1,721,413 research outputs found
Discovering Structure by Learning Sparse Graphs
Full text linkSystems of concepts such as colors, animals, cities, and artifacts are richly structured, and people discover the structure of these domains throughout a lifetime of experience. Discovering structure can be formalized as probabilistic inference about the organization of entities, and previous work has operationalized learning as selection amongst specific candidate
hypotheses such as rings, trees, chains, grids, etc. defined by graph grammars (Kemp & Tenenbaum, 2008). While this model makes discrete choices from a limited set, humans appear to entertain an unlimited range of hypotheses, many without an obvious grammatical description. In this paper, we approach structure discovery as optimization in a continuous space of all possible structures, while encouraging structures to be sparsely connected. When reasoning about animals and cities, the sparse model achieves performance equivalent to more structured approaches. We also explore a large domain of 1000 concepts with broad semantic coverage and no simple structure
Encoding higher-order structure in visual working memory: A probabilistic model
Full text linkWhen encoding a scene into memory, people store both theoverall gist of the scene and detailed information about a few specific objects. Moreover, they use the gist to guide their choice of which specific objects to remember. However, formal models of change detection, like those used to estimate visual working memory capacity, generally assume people represent no higher-order structure about the display and choose which items to encode at random. We present a probabilistic model of change detection that attempts to bridge this gap by formalizing the encoding of both specific items and higher-order information about simple working memory displays. We show that this model successfully predicts change detection performance for individual displays of patterned dots. More generally, we show that it is necessary for the model to encode higher-order structure in order to accurately predict human performance in the change detection task. This work thus confirms and formalizes the role of higher-order structure in visual working memory
Dynamic infinite relational model for time-varying relational data analysis
Full text linkWe propose a new probabilistic model for analyzing dynamic evolutions of relational data, such as additions, deletions and split & merge, of relation clusters like communities in social networks. Our proposed model abstracts observed timevarying object-object relationships into relationships between object clusters. We extend the infinite Hidden Markov model to follow dynamic and time-sensitive changes in the structure of the relational data and to estimate a number of clusters simultaneously. We show the usefulness of the model through experiments with synthetic and real-world data sets
Social and Discourse Contributions to the Determination of Reference in Cross-Situational Word Learning
Full text linkHow do children infer the meanings of their first words? Even in infant-directed speech, object nouns are often used in complex contexts with many possible referents and in sentences with many other words. Previous work has argued that children can learn word meanings via cross-situational observation of correlations between words and their referents. While cross-situational associations can sometimes be informative, social cues to what a speaker is talking about can provide a powerful shortcut to word meaning. The current study takes steps toward quantifying the informativeness of cues that signal speakers' chosen referent, including their eye-gaze, the position of their hands, and the referents of their previous utterances. We present results based on a hand-annotated corpus of 24 videos of child-caregiver play sessions with children from 6 to 18 months old, which we make available to researchers interested in similar issues. Our analyses suggest that although they can be more useful than cross-situational information in some contexts, social and discourse information must also be combined probabilistically to be effective in determining reference.National Science Foundation (U.S.) (NSF #DDRIG #0746251)United States. Department of Education (Jacob K. Javits Graduate Fellowship
Surprise! Infants consider possible bases of generalization for a single input example
Full text linkInfants have been shown to generalize from a small number of input examples. However, existing studies allow two possible means of generalization. One is via a process of noting similarities shared by several examples. Alternatively, generalization may reflect an implicit desire to explain the input. The latter view suggests that generalization might occur when even a single input example is surprising, given the learner's current model of the domain. To test the possibility that infants are able to generalize based on a single example, we familiarized 9-month-olds with a single three-syllable input example that contained either one surprising feature (syllable repetition, Experiment 1) or two features (repetition and a rare syllable, Experiment 2). In both experiments, infants generalized only to new strings that maintained all of the surprising features from familiarization. This research suggests that surprise can promote very rapid generalization. Infants have been shown to generalize from a small number of input examples. However, existing studies allow two possible means of generalization
