skip to main content

Title: Quantifying Individuals’ Theory-Based Knowledge Using Probabilistic Causal Graphs: A Bayesian Hierarchical Approach
Abstract

Extracting an individual’s knowledge structure is a challenging task as it requires formalization of many concepts and their interrelationships. While there has been significant research on how to represent knowledge to support computational design tasks, there is limited understanding of the knowledge structures of human designers. This understanding is necessary for comprehension of cognitive tasks such as decision making and reasoning, and for improving educational programs. In this paper, we focus on quantifying theory-based causal knowledge, which is a specific type of knowledge held by human designers. We develop a probabilistic graph-based model for representing individuals’ concept-specific causal knowledge for a given theory. We propose a methodology based on probabilistic directed acyclic graphs (DAGs) that uses logistic likelihood function for calculating the probability of a correct response. The approach involves a set of questions for gathering responses from 205 engineering students, and a hierarchical Bayesian approach for inferring individuals’ DAGs from the observed responses. We compare the proposed model to a baseline three-parameter logistic (3PL) model from the item response theory. The results suggest that the graph-based logistic model can estimate individual students’ knowledge graphs. Comparisons with the 3PL model indicate that knowledge assessment is more accurate when quantifying more » knowledge at the level of causal relations than quantifying it using a scalar ability parameter. The proposed model allows identification of parts of the curriculum that a student struggles with and parts they have already mastered which is essential for remediation.

« less
Authors:
; ; ;
Award ID(s):
1662230 1728165
Publication Date:
NSF-PAR ID:
10281761
Journal Name:
ASME 2020 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
Sponsoring Org:
National Science Foundation
More Like this
  1. Extracting an individual’s knowledge structure is a challenging task as it requires formalization of many concepts and their interrelationships. While there has been significant research on how to represent knowledge to support computational design tasks, there is limited understanding of the knowledge structures of human designers. This understanding is necessary for comprehension of cognitive tasks such as decision making and reasoning, and for improving educational programs. In this paper, we focus on quantifying theory-based causal knowledge, which is a specific type of knowledge held by human designers. We develop a probabilistic graph-based model for representing individuals’ concept-specific causal knowledge formore »a given theory. We propose a methodology based on probabilistic directed acyclic graphs (DAGs) that uses logistic likelihood function for calculating the probability of a correct response. The approach involves a set of questions for gathering responses from 205 engineering students, and a hierarchical Bayesian approach for inferring individuals’ DAGs from the observed responses. We compare the proposed model to a baseline three-parameter logistic (3PL) model from the item response theory. The results suggest that the graph-based logistic model can estimate individual students’ knowledge graphs. Comparisons with the 3PL model indicate that knowledge assessment is more accurate when quantifying knowledge at the level of causal relations than quantifying it using a scalar ability parameter. The proposed model allows identification of parts of the curriculum that a student struggles with and parts they have already mastered which is essential for remediation.« less
  2. One hallmark of human reasoning is that we can bring to bear a diverse web of common-sense knowledge in any situation. The vastness of our knowledge poses a challenge for the practical implementation of reasoning systems as well as for our cognitive theories – how do people represent their common-sense knowledge? On the one hand, our best models of sophisticated reasoning are top-down, making use primarily of symbolically-encoded knowledge. On the other, much of our understanding of the statistical properties of our environment may arise in a bottom-up fashion, for example through asso- ciationist learning mechanisms. Indeed, recent advances inmore »AI have enabled the development of billion-parameter language models that can scour for patterns in gigabytes of text from the web, picking up a surprising amount of common-sense knowledge along the way—but they fail to learn the structure of coherent reasoning. We propose combining these approaches, by embedding language-model-backed primitives into a state- of-the-art probabilistic programming language (PPL). On two open-ended reasoning tasks, we show that our PPL models with neural knowledge components characterize the distribution of human responses more accurately than the neural language models alone, raising interesting questions about how people might use language as an interface to common-sense knowledge, and suggesting that building probabilistic models with neural language-model components may be a promising approach for more human-like AI.« less
  3. Cognitive flexibility is a core component of executive function, a suite of cognitive capacities that enables individuals to update their behavior in dynamic environments. Human executive functions are proposed to be enhanced compared to other species, but this inference is based primarily on neuroanatomical studies. To address this, we examined the nature and origins of cognitive flexibility in chimpanzees, our closest living relatives. Across three studies, we examined different components of cognitive flexibility using reversal learning tasks where individuals first learned one contingency and then had to shift responses when contingencies flipped. In Study 1, we tested n = 82more »chimpanzees ranging from juvenility to adulthood on a spatial reversal task, to characterize the development of basic shifting skills. In Study 2, we tested how n = 24 chimpanzees use spatial versus arbitrary perceptual information to shift, a proposed difference between human and nonhuman cognition. In Study 3, we tested n = 40 chimpanzees on a probabilistic reversal task. We found an extended developmental trajectory for basic shifting and shifting in response to probabilistic feedback—chimpanzees did not reach mature performance until late in ontogeny. Additionally, females were faster to shift than males were. We also found that chimpanzees were much more successful when using spatial versus perceptual cues, and highly perseverative when faced with probabilistic versus consistent outcomes. These results identify both core features of chimpanzee cognitive flexibility that are shared with humans, as well as constraints on chimpanzee cognitive flexibility that may represent evolutionary changes in human cognitive development.« less
  4. Graph structured data are abundant in the real world. Among different graph types, directed acyclic graphs (DAGs) are of particular interest to machine learning researchers, as many machine learning models are realized as computations on DAGs, including neural networks and Bayesian networks. In this paper, we study deep generative models for DAGs, and propose a novel DAG variational autoencoder (D-VAE). To encode DAGs into the latent space, we leverage graph neural networks. We propose an asynchronous message passing scheme that allows encoding the computations on DAGs, rather than using existing simultaneous message passing schemes to encode local graph structures. Wemore »demonstrate the effectiveness of our proposed DVAE through two tasks: neural architecture search and Bayesian network structure learning. Experiments show that our model not only generates novel and valid DAGs, but also produces a smooth latent space that facilitates searching for DAGs with better performance through Bayesian optimization.« less
  5. Few-shot graph classification aims at predicting classes for graphs, given limited labeled graphs for each class. To tackle the bottleneck of label scarcity, recent works propose to incorporate few-shot learning frameworks for fast adaptations to graph classes with limited labeled graphs. Specifically, these works propose to accumulate meta-knowledge across diverse meta-training tasks, and then generalize such meta-knowledge to the target task with a disjoint label set. However, existing methods generally ignore task correlations among meta-training tasks while treating them independently. Nevertheless, such task correlations can advance the model generalization to the target task for better classification performance. On the othermore »hand, it remains non-trivial to utilize task correlations due to the complex components in a large number of meta-training tasks. To deal with this, we propose a novel few-shot learning framework FAITH that captures task correlations via constructing a hierarchical task graph at different granularities. Then we further design a loss-based sampling strategy to select tasks with more correlated classes. Moreover, a task-specific classifier is proposed to utilize the learned task correlations for few-shot classification. Extensive experiments on four prevalent few-shot graph classification datasets demonstrate the superiority of FAITH over other state-of-the-art baselines.« less