Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (7 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2017)

At any given moment, the mind needs to decide how to think about what, and for how long. The mind's ability to manage itself is one of the hallmarks of human cognition, and these meta-level questions are crucially important to understanding how cognition is so fluid and flexible across so many situations. In this thesis, I investigate the problem of cognitive resource management by focusing in particular on the domain of mental simulation. Mental simulation is a phenomenon in which people can perceive and manipulate objects and scenes in their imagination in order to make decisions, predictions, and inferences about the world. Importantly, if thinking is computation, then mental simulation is one particular type of computation analogous to having a rich, forward model of the world.

Given access to such a model as rich and flexible as mental simulation, how should the mind use it? How does the mind infer anything from the outcomes of its simulations? How many resources should be allocated to running which simulations? When should such a rich forward model be used in the first place, in contrast to other types of computation such as heuristics or rules? Understanding the answers to these questions provides broad insight into people's meta-level reasoning because mental simulation is involved in almost every aspect of cognition, including perception, memory, planning, physical reasoning, language, social cognition, problem solving, scientific reasoning, and even creativity. Through a series of behavioral experiments combined with machine learning models, I show how people adaptively use their mental simulations to learn new things about the world; that they choose which simulations to run based on which they think will be more informative; and that they allocate their cognitive resources to spend less time on easy problems and more time on hard problems.

    Cover page: Metareasoning and Mental Simulation
    • Article
    • Peer Reviewed
    Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 41 (2019)
    Humans regularly perform tasks that require combining infor-mation across several sources of information to learn, reason,and make decisions. Bayesian models provide a computa-tional framework, and a normative account, for how humanscarry out these tasks. However, exact inference is intractablein most real-world situations, and extensive empirical workshows that human behavior often deviates significantly fromthe Bayesian optimum. A promising possibility is that peopleinstead approximate rational solutions using bounded avail-able resources. In this workshop, we bring together lead-ing researchers from cognitive science, neuroscience and ma-chine learning to build a better understanding of bounded op-timality in how humans learn, reason and make decisions.
      Cover page: Heuristics, hacks, and habits:Boundedly optimal approaches to learning, reasoning and decision making
      • Article
      • Peer Reviewed
      Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 37 (2015)
      In this paper, we investigate how people use mental simulations: do people vary the number of simulations that they run in order to optimally balance speed and accuracy? We combined a model of noisy physical simulation with a decision making strategy called the sequential probability ratio test, or SPRT (Wald, 1947). Our model predicted that people should use more samples when it is harder to make an accurate prediction due to higher simulation uncertainty. We tested this through a task in which people had to judge whether a ball bouncing in a box would go through a hole or not. We varied the uncertainty across trials by changing the size of the holes and the margin by which the ball went through or missed the hole. Both people’s judgments and response times were well-predicted by our model, demonstrating that people have a systematic strategy to allocate resources for mental simulation
        Cover page: Think again? The amount of mental simulation tracks uncertainty in the outcome
        • Article
        • Peer Reviewed
        Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 39 (2017)
        In the history of cognitive science, there have been two com-peting philosophies regarding how people reason about theworld. In one, people rely on rich, generative models to makepredictions about a wide range of scenarios; while in the other,people have a large “bag of tricks”, idiosyncratic heuristics thattend to work well in practice. In this paper, we suggest thatrather than being in opposition to one another, these two ideascomplement each other. We argue that people’s capacity formental simulation may support their ability to learn new cue-based heuristics, and demonstrate this phenomenon in two ex-periments. However, our results also indicate that participantsare far less likely to learn a heuristic when there is no logical orexplicitly conveyed relationship between the cue and the rele-vant outcome. Furthermore, simulation—while a potentiallyuseful tool—is no substitute for real world experience
          Cover page: Discovering simple heuristics from mental simulation
          • Article
          Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 40 (2018)
            Cover page: Strategies and Representations in Physical Inference
            • Article
            Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 40 (2018)
            While current deep learning systems excel at tasks such asobject classification, language processing, and gameplay, fewcan construct or modify a complex system such as a tower ofblocks. We hypothesize that what these systems lack is a “re-lational inductive bias”: a capacity for reasoning about inter-object relations and making choices over a structured descrip-tion of a scene. To test this hypothesis, we focus on a task thatinvolves gluing pairs of blocks together to stabilize a tower,and quantify how well humans perform. We then introducea deep reinforcement learning agent which uses object- andrelation-centric scene and policy representations and apply itto the task. Our results show that these structured represen-tations allow the agent to outperform both humans and morena ̈ıve approaches, suggesting that relational inductive bias isan important component in solving structured reasoning prob-lems and for building more intelligent, flexible machines.
              Cover page: Relational inductive bias for physical construction in humans and machines
              • Article
              • Peer Reviewed
              Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 38 (2016)
                Cover page: Wallace: Automating Cultural Evolution Experiments Through Crowdsourcing
                Top