachnoid hemmorhage

Cerebral ischemia and subarachnoid haemorrhage are conditions that can lead to permanent brain injuries if not detected and treated early. Today’s medical practice is to use CT-scans to detect ischemia’s and haemorrhages, which is a costly procedure. However, the neurointensive care units usually collect large amounts of physiological data, used to monitor patients. This data can be used by machine learning algorithms to detect cerebral ischemia and subarachnoid haemorrhage. This project investigate to which degree it is possible to generate heartrate variability data using generative adversarial networks in order to model the variability.

Miroslaw Staron, University of Gothenburg and Emi Yuda, Tohoku University

Perceptions of AI and robotics in Sweden and Japan

The purpose of this project is to develop comparative studies between Sweden and Japan in the intersection of human social norms and AI-supported robots. This includes taking part of the youth- and AI cross-cultural approach and through comparisons of empirical data make a contribution to understanding AI and robot ethics within both European and Asian cultural settings, with an aspiration to deepen this approach.

Read more here

Stefan Larsson, Lund University and Toshie Takahashi, Waseda University
In collaboration with Uppsala University