Karlo Crnek
Researcher ARRS, 2020-
Topic of the thesis
Advanced automatic conversation language generation (CLG)
The doctoral study is focused on the development of multimodal communication interfaces, more specifically on the field of understanding natural language and generating conversational language. The problem of advanced response generation, using artificial intelligence (AI) procedures, which would be more similar to the natural response of people in conversational situations, is a very hot research field. Most research topics are still related to generating more natural language at the level of text (natural language generation NLG), which is otherwise a sub-field of natural language processing NLP. The starting point of scientific research work includes not only the basic aspects of artificial intelligence in the field of UFOs, but also the understanding of verbal and nonverbal content and their inclusion in the automatic generation of responses, and also includes elements of cognitive science. Today, this is one of the biggest challenges in the development of artificial general intelligence AGI. Among the targeted uses of the results of the work can be placed applications for dialogue systems – goal-oriented systems and open-domain systems (goal oriented systems, and open-domain systems)
.
AVAILABLE!
Topic of the thesis
Advanced platform for tracking and analyzing people’s conversational behavior in real time
A research problem is conversational artificial intelligence, which today is one of the very important areas of artificial intelligence development. Comprehensive tracking of people’s conversational behavior as well as nonverbal behavior is necessity for many applications of the future. Tracking nonverbal behavior should focus on assessing the positions and orientations of different parts of the body. In the field of wearable sensors, it is possible to track joint angles and speed of movement, and on this basis to develop more advanced interaction systems with recreation of conversational behavior as actually present in humans, as well as learning deep models to generate such behavior, as well as synchronization of conversational agents. The work is thus focused on the development of a more advanced development platform for tracking and analyzing user conversational behavior.
Dr. Izidor Mlakar
Researcher, 2009-2013
Operational Program for Human Resources Development for the period 2007 – 2013
Topic of the thesis
An expressive model of generating conversational behavior in multimodal user interfaces
The development of multimodal interfaces is a multi-disciplinary process that brings together many fields from psychology, sociology, as well as technology (e.g., artificial intelligence, signal processing and generation, etc.). Psychological and sociological research defines the form, content and “scenarios” (behavior) of interfaces. Technical research defines procedures and algorithms that approximate extracts of psychological and sociological research in the form of hardware / software responses. Nonverbal behavior in multimodal interfaces is understood as the use of non-speech-related information channels. Nonverbal behavior in spontaneous dialogue is a set of mental (conscious and subconscious) processes that act outward synchronously. Correlations between individual processes, however, are carried out at different levels of thought and are generally not defined by general patterns. In addition to one’s own emotional state, well-being, behavioral characteristics, and communicative and non-communicative functions, the formation of nonverbal behavior is influenced by many other factors, e.g. age, social status, attitude towards the topic (whether we are interested or not), attitude towards interlocutors, health status, purpose of the message and interaction, and other empirically measurable and empirically immeasurable factors. The main goal of conversational agents in multimodal interfaces is to provide “natural” and credible behavior. In general, therefore, that the user believes the agent’s statement, even if there is a lack of evidence for its truthfulness (adequacy). The naturalness of behavior is described, by the theory of behavior, in the form of synchronicity of modalities shown by the agent and spontaneity, adequacy and repeatability of responses through which the agent expresses motivation, training, etc. Synchronicity of modality is presented as the connection of modalities that are simultaneously implemented. The modalities through which the virtual agent performs natural behavior are summarized as lip synchronization, emotions and facial expressions, and hand and body gestures.
Vabljeni študentje Fakultete za elektrotehniko, računalništvo in informatiko, ki jih zanima programiranje z vsebino in umetno inteligenco, internet stvari, napredne interakcijske rešitve, avtonomnost sistemov, razvoj naprednih elektronskih sklopov itd.
