No abstract

No abstract

Abstract. ObjectiveWe aim to develop and evaluate an affective Brain-computer music interface (aBCMI) modulating the affective states of its users. ApproachAn aBCMI is constructed to detect a user's current affective state and attempt to modulate it in order to achieve specific objectives (for example, making the user calmer or happier) by playing music which is generated according to a specific affective target by an algorithm music composition system and a case-based reasoning system. The system is trained and tested in a longitudinal study on a population of 8 healthy participants, with each participant returning for multiple sessions. Main ResultsThe final online aBCMI is able to detect its users current affective states with classification accuracies of up to 65% (3 class, p< 0.01) and modulate its user's affective states significantly above chance level (p< 0.05). SignificanceOur system represents one of the first demonstrations of an online affective braincomputer music interface that is able to accurately detect and respond to user's affective states. Possible applications include use in music therapy and entertainment.

A common vision from science fiction is that robots will one day inhabit our physical spaces, sense the world as we do, assist our physical labours, and communicate with us through natural language. Here we study how to design artificial agents that can interact naturally with humans using the simplification of a virtual environment. This setting nevertheless integrates a number of the central challenges of artificial intelligence (AI) research: complex visual perception and goal-directed physical control, grounded language comprehension and production, and multi-agent social interaction. To build agents that can robustly interact with humans, we would ideally train them while they interact with humans. However, this is presently impractical. Therefore, we approximate the role of the human with another learned agent, and use ideas from inverse reinforcement learning to reduce the disparities between human-human and agent-agent interactive behaviour. Rigorously evaluating our agents poses a great challenge, so we develop a variety of behavioural tests, including evaluation by humans who watch videos of agents or interact directly with them. These evaluations convincingly demonstrate that interactive training and auxiliary losses improve agent behaviour beyond what is achieved by supervised learning of actions alone. Further, we demonstrate that agent capabilities generalise beyond literal experiences in the dataset. Finally, we train evaluation models whose ratings of agents agree well with human judgement, thus permitting the evaluation of new agent models without additional effort. Taken together, our results in this virtual environment provide evidence that large-scale human behavioural imitation is a promising tool to create intelligent, interactive agents, and the challenge of reliably evaluating such agents is possible to surmount. See videos for an overview of the manuscript, training time-lapse, and human-agent interactions.

It is widely acknowledged that music can communicate and induce a wide range of emotions in the listener. However, music is a highly-complex audio signal composed of a wide range of complex time- and frequency-varying components. Additionally, music-induced emotions are known to differ greatly between listeners. Therefore, it is not immediately clear what emotions will be induced in a given individual by a piece of music. We attempt to predict the music-induced emotional response in a listener by measuring the activity in the listeners electroencephalogram (EEG). We combine these measures with acoustic descriptors of the music, an approach that allows us to consider music as a complex set of time-varying acoustic features, independently of any specific music theory. Regression models are found which allow us to predict the music-induced emotions of our participants with a correlation between the actual and predicted responses of up to r=0.234,p<0.001. This regression fit suggests that over 20% of the variance of the participant's music induced emotions can be predicted by their neural activity and the properties of the music. Given the large amount of noise, non-stationarity, and non-linearity in both EEG and music, this is an encouraging result. Additionally, the combination of measures of brain activity and acoustic features describing the music played to our participants allows us to predict music-induced emotions with significantly higher accuracies than either feature type alone (p<0.01).