Automated systems are becoming increasingly more prominent in our lives. As this continues to happen, we see more interactions between humans and machines in a wider variety of contexts. This raises questions about the extent to which interactions between humans and machines will translate to interactions between humans, and how robots can be designed aesthetically to facilitate those interactions. I was interested in exploring how a robot’s physical appearance might influence people’s attitudes towards human-robot interaction, and particularly how much people would trust a given robot with certain tasks. Established literature on the subject points to many potential determinants of such attitudes, but I was primarily interested in three: anthropomorphism of a robot, gender presentation of robot, and racial presentation of the robot. To explore these determinants, I conducted a series of experiments in which I presented participants with videos of robots with differing levels of human resemblance, different gender presentations, and different racial presentations in online surveys. Each video would comprise a robot assuming the fictional role of a household caretaker giving a brief speech to the viewer about its capabilities. Participants then answered a series of questions about how much they would trust the robot in their homes performing such tasks, and they would then rate their perceptions of the robot along a variety of criteria such as how likable or how intelligent they perceived the robot to be, all on a series of Likert scales. I conducted two sets of these studies: one in which the robots assigned to participants differed based on level of anthropomorphism and male or female gender presentation, and another in which the assigned robots differed based on level of anthropomorphism and white or Black racial presentation. The results of these experiments broadly indicated that anthropomorphism at extreme levels significantly influenced trust in the robots as well as perceptions of them, while gender and racial variations did not. The data also demonstrated strong correlations between the trust metrics and perceptual appraisals that I used, potentially suggesting that such metrics are reliable indicators of attitudes towards robots. This dissertation, The Impact of Anthropomorphism on Trust in Human-Robot Interaction, is submitted by Umesh Krishnamurthy in 2021 in partial fulfillment of the degree Doctor of Philosophy in Cognitive and Information Sciences at the University of California, Merced under the guidance of dissertation committee chair Paul Maglio.

Ordinary differential equations (ODE) are used extensively in science and engineering to model dynamical systems. In this dissertation, we use noisy observations to address two learning tasks regarding dynamical systems. The first one is called ODE parameter estimation, where the shape of the ODEs are known, and we try to learn (estimate) the parameters. The second task is called learning the governing equations, where we learn both the shape and the parameters of the ODEs.

In the first part of this dissertation, we address the ODE parameter estimation problem. We propose and analyze a block coordinate descent proximal algorithm (BCD-prox) for simultaneous filtering and parameter estimation of ODE models. The main idea is to learn the states and the parameters in an alternation, where the states are restricted to change slowly from one iteration to the next. As we show on ODE systems with up to $d=40$ dimensions, as compared to state-of-the-art methods, BCD-prox exhibits increased robustness (to noise, parameter initialization, and hyperparameters), decreased training times, and improved accuracy of both filtered states and estimated parameters. We show how BCD-prox can be used with multistep numerical discretizations, and we establish convergence of BCD-prox under hypotheses that include real systems of interest.

In the second part of this dissertation, we address the problem of learning the governing equations given the noisy observations. While we use powerful neural networks to learn the ODEs, we propose a novel structure that makes our network interpretable. The idea is to use the neural network to learn a set of one-dimensional and multi-dimensional shape functions, whose linear combinations give use the equations. To make our method robust to the noise in the observations, we learn the clean states and the parameters of the network simultaneously using the block coordinate descent proximal algorithm (BCD-prox). As we show in our experiments, our method is robust to its hyperparameter, robust to the noise, and outperforms the state-of-the-art methods by achieving more accurate state predictions.

Tools help us perform tasks. But they also change the way we think. In five studies, I explored the lasting effects — the cognitive residuals — of using two common forms of automated spelling assistance, spell-checking and autocorrect. These different spelling assistance systems aim to provide similar spelling outcomes, but users interact quite differently with each. By comparing differences in spelling skills after using these different types of spelling assistance, I assessed differences in the cognitive residuals that arise from direct interaction (spell-checking) and passive support (autocorrect). Results suggest that the passive support provided by autocorrect is just as good as for the development of spelling skill as the direct interaction offered by traditional forms of spell-checking. However, I also demonstrate a new type of direct-interaction spell-checking that outperforms traditional approaches in creating cognitive residuals for users that help improve their spelling skills.