Bonan ZHAO / 赵博囡

I am a postdoctoral researcher working with Tom Griffiths and Natalia Velez at Princeton University. I am broadly interested in the cognitive mechanisms that drive conceptual discoveries, and in particular how individual cognitive constraints shape innovations and diverse understandings.

I completed my PhD at the University of Edinburgh advised by Neil Bramley and Chris Lucas. Previously I worked in data science at an adtech start-up. I received my MSc in Logic at the ILLC from University of Amsterdam, and B.A. in philosophy from Tsinghua University in Beijing, China.

Feel free to contact me at bnz@princeton.edu.

CV  /  Google Scholar  /  Twitter

* denotes equal contribution. Representative papers are highlighted.

We formalize a crafting game as a Markov decision process (MDP) and find that people can make near-optimal decisions in various game environments.

We show that people can learn 22 categories from just 4 examples, by leveraging soft compositional labels.

We present a model of conceptual bootstrapping - learning complex concepts by recursively combining simpler concepts. Our model predicts systematically different learned concepts when the same evidence is processed in different orders.

We propose that stochastic program induction algorithms (MCMC & adaptor grammars) help explain how cognition innovates, via incremental selection among random local mutations and recombinations of (parts of) a cognizer's current world model.

We propose a Bayesian computational model for the line bisection task, modeling neglect as rational inference in the face of uncertain information.

We find causal generalizations benefit from facilitory curriculums, neatly captured by an adaptor grammar model's native chunking mechanism.

To induce causal asymmetries in object-based causal generalization tasks, the agent object needs to possess three interactions cues altogether: movement, stability, and visual-nominal indicator.

Perceptual causal stimuli work differently from well-established verbal stimuli in causal categorization tasks.

We account for generalization-order effects in one-shot causal generalization and causal asymmetries in few-shot tasks with a computational framework and its process variant.

We integrate a symbolic causal function generator with a Dirichlet Process to model causal generalizations that well-match people's.

We invite a wide range of speakers to debate and explore hybrid architectures linking symbolic systems with neural approaches.

We find that people's causal generalization predictions are affected by the presentation order of generalization tasks.

We solve the deductive Mastermind game using DEL and find that it can explain children's performance in this game.

We propose a sound and complete logic where stubborn agents revise their friendship network according to their own opinions, rather than the other way around.

Resources for those interested in program induction.

A computational modeling framework that combines program induction (about the hidden causal laws) with non-parametric category inference (about their domains of influence) to account for unique patterns in human causal generalization.

We examined Sun Tzu's 2500-year-old Art of War from the perspectives of logic, mathematics, and computer science, using contemporary mind mapping methods.

Template from here