Background and goals
This work package is based on the continuation of two main existing research tracks in AI research at Arcada. AI is truly interdisciplinary and methods that have been developed for one domain can be used in many other domains (such as well-being). The objective for this WP is to further develop core AI methods that have already a track record in Arcada. We will have two tracks in this project and they are called i) Environmentally responsible AI methods for different applications on the edge devices and ii) Trustworthy identification for privacy enabled applications. In many domains (such as also health care and well-being) efficient algorithms in the handheld devices and privacy enabled applications are needed for reliable, efficient and trustworthy AI enables services. Thus we need a work package that looks beyond the scope of well-being.
Objectives and benefits
Global warming and environmental pollution became the main challenges of XXI century. Energy production plays a significant part in both of them . Data centers became significant global energy consumers ; their share of energy demand is destined to only grow in the near future as larger datasets and more sophisticated algorithms come into existence – consuming up to 8% of world energy demand by 2030. We as machine learning researchers, data analysts, and future industrial developers should understand the need for energy efficient computations. The energy cost of our methods ought to be considered, and purposely wasteful approaches avoided. This is easiest done by having alternative energy-conserving methods available; finding them is the goal of the research project.
The widespread of digitally controlled devices places higher trust and stricter time constraints into our digital identities. While the existing applications allow for a significant time window up to few days for identity verification and decision roll-back (like online loan application), the upcoming use cases demand real-time verification and the decisions are irreversible. Examples range from trivial like digitally controlled cars or drones (that place legal responsibilities on their operator), to specific ones in remote surgery, industrial remotely operated machines, business applications and etc. The authors acknowledge a possible application for improved safety of the remotely operated military machines.
Edge devices are inherently energy aware as they mostly work from battery power, and their processors operate at lower frequencies and voltages providing much higher computational efficiency per unit of electrical power spent. The second goal of this project would be research on large-scale and energy-efficient machine learning algorithms that are feasible to run directly on edge computing devices, enabling new services and business opportunities that are truly eco-friendly.
Arcada will develop advanced analytical methods for accurate identification based on multi-task deep networks that learn task-independent user characteristics and construct a “signature” for each user derived from this biometric data. More use cases are considered by Arcada, such as counterfeit signature detection leveraging recently published large datasets, or continuous mobile phone user verification based on accelerometer recordings. The data, underlying methodology and processing power are all available today, but there is a lack of applied research that is needed in order to fully enhance the business opportunities with a trustworthy eco-system for our digital economy.
We aim to develop user-centered approaches to health and welfare technology, using AI methods and technical solutions to health promotion. We will explore the human aspects on the possibilities of using AI methods and technical solutions in health care services.
This work package is based on the continuation of two main existing research tracks in AI research at Arcada. AI is truly interdisciplinary and methods that have been developed for one domain can be used in many other domains (such as well-being). We will have two tracks in this project and they are called i) Environmentally responsible AI methods for different applications on the edge devices and ii) Trustworthy identification for privacy enabled applications.