Bootstrapping in a language of thought: A formal model of numerical concept learning
No full textIn acquiring number words, children exhibit a qualitative leap in which they transition from understanding a few number words, to possessing a rich system of interrelated numerical concepts. We present a computational framework for understanding this inductive leap as the consequence of statistical inference over a sufficiently powerful representational system. We provide an implemented model that is powerful enough to learn number word meanings and other related conceptual systems from naturalistic data. The model shows that bootstrapping can be made computationally and philosophically well-founded as a theory of number learning. Our approach demonstrates how learners may combine core cognitive operations to build sophisticated representations during the course of development, and how this process explains observed developmental patterns in number word learning.United States. Air Force Office of Scientific Research (Grant FA9550-07-1-0075
Theory learning as stochastic search in the language of thought
Full text linkWe present an algorithmic model for the development of children's intuitive theories within a hierarchical Bayesian framework, where theories are described as sets of logical laws generated by a probabilistic context-free grammar. We contrast our approach with connectionist and other emergentist approaches to modeling cognitive development. While their subsymbolic representations provide a smooth error surface that supports efficient gradient-based learning, our symbolic representations are better suited to capturing children's intuitive theories but give rise to a harder learning problem, which can only be solved by exploratory search. Our algorithm attempts to discover the theory that best explains a set of observed data by performing stochastic search at two levels of abstraction: an outer loop in the space of theories and an inner loop in the space of explanations or models generated by each theory given a particular dataset. We show that this stochastic search is capable of learning appropriate theories in several everyday domains and discuss its dynamics in the context of empirical studies of children's learning.James S. McDonnell Foundation. Causal Learning CollaborativeUnited States. Office of Naval Research (N00014-09-0124)United States. Army Research Office (W911NF-08-1-0242)National Science Foundation (U.S.). Graduate Research Fellowshi
Perceptual multistability as Markov Chain Monte Carlo inference
Full text linkWhile many perceptual and cognitive phenomena are well described in terms of Bayesian inference, the necessary computations are intractable at the scale of real-world tasks, and it remains unclear how the human mind approximates Bayesian computations algorithmically. We explore the proposal that for some tasks, humans use a form of Markov Chain Monte Carlo to approximate the posterior distribution over hidden variables. As a case study, we show how several phenomena of perceptual multistability can be explained as MCMC inference in simple graphical models for low-level vision
One-shot learning by inverting a compositional causal process
Full text linkPeople can learn a new visual class from just one example, yet machine learning algorithms typically require hundreds or thousands of examples to tackle the same problems. Here we present a Hierarchical Bayesian model based on compositionality and causality that can learn a wide range of natural (although simple) visual concepts, generalizing in human-like ways from just one image. We evaluated performance on a challenging one-shot classification task, where our model achieved a human-level error rate while substantially outperforming two deep learning models. We also used a visual Turing test "to show that our model produces human-like performance on other conceptual tasks, including generating new examples and parsing."National Science Foundation (U.S.) (NSF Graduate Research Fellowship)United States. Army Research Office (ARO MURI contract W911NF-08-1-0242
Probing the compositionality of intuitive functions
Full text linkHow do people learn about complex functional structure? Taking inspiration from other areas of cognitive science, we propose that this is accomplished by harnessing compositionality: complex structure is decomposed into simpler building blocks. We formalize this idea within the framework of Bayesian regression using a grammar over Gaussian process kernels. We show that participants prefer compositional over non-compositional function extrapolations, that samples from the human prior over functions are best described by a compositional model, and that people perceive compositional functions as more predictable than their non-compositional but otherwise similar counterparts. We argue that the compositional nature of intuitive functions is consistent with broad principles of human cognition.This work was supported by the Center for Brains, Minds and Machines (CBMM), funded by NSF STC award CCF – 1231216
- …
