Pozývame Vás na dva doktorandské semináre, ktoré sa uskutočnia v utorok 28.05.2024 na ÚI SAV v zasadačke č.102:
Program: Jay Kejriwal, MA (odd. M. Rusko, Š. Beňuš)
Alignment in human-human and human-machine spoken interaction
Abstract:
Entrainment is the tendency of a speaker to adjust some properties of a speaker’s features to match the interlocutor’s characteristics. It affects various linguistic dimensions and correlates with positive social attributes, allowing for rich human-machine interaction (HMI) applications. This thesis primarily focuses on improving spoken dialogue systems (SDS) by gaining an in-depth understanding of entrainment behavior in Human-Human interaction (HHI) and then applying this understanding to improve Human-Machine interaction (HMI). The thesis has three primary goals. First, a deep learning framework is presented to develop an entrainment detection system that identifies entrainment in four linguistic dimensions. The second primary goal is to understand cross-linguistic differences in entrainment across various linguistic dimensions in different languages. Lastly, the relationship between entrainment and non-verbal social cues and emotions was examined using different datasets. To sum up, the thesis contributed to a better understanding of entrainment in HHI and HMI that can be useful in developing entrainment functionality in existing SDS.
Program: Ing. Sepideh Hassankhani Dolatabadi (odd. J. Zelenka, I. Budinská)
APPLYING MACHINE LEARNING ALGORITHMS IN PREDICTIVE MAINTENANCE (ANOMALY PREDICTION)
Abstract:
In today’s manufacturing landscape, optimizing operational efficiency and ensuring machinery’s
uninterrupted performance are critical objectives. This research embarks on a journey to develop an intelligent Predictive Maintenance System (PMS) for a major low-grade hematite iron ore processing plant. Focused on the principles of Industry 4.0, our work leverages real-world data collected from sensors and repair logs, serving as a foundational dataset for comprehensive analysis and predictive modeling. The study begins by addressing the challenge of missing data, employing sophisticated imputation algorithms to bridge temporal data gaps effectively. Subsequently, our work delves into predicting potential machinery malfunctions based on advanced maintenance strategies guided by machine learning algorithms. As a central theme, our methodology involves data collection, preprocessing, feature engineering, labeling, and model training. We validate and verify the proposed methods through rigorous model evaluation and hyperparameter tuning. Additionally, our research paves the way for the development of a novel approach for developing a maintenance dashboard in the context of predictive maintenance. This dashboard is designed to deploy, monitor, and manage alerts and actions efficiently, creating a seamless feedback loop for continuous improvement.
