Person:
Feigh,
Karen M.
Feigh,
Karen M.
Permanent Link
Associated Organization(s)
Organizational Unit
ORCID
ArchiveSpace Name Record
Publication Search Results
Now showing
1 - 9 of 9
-
ItemHuman Teacher’s Perception of Teaching Methods for Machine Learning Algorithms(Georgia Institute of Technology, 2019-02-06) Feigh, Karen M.A goal of interactive machine learning (IML) is to create robots or intelligent agents that can be easily taught how to perform tasks by individuals with no specialized training. To achieve that goal, researchers and designers must understand how certain design decisions impact the human’s experience of teaching the agent, such as influencing the agent’s perceived intelligence. We posit that the type of feedback a robot can learn from effects the perceived intelligence of the robot, similar to its physical appearance. This talk will discuss different methods of natural language instruction including critique and action advice. We conducted multiple human-in-the-loop experiments, in which people trained agents with different teaching methods but, unknown to each participant, the same underlying machine learning algorithm. The results show that the mechanism of teaching has an impact on a human teacher’s perception of the agent, including feelings of frustration, perceptions of intelligence, and performance, while only minimally impacting the agent’s performance.
-
ItemJudicial Evidential Reasoning for Decision Support Applied to Orbit Insertion Failure(Georgia Institute of Technology, 2017-11) Jaunzemis, Andris D. ; Minotra, Dev ; Holzinger, Marcus J. ; Feigh, Karen M. ; Chan, Moses W. ; Shenoy, Prakash P.Realistic decision-making often occurs with insufficient time to gather all possible evidence before a decision must be rendered, requiring an efficient process for prioritizing between potential action sequences. This work aims to develop a decision support system for tasking sensor networks to gather evidence to resolve hypotheses in the face of ambiguous, incomplete, and uncertain evidence. Studies have shown that decision-makers demonstrate several biases in decisions involving probability judgement, so decision-makers must be confident that the evidence-based hypothesis resolution is strong and impartial before declaring an anomaly or reacting to a conjunction analysis. Providing decision-makers with the ability to estimate uncertainty and ambiguity in knowledge has been shown to augment effectiveness. The proposed framework, judicial evidential reasoning (JER), frames decision-maker questions as rigorously testable hypotheses and employs an alternating-agent minimax optimization on belief in the null proposition. This approach values impartiality in addition to time efficiency: an ideal action sequence gathers evidence to quickly resolve hypotheses while guarding against bias. JER applies the Dempster-Shafer theory of belief functions to model knowledge about hypotheses and quantify ambiguity, and adversarial optimization techniques are used to make many-hypothesis resolution computationally tractable. This work includes derivation and application of the JER formulation to a GTO insertion maneuver anomaly scenario.
-
ItemEvaluation of the Usability and Learnability of Vehicle Sketch Pad in Academia(Georgia Institute of Technology, 2016-06-02) MacLeod, Stephanie F. ; Feigh, Karen M.This paper describes an evaluation of the usability and learnability of Vehicle Sketch Pad (VSP) within the academic aerospace engineering field. In addition, this paper provides suggestions on how Vehicle Sketch Pad should develop to better provide a desirable graphical user interface (GUI). Aerospace Engineering graduate students were asked to perform specific tasks within Vehicle Sketch Pad, and their performance, efficiency, and workload while completing these tasks were measured. The same students were asked to perform similar tasks within a more traditional computer aided design (CAD) tool, AutoDesk Inventor. At this time, how the user interfaced with the Inventor GUI and what features within the interface they liked and did not like were measured. Analysis of the results of this experiment showed that all three experience-based noise variables, undergraduate major, prior courses taken, and prior VSP experience, do impact the user’s performance, efficiency, and workload while completing a task in VSP. For example, those with a more in-depth Aerospace-related background perceived less workload than those with alternative backgrounds. Additionally, the VSP GUI proved to have a statistically significant impact on the user’s performance, efficiency, and workload. For example, those who experienced confusion due to the workspace layout expressed they felt utilizing VSP was a very mentally demanding task. As a result this analysis, a roadmap for VSP development opportunities was created. This roadmap included users’ opinions and the researcher’s observances of gaps within the VSP GUI in addition to suggestions based on the Inventor GUI.
