Select a calendar:
Filter May Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for May
-
PhD Defense - Elaine Short
Wed, May 03, 2017 @ 02:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Elaine Short
Title: Managing Multi-Party Social Dynamics for Socially Assistive Robotics
Date: 05/03/17
Time: 2-4pm
Location: RTH 406
Committee:
Maja MatariÄ (Chair)
David Traum
Gaurav Sukhatme
Gisele Ragusa (External)
Abstract:
This dissertation presents a domain-independent computational model of moderation of multi-party human-machine interactions that enables a robot or virtual agent to act as a moderator in a group interaction.
A moderator is defined in this work as an agent that regulates social and task outcomes in a goal-oriented social interaction. This model has multiple applications in human-machine interaction: groups of people often require some management or facilitation to ensure smooth and productive interaction, especially when the context is emotionally fraught or the participants do not know each other well. A particularly relevant application domain for moderation is in Socially Assistive Robotics (SAR), where systems are frequently deployed without complex speech understanding or dialogue management, but where group interactions can benefit from a moderator's participation. The evaluation of the model focuses on intergenerational interactions, but the model is applicable to various other SAR domains as well, including group therapy, informal teaching between peers, and social skills therapy.
Moderation is formalized as a decision-making problem, where measures of task performance and positive social interaction in a group are maximized through the behavior of a social moderator. This framework provides a basis for the development of a series of control algorithms for robot moderators to assist groups of people in improving task performance and managing the social dynamics of interactions in diverse domains. Based on reliably-sensed features of the interaction such as task state and voice activity, the moderator takes social actions that can predictably alter task performance and the social dynamics of the interaction. Thus the moderator is able to support human-human interaction in unpredictable, open-ended, real-world contexts.
The model of moderation provides a framework for developing algorithms that enable robots to moderate group interactions without the need for speech recognition; it complements dialogue systems and human-computer interaction, providing conversational agents with additional strategies for managing dynamics of group interaction. Four algorithms are developed based on the model: a basic moderation algorithm, a task-goal-based moderation algorithm, a social-feature-based moderation algorithm, and a combined algorithm that takes into account both task goals and social features. These algorithms are validated in both peer-group interactions and inter-generational family interactions where the moderator supports interactions including members of multiple generations within the same family. The work is intended for short- and long-term deployments of socially assistive robots and virtual agents, and can be applied across assistive domains to facilitate social interactions and improve task performance.
Location: Ronald Tutor Hall of Engineering (RTH) - 406
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Hao Wu
Wed, May 10, 2017 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Hao Wu
Committee:
Kristina Lerman (chair)
Kevin Knight
Florenta Teodoridis (external)
Title: Learning Distributed Representations from Network Data and Human Navigation
Time: May 10 (Wed) 1:00-3:00pm
Room: SAL 322
Abstract:
The increasing growth of network data in online social networks and linked documents on the Web, presents challenges for automatic feature generation for data analysis. We study the problem of learning representations from network data, which is of critical importance for real world applications, including document search, personalized recommendation and role discovery. Most existing approaches do not characterize the surrounding network structure that serves as context for each data point, or they cannot scale well to massive data in real world scenarios. We present novel neural network algorithms that learn distributed representations of network data by exploiting network structure and human navigation. The algorithms embed data into a common low-dimensional continuous vector space, which facilitates predictive tasks, such as classification, relational learning and analogy. Efficient optimization and sampling methods improve the scalability of our algorithms.
First, we propose a neural embedding algorithm to learn distributed representations of generic graphs with global context. To capture the local network structure of each data point, we use random walks to sample nodes in a network neighborhood. Our algorithm is scale-invariant and the learned global representations can be used for similarity measurement of networks. We evaluate our model against state-of-the-art methods on node classification, role discovery and analogy tasks.
Second, we present a neural language model for generating text in networked documents. The model can capture both the local context of word sequences and the semantic influence between linked documents. The approach is based on an intuition that authors are influenced by words in the documents they cite and readers usually read the words in paragraphs by referring to those cited concepts or documents. We show improved performance in document classification and link prediction with our model.
Third, the information of how people navigate the network data online provides clues about missing links between cognitively similar concepts. Learning human navigation can also help characterizing human behavior and improving recommendation. We devise another neural network algorithm that accounts for human navigation patterns to learn better representations of text documents. We present empirical results of our algorithm on online news and movie review data, and show its effectiveness on real world applications.
Location: Henry Salvatori Computer Science Center (SAL) - 322
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
-
PhD Defense - Luenin Barrios
Thu, May 25, 2017 @ 10:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Luenin Barrios
Committee: Wei-Min Shen (chair), Stephan Haas, Aiichiro Nakano.
Title: Simultaneous Center of Mass Estimation and Foot Placement Selection in Complex Planar Terrains for Legged Architectures
Time: Thursday, May 25 at 10am
Room: SAL 213
Abstract:
Center of Mass (CoM) path planning and foot placement selection in complex and rough terrains remains an important goal in the development of motion plans for legged robots. Precise CoM measurements and percipient foot placements are essential in understanding the behavior of a system, for example in gait selection or in extreme locomotion maneuvers. However, operating and maneuvering in difficult terrains has remained a challenging problem due to the diversity of environments and the complex interplay of foot placements and CoM motions. These locomotion maneuvers involve complex forces and movements that make analysis of CoM behavior a challenging task. Nevertheless, understanding CoM dynamics remains pivotal in locomotion planning for both humans and robots. Indeed, the critical element in robot and human motion planning revolves around the ability to accurately measure and describe the CoM. But given the cyclopean space of natural terrains available and the large number of kinematic shapes and sizes possible, the question arises: Is it conceivable to create a generalized framework for CoM construction and estimation with optimal foot placement selection that incorporates the large variety of kinematic architectures and terrains? The work described in this research addresses this issue by presenting a generalized geometric framework from which accurate CoM estimates are produced for the case of bipedal locomotion in complex planar terrains. This framework allows for the simultaneous treatment of CoM estimation and foot placement selection in legged architectures in an efficient and straightforward manner. This is a marked change from current methods for CoM position estimation that rely heavily on expensive and ungainly tools, for example force plates and motion capture video. These render CoM analysis impractical and time consuming and serve as an impediment to understanding locomotion maneuvers in uneven terrains. To tackle these challenges, this work proposes a reliable geometric approach for CoM estimation that delivers accurate CoM behavior in complex planar terrains. The geometric approach depends only on terrain geometry information and essential kinematic data of the moving body. Using this key information in conjunction with an Optimized Geometric Hermite (OGH) curve, a model is developed that produces accurate CoM position and phase space behavior. This phase space behavior is simultaneously optimized during CoM estimation to find candidate foot locations that produce an overall plan with minimum energy. This provides a way to synthesize complex maneuvers in rough terrains and to develop accurate CoM estimates and foot placement plans. Various human case studies were analyzed to validate the effectiveness of the approach. The results show that for natural walking over complex planar terrains, the geometric approach generates accurate CoM path approximations and state space trajectories and is a powerful tool for understanding CoM behavior and foot placements in irregular planar terrains.
Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
