Institute of Cognitive and Brain Sciences
University of California at Berkeley
3210 Tolman Hall MC 1650
Berkeley, CA 94720-1650
Administration support for the Institute is provided by the staff of the Helen Wills Neuroscience Institute. See the administration page for help and information.
All talks are in 5101 Tolman Hall, 11am-12:30pm.
Presenter: Mark Bickhard, LeHigh University
Title: Interactivism and Central Nervous System Dynamics
Abstract: There are at least two problems with standard computational/information processing models of brain functioning: 1) the framework is metaphysically incoherent, and 2) what we know about brain functioning is at best anomalous for, if not contradictory to, these models. I will outline a pragmatism based interactivist model of cognition as an alternative to computational models, and a model of central nervous system functioning derived from it. These models are jointly theoretically coherent and comport with such CNS phenomena as volume transmitters, functional astrocytes, non-zero baseline rates of firing, and so on. The central functional relationship in the brain is modulation, not semantic information processing.
Presenter: Tyler Burge, UCLA
Title: Do Apes and Very Young Children Attribute Mental States?
Presenter: Peter Richerson, UC Davis
Title: Cumulative cultural evolution as collective cognition
Abstract: Humans do an amazing variety of very clever things. We are said to be the most intelligent animal. Some writers argue that individual intelligence is responsible for our cleverness. A competing hypothesis is that it is not so much that individuals are intelligent but that populations of individuals collaborating over many generations that are the more important root of our genius. Even moderately complex bits of technology like stone spears were not invented by any one person. Rather, they evolved as component technologies like stone- and wood-working, adhesives and cordage evolved. Once ancient humans became sufficiently skilled social learners to acquire the previous generation's culture sufficiently accurately, the evolution of culture became cumulative. Relatively rare innovative improvements in technology, social organization, and other cultural traditions, could lead, under favorable conditions, to more sophisticated traditions and the diversification of traditions to adapt to local environments. If innovations are the result of costly search and rare luck, accurate social learning can act as a ratchet, preventing the loss of valuable innovations. Much evidence supports this hypothesis including comparative evidence on the sophistication of human social learning, the presence of phylogenetic patterns in human cultures, and the slippage of the ratchet in small populations.
Presenter: Pieter Abbeel, UC Berkeley
Title: Making Robots Learn
Abstract: Programming robots remains notoriously difficult. Equipping robots with the ability to learn would by-pass the need for what often ends up being time-consuming task specific programming. In this talk I will describe the ideas behind two promising types of robot learning: Apprenticeship learning, in which robots learn from human demonstrations. Reinforcement learning, in which robots learn through their own trial and error. I will highlight capabilities enabled by these approaches, such as autonomous helicopter aerobatics, knot-tying, cloth manipulation, basic suturing, and locomotion. I will also discuss some of our latest work in deep reinforcement learning, which I believe has the potential to significantly advance robotics in the foreseeable future.
Presenter: Janet Werker, University of British Columbia
Title: Critical Periods in Speech Perception Development
Abstract: The foundations of language acquisition begin in perception long before infants produce or even understand, their first words. In this talk, I will review the preparation infants have at birth for processing language, and the rapid changes in auditory, visual, and multimodal speech perception that occur in the first months of life as infants begin to acquire the native language. I will then present evidence that, while under typical circumstances the timing of perceptual attunement seems to be constrained by maturation, there are identifiable variations in experiences that can accelerate, slow down, or modify this developmental trajectory. Finally, I will introduce new data that question whether our studies on the timing of plasticity, and indeed on the foundations of language, have considered all the relevant input systems.
Presenter: David Bamman, UC Berkeley
Title: Latent Variable Models of People in Text
Abstract: The written text that we interact with on an everyday basis - news articles, emails, social media, books - is the product of a profoundly social phenomenon with people at its core. People intersect with text in multiple, complex ways: they are the authors of nearly all text we see, they are the audience for whom that text is written, and they are the subjects of that content itself. In this talk, I'll outline two latent variable models for reasoning about the interaction of people and text. In the first, I'll discuss the unsupervised induction of biographical structure from Wikipedia, in which we exploit the correlations of event descriptions in biographies to learn the structure of abstract events, grounded in time, from text alone (along with a systematic bias in the presentation of male and female biographies); in the second, I'll discuss unsupervised models of "personas," which capture patterns of identity and behavior as people are described in text, allowing us to learn a set of fine-grained character types in movies and literary novels.
Presenter: Scott Kelso, Florida Atlantic University
Title: Self-organizing agency
Abstract: The question of agency and directedness in living systems has puzzled philosophers and scientists for centuries. But what principles and mechanisms underlie the emergence of agency? In the last 30 years or so the discovery of emergent phenomena regulated by higher order principles has profoundly inluenced how we understand the organization of matter and its behavior. Might such self-organization somehow give rise to agency? Here, using the concepts, methods and tools of evolutionarily constrained self-organizing dynamical systems (coordination dynamics), a theoretical account of the emergence of agency will be presented. The test field for the theory is experiments and observations on human infants. The birth of agency and its causative powers ("I do", "I can do") is shown to correspond to a eureka-like phase transition in a (nonlinearly) coupled dynamical system whose key variables span the interaction between the organism and its environment. The main mechanism underlying the origin of agency is autocatalytic and involves positive feedback: when the baby’s initially spontaneous movements cause the world to change, their perceived consequences have a sudden and sustained amplifying effect on the baby’s further actions. Some implications of this theory will be discussed.
Presenter: Emma Brunskill, Carnegie Mellon University
Title: Automated Decision Making Under Uncertainty for Societal Benefit
Abstract: The potential societal benefit of autonomous agents is enormous: consider lifelong tutoring systems that help speed learning and provide retraining and career advice, or health care decision support agents. Realizing such systems requires fundamental advances in our understanding of how to automatically make good decisions under uncertainty. In this talk I will discuss our progress in several key challenges including how to leverage past experience to improve future performance on related tasks (transfer learning in sequential decision making tasks), and how much experience is needed to learn to make good decisions (sample efficient reinforcement learning). I will also describe some of the educational applications that have inspired, informed and benefited from our technical discoveries.