-
ItemSingle-operator Multi-vehicle Human-Automation Interface Study dataset(Georgia Institute of Technology, 2015-05) Feigh, Karen M. ; Johnson, Eric N. ; Christmann, Hans ClausWith the achievement of autonomous flight for small unmanned aircraft, currently ongoing research is expanding the capabilities of systems utilizing such vehicles for various tasks. This allows shifting the research focus from the individual systems to task execution benefits resulting from interaction and collaboration of several aircraft. Given that some available high-fidelity simulations do not yet support multi-vehicle scenarios, a multi-vehicle framework has been introduced which allows several individual single-vehicle systems to be combined into a larger multi-vehicle scenario with little to no special requirements towards the single-vehicle systems. The created multi-vehicle system offers real-time software-in-the-loop simulations of vehicle teams across multiple hosts and enables a single operator to command and control a several unmanned aircraft beyond line-of-sight in geometrically correct two-dimensional cluttered environments through a multi-hop network of data relaying intermediaries. The related dissertation by Christmann presents the main aspects of the developed system: the underlying software framework and application programming interface, the utilized inter- and intrasystem communication architecture, the graphical user interface, and implemented algorithms and operator aid heuristics to support the management and placement of the vehicles.The effectiveness of the aid heuristics is validated through a human subject study which showed that the provided operator support systems significantly improve the operators' performance in a simulated first responder scenario. This dataset contains the collected data of that human subject study.
-
ItemDecision Making with Incomplete Information Dataset(Georgia Institute of Technology, 2014-07-09) Feigh, Karen M. ; Canellas, Marc ; Chua, Zarrin K.Decision makers are often required to make decisions with incomplete information. In order to design decision support systems (DSSs) utilizing restrictiveness and guidance to assist decision makers in these situations, it is essential to understand how certain decision making strategies are affected by incomplete information. This paper presents the results of a simulation measuring the accuracy and effort of take-the-best (TTB) and Tallying alongside two normative-rational decision making strategies, weighted-additive (WADD) and equal-weighting (EW) in scenarios with varying levels of total information, information imbalance, dispersion, and dominance. The results show there is significant variability in the effort requirements of heuristic strategies which may diminish the arguments for effort-accuracy trade-offs. Additionally, heuristic strategies were shown to be closest in accuracy to normative-rational strategies when context features matched dynamic decision settings. Ultimately, methods for restrictiveness and guidance based on trade-offs between total information and information imbalance were shown to enable reductions in total information that actually increased the accuracy of heuristics.
-
ItemNeural Networks for Assessment of Flight Deck Human-Automation Interaction Dataset(Georgia Institute of Technology, 2013-08-22) Feigh, Karen M. ; Sullivan, Katlyn B. ; Mappus, Rudolph Louis, IV ; Durso, Frank ; Fischer, Ute ; Pop, Vlad ; Mosier, Kathleen T. ; Morrow, Daniel G.
-
ItemDevelopment and Evaluation of an Automated Path Planning Aid(Georgia Institute of Technology., 2012-11) Watts, Robert ; Christmann, Hans Claus ; Johnson, Eric N. ; Feigh, Karen M. ; Tsiotras, PanagiotisHandling en route emergencies in modern transport aircraft through adequate teamwork between the pilot, the crew and the aircraft’s automation systems is an ongoing and active field of research. An automated path planning aid tool can assist pilots with the tasks of selecting a convenient landing site and developing a safe path to land at this site in the event of an onboard emergency. This paper highlights the pilot evaluation results of a human factors study as part of such a proposed automated planning aid. Focusing on the interactions between the pilot and the automated planning aid, the presented results suggest that a particular implementation of the pilot aid interface, which uses a simple dial to sort the most promising landing sites, was effective. This selectable sorting capability, motivated by the anticipated cognitive mode of the pilot crew, improved the quality of the selected site for the majority of the cases tested. Although the presented approach increased the average time required for the selection of an alternate landing site, it decreased the time to complete the task in the case of emergencies unfamiliar to the pilot crew.
-
ItemAssessment of the accuracy of existing real-time wake vortex models(Georgia Institute of Technology, 2011-03-31) Sankar, Lakshmi N. ; Schrage, Daniel P. ; Feigh, Karen M. ; Huff, Brian ; Flick, Ashley ; Manivannan, Vasu
-
ItemExamination of Human Performance During Lunar Landing(Georgia Institute of Technology, 2010-03) Chua, Zarrin K. ; Feigh, Karen M. ; Braun, Robert D.Experimentally derived data was extrapolated to compare the lunar landing performance of human pilots to that of an automated landing system.12 The results of this investigation are presented. Overall, the pilots performed equal to or better than the automated system in 18% of the relevant cases, but required more fuel. Pilot site selections were further investigated as a function of the time to complete. Each hypothetical case was compared to the automated system, across a range of performance criteria weighting distributions. This performance criteria is threefold – proximity to point of interest, safety of the site, and fuel consumed. In general, the pilots perform better than the automated system in terms of safety and proximity to points of interest criteria. However, as the priority of fuel conservation increases, the tradeoff between using an autonomous landing system versus a human-in-command system favors the automation, especially if the pilot is not able to make the proper decision within a performance criteria specific threshold.