Employee performance evaluations have been subject to a lot of criticism and organizations are now leveraging artificial intelligence (AI) to enhance and maximize the efficiency and accuracy of these performance evaluations. Although organizations assume that AI-driven performance evaluation systems will enhance traditional performance evaluation systems, a growing body of research documents the phenomenon of algorithm aversion, the human tendency to discount algorithm/computer generated advice more heavily than human advice. Using an employee performance evaluation setting, I conduct an experiment to examine how managers will resolve differences between two contradictory judgments, their own judgment and an AI's judgment. I find that in addition to algorithm aversion and an individual's attitude towards technology, the performance evaluation measures (objective or subjective), and more importantly, the consequence of the decision on the employee strongly influenced the manager's reliance on AI. Specifically, managers resolved conflict between AI and their own decision by relying on decisions that were in the employee's favor. The study contributes to existing research on the adoption of AI and management accounting research.

Accurate image reconstruction is fundamental to medical imaging diagnostics, involving the transformation of data from the sensor domain to the image domain by solving an inverse problem. In Magnetic Resonance Imaging (MRI), measurements are acquired in the k-space spatial frequency domain, and the inverse Fourier Transform is applied to reconstruct the image for diagnosis. However, exact solutions to the inverse problem using analytical models are often not possible. Partial measurements are often acquired to decrease scanning time, resulting in ill-posed inverse problems that necessitate a series of signal processing steps for optimal reconstructions. Supervised deep learning approaches have been explored for solving such inverse problems, including image reconstruction. While deep learning can tackle these challenges in a single reconstruction step, training deployable models can be challenging due to encountering unseen data distributions that deviate from the training data in real-world scenarios. In this dissertation, we first investigate the impact of complex input data design, data augmentations, adversarial noise, and hallucinations on reconstruction accuracy and robustness of deep learning-based image reconstruction methods. We illustrate how the complex input data design and architectural modifications can notably enhance performance accuracy. We showcase the emergence of artifacts when training lacks proper data augmentations such as multiple field-of-views in the dataset. Additionally, we study the effectiveness of deep learning when exposed to Gaussian versus engineered adversarial noise, proposing a technique to adapt the numerical properties of the training dataset for resilience against adversarial noise. Finally, we investigate the occurrence of hallucinations on undersampled out-of-distribution (OOD) data reconstructions and propose a method for quantifying and mitigating them through domain adaptation techniques. Due to encountering OOD data in real-world settings, it is essential to assess whether a given input falls within the training data distribution, in-distribution (ID), particularly when reconstructing medical images for diagnostic purposes. We propose a single model variance method based on the local Lipschitz metric to distinguish OOD images from ID. Our method achieves an impressive area under the curve of 99.94% for True Positive Rate versus False Positive Rate. Empirically, we demonstrate a very strong relationship between the local Lipschitz value and mean absolute error (MAE), supported by a high Spearman's rank correlation coefficient of 0.8475. Through selective prediction, we demonstrate a method to determine the local Lipschitz threshold for uncertainty as it relates to optimal model performance. Our study was validated using the AUTOMAP architecture for sensor-to-image domain MRI reconstruction. We compare our proposed approach with baseline methods of Monte-Carlo dropout and deep ensembles as well as the state-of-the-art Mean Variance Estimation (MVE) network approach. Furthermore, we showcase the versatility of our approach to other architectures and learned functions, including the UNET architecture for MRI denoising and Computed Tomography (CT) sparse-to-full view reconstruction applications. Lastly, we expand the field of deep learning to solve inverse problems to Nitrogen-vacancy (NV) center diamond magnetometry, a quantum sensing technique that measures the magnetic field produced by circuits using the NV center optical defect. We designed a MAGnetic Inverse Calculation UNET (MAGIC-UNET) to reconstruct current density images using magnetic fields as input by solving the inverse Biot-Savart law and compared it to the analytical Fourier Method. We find that the deep learning solution using the MAGIC-UNET has greater accuracy on simulated and NV-diamond magnetometry experimental data compared to the analytical Fourier Method. It also significantly reduces the magnetometer collection time due to requiring fewer signal averages. These results expand the application scope of NV-diamond magnetometry to weak current sources and the use of DL to solve inverse problems to the quantum sensor domain.

