THIS GRANT IS A CONTINUATION OF N00014-13-1-0470 An Analogical Approach to Autonomy and Social Inference in a Cognitive Architecture
Abstract
Year 1Computational:? Develop a vocabulary of psychologically plausible primitive mental operations and an initial set of higher level plans that combine these plans to perform tasks, to ensure their expressiveness.? Develop internal metrics to support learning self models about task execution? Finish experiments on learning more abstract role relations.? Implement infrastructure to support cognitive state sensing.? Experiment with comparison-based appraisal algorithms.? Develop first-cut representations for Companion episodic memories.? Extend Companions to include web-based data collection capabilities, to support crowdsourcing experiments.? Develop initial set of knowledge capture games to collect data for learning social inference.? Extend the Interaction Manager to support extensible interpretation of imperatives.? Experiment with methods for integrating schemas into relations, to improve accuracy of analogical mappings and retrievals.? Experiment with heuristic strategies for hypothesis updating in category learning.Psychological:? In Stock and Flow studies, experiment on effects of giving qualitative principles on understanding and transfer.? In Causal Systems studies, investigate effects of analogy and explanation on retention over a delay.? Pilot studies varying degree of constraint expressed in higher-order relation in analogical processing.? In studies of higher-order relations in analogy, norm materials for social and metacognitive patterns.Year 2Computational:? Develop methods for learning to estimate properties of problems, e.g. difficulty, probability of a result being correct.? Develop methods for detection and diagnosis of failures in task execution, including both model-based and analogy-based methods.? Develop task models for metaphor interpretation, including rerepresentation and decomposition.? Implement first-cut metacognitive processes for cognitive statesensing related to analogical operations (i.e. tip of the tongue, click of a good answer)? Use crowdsourcing to gather data for experiments in learning social inference.? Develop initial analogy-based social reasoning methods for mindreading.? Develop analogical theory of mind account to the point where it can be tested by experiments using the Companion architecture, using handcoded stimuli.? Extend the notion of generalization context in SAGE to allow superordinate/subordinate relationships between concepts, to support broader-scale learning.? Demonstrate the ability to do rerepresentation in service of understanding metaphors expressed in simplified English.Psychological:? In Stock and Flow studies, factorially vary presence of QP explanations with analogical comparison on understanding and transfer.? In Causal Systems studies, investigate less intensive methods: e.g., adding relational labels to passages? dependent measures include causal sorting and inference.? Test effects of degree of constraint expressed in higher-order relation on inference of the consequent relation as well as on inference rates for the higher-order relation itself.? In studies of social and metacognitive patterns, test effects of higher-orderpatterns on (a) analogical inference and (b) categorization.? Pilot studies of effects of relational labels on encoding and memory, for both social and physical systems.Year 3Computational:? Develop methods to learn to improve information processing plans via experimentation.? Use our improved category learning model to explore additional phenomena in language learning (e.g. adaptive encoding, using changes in SAGE parameters to tune language learning strategies).? Test metaphor interpretation capabilities with crowdsourced data.? Implement first-cut metacognitive processes for cognitive state sensingin overall effort, focusing on surprise, curiosity, frustration, and boredom.? Extend analogy-based social reasoning methods as needed.? Experiment with strategies for automatically splitting concepts, to create subordinate concepts, and autom
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141612613
Entities
People
- Dedre Gentner
Organizations
- Northwestern University
- Office of Naval Research
- United States Navy