An official website of the United States government
Concept-based interpretations of black-box models are often more intuitive for humans to understand. The most widely adopted approach for concept-based interpretation is Concept Activation Vector (CAV). CAV relies on learning a linear relation between some latent representation of a given model and concepts. The linear separability is usually implicitly assumed but does not hold true in general. In this work, we started from the original intent of concept-based interpretation and proposed Concept Gradient (CG), extending concept-based interpretation beyond linear concept functions. We showed that for a general (potentially non-linear) concept, we can mathematically evaluate how a small change of concept affecting the model’s prediction, which leads to an extension of gradient-based interpretation to the concept space. We demonstrated empirically that CG outperforms CAV in both toy examples and real world datasets.
more »
« less
Semantic Fast-Forwarding for Video Training Set Construction
We introduce the concept of semantic fast-forwarding of video streams for efficient labeling of training data for activity recognition. We show that this concept can be realized by combining deep learning within individual frames, with spatial and temporal entity-relationship reasoning about detected objects. We describe a prototype that implements this concept, and present preliminary experimental results on its feasibility and value.
more »
« less
- Award ID(s):
- 2106862
- PAR ID:
- 10409698
- Date Published:
- Journal Name:
- HotMobile '23: Proceedings of the 24th International Workshop on Mobile Computing Systems and Applications
- Page Range / eLocation ID:
- 29 to 35
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Concept maps have emerged as a valid and reliable method for assessing deep conceptual understanding in engineering education within disciplines as well as interdisciplinary knowledge integration across disciplines. Most work on concept maps, however, focuses on undergraduates. In this paper, we use concept maps to examine changes in graduate students’ conceptual understanding and knowledge integration resulting from an interdisciplinary graduate program. Our study context is pair of foundational, team-taught courses in an interdisciplinary Disaster Resilience and Risk Management (DRRM) graduate program. The courses include a 3-hour research course and a 1-hour seminar that aim to build student understanding within and across Urban Affairs and Planning, Civil and Environmental Engineering, Geosciences, and Business Information Technology. The courses introduce core principles of DRRM and relevant research methods in these disciplines, and drive students to understand the intersections of these disciplines in the context of planning for and responding to natural and human-made disasters. To understand graduate student growth from disciplinary-based to interdisciplinary scholars, we pose the research questions: 1) In what ways do graduate students’ understandings of DRRM change as a result of their introduction to an interdisciplinary graduate research program? and 2) To what extent and in what ways do concept maps serve as a tool to capture interdisciplinary learning in this context? Data includes pre/post concept maps centered on disaster resilience and risk management, a one-page explanation of the post-concept map, and ethnographic field notes gathered from class and faculty meetings. Pre-concept maps were collected on the first day of class; post-concept maps will be collected as part of the final course assignment. We assess the students’ concept maps for depth of conceptual understanding within disciplines and interdisciplinary competency across disciplines, using the field notes to provide explanatory context. The results presented in this paper support the inclusion of an explanation component to concept maps, and also suggest that concept maps alone may not be the best measure of student understanding of concepts within and across disciplines in this specific context. If similar programs wish to use concept maps as an assessment method, we suggest the inclusion of an explanation component and suggest providing explicit instructions that specify the intended audience. We also suggest using a holistic scoring method, as it is more likely to capture nuances in the concept maps than traditional scoring methods, which focus solely on counting factors like hierarchies and number of cross-links.more » « less
-
Abstract Ecologists who study human-dominated places have adopted a social–ecological systems framework to recognize the coproduced links between ecological and social processes. However, many social scientists are wary of the way ecologists use the systems concept to represent such links. This wariness is sometimes due to a misunderstanding of the contemporary use of the systems concept in ecology. We aim to overcome this misunderstanding by discussing the contemporary systems concept using refinements from biophysical ecology. These refinements allow the systems concept to be used as a bridge rather than a barrier to social–ecological interaction. We then use recent examples of extraordinary fire to illustrate the usefulness and flexibility of the concept for understanding the dynamism of fire as a social–ecological interaction. The systems idea is a useful interdisciplinary abstraction that can be contextualized to account for societally important problems and dynamics.more » « less
-
Hitzler, Pascal; Sarker, Md Kamruzzaman (Ed.)Understanding complex machine learning models such as deep neural networks with explanations is crucial in various applications. Many explanations stem from the model perspective, and may not necessarily effectively communicate why the model is making its predictions at the right level of abstraction. For example, providing importance weights to individual pixels in an image can only express which parts of that particular image are important to the model, but humans may prefer an explanation which explains the prediction by concept-based thinking. In this work, we review the emerging area of concept based explanations. We start by introducing concept explanations including the class of Concept Activation Vectors (CAV) which characterize concepts using vectors in appropriate spaces of neural activations, and discuss different properties of useful concepts, and approaches to measure the usefulness of concept vectors. We then discuss approaches to automatically extract concepts, and approaches to address some of their caveats. Finally, we discuss some case studies that showcase the utility of such concept-based explanations in synthetic settings and real world applications.more » « less
-
null (Ed.)Abstract Objective The Unified Medical Language System (UMLS) integrates various source terminologies to support interoperability between biomedical information systems. In this article, we introduce a novel transformation-based auditing method that leverages the UMLS knowledge to systematically identify missing hierarchical IS-A relations in the source terminologies. Materials and Methods Given a concept name in the UMLS, we first identify its base and secondary noun chunks. For each identified noun chunk, we generate replacement candidates that are more general than the noun chunk. Then, we replace the noun chunks with their replacement candidates to generate new potential concept names that may serve as supertypes of the original concept. If a newly generated name is an existing concept name in the same source terminology with the original concept, then a potentially missing IS-A relation between the original and the new concept is identified. Results Applying our transformation-based method to English-language concept names in the UMLS (2019AB release), a total of 39 359 potentially missing IS-A relations were detected in 13 source terminologies. Domain experts evaluated a random sample of 200 potentially missing IS-A relations identified in the SNOMED CT (U.S. edition) and 100 in Gene Ontology. A total of 173 of 200 and 63 of 100 potentially missing IS-A relations were confirmed by domain experts, indicating that our method achieved a precision of 86.5% and 63% for the SNOMED CT and Gene Ontology, respectively. Conclusions Our results showed that our transformation-based method is effective in identifying missing IS-A relations in the UMLS source terminologies.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3572864.3580331
