Maria D Chang  Maria D Chang photo         

contact information

Research Staff Member, AI
Almaden Research Center, San Jose, CA, USA
  

links

Professional Associations

Professional Associations:  Association for the Advancement of Artificial Intelligence (AAAI)  |  Cognitive Science Society  |  CRA-W

profile


Publications: Maria's Google Scholar Profile

 

I am a Research Staff Member specializing in knowledge representation and reasoning, especially as it pertains to natural language understanding, spatial reasoning, analogical reasoning, and reasoning about events.

My work touches on the following questions:

1) How do we build robust knowledge representations that can support expressive reasoning? Distributed representations are robust, compact (relative to the number of concepts they can represent), and learnable. Structured symbolic representations are typically used in a non-distributed fashion, yet they are the foundation of logical reasoning, they lend themselves to explanations, and they can enable other types of sophisticated reasoning (e.g. analogical reasoning). One of the fundamental questions of knowledge representation is how to extend these two seemingly opposing categories to capture the robustness, compositionality, interpretability, and expressivity needed for human-like learning and reasoning. In recent work, we have explored how to encode external symbolic knowledge graphs as distributed representations to improve performance on natural language inference (Wang et al, 2019; Kapanipathi et al. 2019). We plan to explore other neurosymbolic approaches for complex question answering as well as reasoning about events appearing in natural language, images, and video.

2) Can analogical reasoning enable human-like reasoning and knowledge acquisition? Analogical reasoning is an incredibly powerful part of human cognition. My work has explored how symbolic models of analogy can be used to judge similarity between hand-drawn sketches (using a sketching interface that facilitates semantic parsing) in educational software (Chang & Forbus, 2012; Chang & Forbus, 2014). Those same models of analogical reasoning can be used to capture qualitative science knowledge from sketches and natural language (Chang & Forbus, 2015; Chang, 2016).

3) How can AI enable engaging and effective learning environments? In prior work (at the qualitative reasoning group at Northwestern University) I helped to develop and evaluate a sketch-based tutoring system, called Sketch Worksheets, that used computational models of spatial reasoning and analogy to provide students with timely feedback on sketching exercises (Forbus et al., 2017; Garnier et al. 2017). More recently, in a collaboration with Pearson Education, we developed a dialogue-based tutoring system with a particular focus on domain adaptation (Chang et al, 2018; Ventura et al. 2018). We also explored innovative ways of enhancing the dialogue experience with visual components (Ahn et al 2018). 

 

Ahn, J.-w.; Chang, M.; Watson, P.; Tejwani, R.; Sundararajan, S.; Abuelsaad, T.; and Prabhu, S. 2018. Adaptive visual dialog for intelligent tutoring systems. In International Conference on Artificial Intelligence in Education, 413–418. Springer, Cham.

Chang, M. D., and Forbus, K. D. 2014. Using analogy to cluster hand-drawn sketches for sketch-based educational software. AI Magazine 35(1):76.

Chang, M. D., and Forbus, K. D. 2015. Towards interpretation strategies for multimodal instructional analogies. In Proceedings of the 28th International Workshop on Qualitative Reasoning (QR2015).

Chang, M.; Ventura, M.; Ahn, J.-w.; Foltz, P.; Ma, T.; Dhamecha, T. I.; Marvaniya, S.; Watson, P.; D’helon, C.; Wetzel, A.; et al. 2018. Dialogue-based tutoring at scale: Design and challenges. ICLS, London.

Chang, M., and Forbus, K. D. 2012. Using quantitative information to improve analogical matching between sketches. In Twenty-Fourth IAAI Conference.

Chang, M. 2016. Capturing qualitative science knowledge with multimodal instructional analogies. Ph.D. Dissertation, Northwestern University.

Forbus, K. D.; Chang, M.; McLure, M.; and Usher, M. 2017. The cognitive science of sketch worksheets. Topics in cognitive science 9(4):921–942.

Garnier, B.; Chang, M.; Ormand, C.; Matlen, B.; Tikoff, B.; and Shipley, T. F. 2017. Promoting sketching in introductory geoscience courses: Cogsketch geoscience worksheets. Topics in cognitive science 9(4):943–969.

Kapanipathi, P.; Thost, V.; Patel, S. S.; Whitehead, S.; Abdelaziz, I.; Balakrishnan, A.; Chang, M.; Fadnis, K.; Gunasekara, C.; Makni, B.; et al. 2019. Infusing knowledge into the textual entailment task using graph convolutional networks. arXiv preprint arXiv:1911.02060.  (to appear in AAAI 2020)

Ventura, M.; Chang, M.; Foltz, P.; Mukhi, N.; Yarbro, J.; Salverda, A. P.; Behrens, J.; Ahn, J.-w.; Ma, T.; Dhamecha, T. I.; et al. 2018. Preliminary evaluations of a dialoguebased digital tutor. In International Conference on Artificial Intelligence in Education, 480–483. Springer, Cham.

Wang, X.; Kapanipathi, P.; Musa, R.; Yu, M.; Talamadupula, K.; Abdelaziz, I.; Chang, M.; Fokoue, A.; Makni, B.; Mattei, N.; et al. 2019. Improving natural language inference using external knowledge in the science questions domain. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, 7208–7215.