skip to main content

Title: Why? Why not? When? Visual Explanations of Agent Behaviour in Reinforcement Learning
Award ID(s):
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
2022 IEEE 15th Pacific Visualization Symposium (PacificVis)
Page Range / eLocation ID:
111 to 120
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Explaining why an answer is (or is not) returned by a query is important for many applications including auditing, debugging data and queries, and answering hypothetical questions about data. In this work, we present the first practical approach for answering such questions for queries with negation (first-order queries). Specifically, we introduce a graph-based provenance model that, while syntactic in nature, supports reverse reasoning and is proven to encode a wide range of provenance models from the literature. The implementation of this model in our PUG (Provenance Unification through Graphs) system takes a provenance question and Datalog query as an input and generates a Datalog program that computes an explanation, i.e., the part of the provenance that is relevant to answer the question. Furthermore, we demonstrate how a desirable factorization of provenance can be achieved by rewriting an input query. We experimentally evaluate our approach demonstrating its efficiency. 
    more » « less
  2. Scenario-inding tools like the Alloy Analyzer are widely used in numerous concrete domains like security, network analysis, UML analysis, and so on. They can help to verify properties and, more generally, aid in exploring a system’s behavior. While scenario inders are valuable for their ability to produce concrete examples, individual scenarios only give insight into what is possible, leaving the user to make their own conclusions about what might be necessary. This paper enriches scenario inding by allowing users to ask łwhy?ž and łwhy not?ž questions about the examples they are given. We show how to distinguish parts of an example that cannot be consistently removed (or changed) from those that merely relect underconstraint in the speciication. In the former case we show how to determine which elements of the speciication and which other components of the example together explain the presence of such facts. This paper formalizes the act of computing provenance in scenario- inding. We present Amalgam, an extension of the popular Alloy scenario-inder, which implements these foundations and provides interactive exploration of examples. We also evaluate Amalgam’s algorithmics on a variety of both textbook and real-world examples. 
    more » « less
  3. null (Ed.)
  4. This research paper examines retaining traditionally underrepresented minorities (URM) in STEM fields. The retention of URM students in STEM fields is a current area of focus for engineering education research. After an extensive literature review and examination of best practices in retaining the targeted group, a cohort-based, professional development program with a summer bridge component was developed at a large land grant institution in the Mid-Atlantic region. One programmatic goal was to increase retention of underrepresented students in the engineering college which, ultimately, is expected to increase diversity in the engineering workforce. The program has a strong focus on cohort building, teamwork, mentorship, and developing an engineering identity. Students participate in a week-long summer bridge component prior to the start of their first semester. During their first year, students take a class as a cohort each semester, participate in an industrial site visit, and interact with faculty mentors. Since 2016 the program has been funded by a National Science Foundation S-STEM grant, which provides scholarships to eligible program participants. Scholarships start at $4,500 during year one, and are renewable for up to five years, with an incremental increase of $1000 annually for years one through four. Even with the professional development program providing support and scholarships alleviating the financial burden of higher education, students are still leaving engineering. The 2016-2017 cohort consisted of five scholarship recipients, of which three remained in engineering as of fall 2018, the beginning of their third year. The 2017-2018 cohort consisted of seven scholarship recipients, of which five remained in engineering as of fall 2018, their second year. While the numbers of this scholarship group are small, their retention rate is alarmingly below the engineering college retention rate. Why? This paper presents the results of additional investigations of the overall program cohorts (not only the scholarship recipients) and their non-program peers with the aim of determining predictors of retention in the targeted demographic. Student responses to three survey instruments: GRIT, MSLQ, and LAESE were analyzed to determine why students were leaving engineering, even though the program they participated in was strongly rooted in retention based literature. Student responses on program exit surveys were also analyzed to determine non-programmatic elements that may cause students to leave engineering. Results of this research is presented along with “lessons learned” and suggested actions to increase retention among the targeted population. 
    more » « less
  5. null (Ed.)