FCAI researchers advance applications of interactive AI

CHI2020, the leading conference of the field of HCI, accepted several research papers from FCAI researchers on topics such as AI for designers, models of how people move, and brain-computer interfaces.

Derived from the paper Optimal Sensor for a Computer Mouse

Derived from the paper Optimal Sensor for a Computer Mouse

Many AI-related research projects conducted at, or in collaboration with, FCAI were accepted for presentation at the CHI conference. The ACM CHI Conference on Human Factors in Computing Systems is the premier venue for publishing research work in human-computer interaction (HCI). This year, CHI was cancelled due to the COVID-19 pandemic, but some events will be, or have already been, held virtually.

Interactive AI is at the core of FCAI research, and it is the theme of one of FCAI’s seven research programs. The goal of the Interactive AI program is to create AI that people can naturally work and solve problems with. This means developing AI that has a better understanding of humans’ goals and abilities, takes initiative more sensitively, and supports people.

FCAI’s main expertise regarding interactive AI lies in Bayesian and reinforcement learning-based models of foundational capabilities in AI-human interaction. The Interactive AI program is coordinated by Associate Professor Antti Oulasvirta at Aalto University.

The following research articles were accepted to CHI2020:

AI helps design better graphical interfaces

Niraj Dayama, Kashyap Todi, Taru Saarelainen and Antti Oulasvirta present an interactive optimization method for the design of graphical layouts. The method helps designers explore possible designs and automatically complete partial designs, which has not been demonstrated before.

The paper presents a new formulation of graphical layout generation for integer programming, a mathematical optimization method that provides guarantees for solution quality. It offers a solution method that permits fast solutions and thereby integration in design tools.

The authors tested the tool with professional designers, who provided positive feedback for its ability to support designers’ exploration of solutions.

Read the research paper at https://kashyaptodi.com/grids/

 

FCAI researchers demonstrate crowdsourcing via brain-computer interfacing

In their research paper, Keith M. Davis, Lauri Kangassalo, Michiel Spapé and Tuukka Ruotsalo introduce brainsourcing; a novel crowdsourcing paradigm that utilizes brain responses recorded via EEG of a group of people. Brainsourcing allows the direct mapping of users’ implicit, neutrally measured reactions within recognition tasks to predict when targets of interest appear.

Read the research paper at https://dl.acm.org/doi/abs/10.1145/3313831.3376288

 

Biomechanically simulated AI agents can be used to test bodily interactions in virtual reality

Noshaba Cheema, Laura A. Frey-Law, Kourosh Naderi, Jaakko Lehtinen, Philipp Slusallek and Perttu Hämäläinen demonstrated in their cross-university collaboration project that biomechanically simulated AI agents can be used to test bodily interactions in, for example, VR reality.

A common problem of mid-air interaction is excessive arm fatigue, known as the “Gorilla arm” effect. To predict and prevent such problems at low cost, researchers investigated user testing of mid-air interaction without real users, utilizing biomechanically simulated AI agents trained using deep Reinforcement Learning (RL). They implemented this in a pointing task and four experimental conditions, demonstrating that the simulated fatigue data matches human fatigue data.

The researchers’ work demonstrates that deep RL combined with Three Compartment Controller provides a viable tool for predicting both interaction movements and user experience in silico, without users. Although real user testing cannot and should not be completely replaced, testing with simulated users has the potential to provide valuable data with less delay and at a lower cost.

Read the research paper at https://dl.acm.org/doi/abs/10.1145/3313831.3376701


Optimal sensor position for a computer mouse

In their research paper, Sunjun Kim, Byungjoo Lee, Thomas van Gemert and Antti Oulasvirta present a method for in-application personalization of input devices, such as the mouse. When deployed on a mouse, the method can improve the accuracy and speed of pointing, even for experienced users.

The authors use machine learning to optimize sensor parameters for high-end users, such as gamers. A virtual model of the input device that can be parametrized is developed. It is tuned by using Bayesian optimization

Read the research paper at https://userinterfaces.aalto.fi/mouse_sensor_position/