Conditional preference networks (CP-nets) model user preferences over objects described in terms of values assigned to discrete features, where the preference for one feature may depend on the values of other features. Most existing algorithms for learning CP-nets from the user's choices assume that the user chooses between pairs of objects. However, many real-world applications involve the the user choosing from all combinatorial possibilities or a very large subset. We introduce a CP-net learning algorithm for the latter type of choice, and study its properties formally and empirically.

This thesis is about a design process for creating educationally relevant tools. I submit that the key to creating tools that are educationally relevant is to focus on ensuring a high degree of alignment between the designed tool and the broader educational context into which the tool will be integrated. The thesis presents methods and processes for creating a tool that is both well aligned and relevant. The design domain of the thesis is described by a set of tools I refer to as “Configurable Embodied Interfaces”. Configurable embodied interfaces have a number of key features, they: Can sense their local surroundings through the detection of such environmental and physical parameters as light, sound, imagery, device acceleration, etc. Act on their local environment by outputting sound, light, imagery, motion of the device, etc. Are configurable in such a way as to link these inputs and outputs in a nearly unlimited number of ways. Contain active ways for users to either directly create new programs linking input and output, or to easily re-configure them by running different programs on them. Are user focused; they assume that a human being is manipulating them in some way, through affecting input and observing output of the interface. Spurred by the growth of cheap computation and sensing, a large number of educational programs have been built around use of configurable embodied interfaces in the last three decades. These programs exist in both formal and informal educational settings and are in use from early childhood through adult and community education. Typically, configurable embodied interfaces are used as tools in three major and sometimes overlapping areas: computer Science education, creative and engineering design education, and traditional science and math education. This work details three examples of collaborations between technologists and educators that led to the creation of educationally successful tools; these three examples share a focus on creating a configurable embodied interface to tackle a specific cognitive and affective set of learning goals, but differ completely in the location of the learning environment, the age and interests of the learners, and the nature of the learning goals. Through the exploration of the methods used, an analysis of the general and context-specific features of the design processes of the three accounts, and a comparison of the process used in this thesis to a conventional engineering design process, this work provides case studies and a set of guidelines that can inform technologists interested in designing educationally relevant embodied interfaces

Mobile robots that encounter people on a regular basis must react to them in some way. While traditional robot control algorithms treat all unexpected sensor readings as objects to be avoided, we argue that robots that operate around people should react socially to those people, following the same social conventions that people use around each other. This thesis presents our COMPANION framework: a Constraint-Optimizing Method for Person–Acceptable NavigatION. COMPANION is a generalized framework for representing social conventions as components of a constraint optimization problem, which is used for path planning and navigation. Social conventions, such as personal space and tending to the right, are described as mathematical cost functions that can be used by an optimal path planner. These social conventions are combined with more traditional constraints, such as minimizing distance, in a flexible way, so that additional constraints can be added easily. We present a set of constraints that specify the social task of traveling around people. We explore the implementation of this task first in simulation, where we demonstrate a robot’s behavior in a wide variety of scenarios. We also detail how a robot’s behavior can be changed by using different relative weights between the constraints or by using constraints representing different sociocultural conventions. We then explore the specific case of passing a person in a hallway, using the robot Grace. Through a user study, we show that people interpret the robot’s behavior according to human social norms, and also that people ascribe different personalities to the robot depending on its level of social behavior. In addition, we present an extension of the COMPANION framework that is able to represent joint tasks between the robot and a person. We identify the constraints necessary to represent the task of having a robot escort a person while traveling side-by-side. In simulation, we show the capability of this representation to produce behaviors such as speeding up or slowing down to travel together around corners, as well as complex maneuvers to travel through narrow chokepoints and return to a side-by-side formation. Finally, we present a newly designed robot, Companion, that is intended as a platform for general social human–robot research. Companion is a holonomic robot, able to move sideways without turning first, which we believe is an important social capability. We detail the design and capabilities of this new platform. As a whole, this thesis demonstrates both a need for, and an implementation and evaluation of, robots that navigate around people according to social norms.

No description available.

No description available.

No description available.

No description available.

No description available.