The COMCO 2021 workshop pursues to contribute to the
re-integration of Cognitive Science and Artificial Intelligence. There is a
schism between low- and high-level cognition: a lot is known about the neural
signals underlying basic sensorimotor processes and also a fair bit about the
cognitive processes involved in reasoning, problem solving, and language.
However, explaining how high-level cognition can arise from low-level
mechanisms is a long-standing open problem in Cognitive Science.
In order to bridge this gap, this workshop tackles problems such as grammar
learning, structured representations, and the production of complex behaviors
with neural modeling. With COMCO 2021 we are bringing together
experts studying the mind from a computational point of view to better understand
human and machine intelligence.
Director of the Computational Cognitive Science Lab at Princeton
University. Research focused on developing mathematical models of
higher level cognition and understanding the underlying principles
of our ability to solve computational problems in everyday life.
Co-author of the best-selling book "Algorithms to Live By: the Computer Science of
Human Decisions".
Associated Talk: Understanding human intelligence through human limitations
Victor E. Cameron Professor of Electrical and Computer
Engineering at Rice University, Founding Director of OpenStax, Fellow of the American Academy of Arts and Sciences, National Academy of
Inventors, and American Association for the Advancement of Science,
and recipient of the DOD Vannevar Bush Faculty Fellow Award, the IEEE Signal Processing Society
Technical Achievement Award, and the IEEE James H. Mulligan, Jr. Education Medal.
Research focused in new theory, algorithms, and hardware for sensing, signal processing, and machine
learning.
Group leader at Max Planck Institute for Psycholinguistics,
Principal investigator in the Language and Computation in Neural
Systems Group at Donders Centre for Cognitive Neuroimaging. Research
focused on how language is represented and processed in the mind and
brain.
Associated Talk: Boundary conditions for language in biological and artificial neural systems
Professor at the Italian Institute of Technology heading the
research group "Social Cognition in Human-Robot Interaction",
Editor-in-Chief of International Journal of Social Robotics and
Associate Editor of Frontiers in Psychology, President-elect of
European Society for Cognitive and Affective Neuroscience (ESCAN).
Research focused on human-robot interaction, cognitive and social
neuroscience, and intentional cognition.
X Consortium Assistant Professor at MIT. Research focused on
understanding the computational foundations of efficient language
learning and building general-purpose intelligent systems that can
communicate effictively with humans and learn from human feedback.
Associated Talk: Implicit representations of meaning in neural language models
Professor and researcher in the Integrative Neuroscience and
Cognition Center at the Universite dé Paris. Research focuses
on perception attention and consciousness using cognitive
psychology, EEG, MEG and fMRI.
Associated Talk: Brain dynamics associated with conscious processing
Agenda
Day 1, Thursday, September 23 (CEST)
Welcome
Using robots to study mechanisms of human cognition
Speaker: Agnieszka Wykowska, Professor Affiliation: Italian Institute of Technology, Italy
Robots receive increasingly more attention in scientific areas beyond
robotics. The field of human-robot interaction, for example, focuses not only on developing new
technological solutions for robots, but also on how the human interacts with such artificial entities.
In my research, I use robots as sophisticated stimuli for examining the mechanisms of human cognition.
As such, robots allow for more ecological validity than screen-based stimuli and for excellent
experimental control at the same time.
In this talk, I will present a series of studies that, with the use of a humanoid robot iCub, addressed
specific cognitive mechanisms, such as attention, cognitive control, decision making processes, and theory of mind.
Results of these studies reveal that the social component should be taken into account in models of cognition,
specifically attention and cognitive control. This might also inspire the way in which artificial cognitive architectures are built.
Speaker: Johannes Niediek, Postdoctoral researcher Affiliation: The Hebrew University of Jerusalem, Israel Collaborators: Maciej M Jankowski, Ana Polterovich, Alex Kazakov and Israel Nelken
Speaker: Jasmin Walter, PhD student Affiliation: Osnabrück University, Germany Collaborators: Lucas Essmann, Sabine U. König and Peter König
Lunch break
Brain dynamics associated with conscious processing
Speaker: Claire Sergent, Professor Affiliation: Université de Paris, France
Using experimental psychology and neuroimaging, my team and I
investigate the brain dynamics associated with conscious processing of sensory stimuli. I will present
empirical evidence suggesting that conscious processing might specifically relate to a bifurcation in
global brain activity following the first 200ms of sensory processing. Our results also suggest that,
contrary to the first stages of sensory processing, the onset of these “conscious” mechanisms can be
quite flexible in time. I will discuss how these findings might help update the global neuronal
workspace model of conscious access.
Speaker: Huang Ham, Research Specialist Affiliation: University of Pennsylvania, USA Collaborator: Adrianna Jenkins
Coffee break & chat in Gathertown
Poster session in Gathertown
Deep Network Spline Geometry
Speaker: Richard Baraniuk, Professor Affiliation: Rice University, USA
We study the geometry of deep learning through the lens of approximation theory via spline functions and operators. Our key result is that a large class of DNs can be written as a composition of max-affine spline operators (MASOs), which provide a powerful portal through which to view and analyze their inner workings. For instance, conditioned on the input signal, the output of a MASO DN can be written as a simple affine transformation of the input. This implies that a DN constructs a set of signal-dependent, class-specific templates against which the signal is compared via a simple inner product; we explore the links to the classical theory of optimal classification via matched filters and the effects of data memorization. The spline partition of the input signal space that is implicitly induced by a MASO directly links DNs to the theory of vector quantization (VQ) and K-means clustering, which opens up new geometric avenue to study how DNs organize signals in a hierarchical and multiscale fashion.
Session close of Day 1
Day 2, Friday, September 24 (CEST)
Boundary conditions for language in biological and artificial neural systems
Speaker: Andrea E. Martin, Group Leader/ PI Affiliation: Max Planck Institute for Psycholinguistics, Netherlands
Human language is a fundamental biological signal with computational properties that are markedly different than in other perception-action systems: hierarchical relationships between sounds, words, phrases, and sentences, and the unbounded ability to combine smaller units into larger ones. These and other formal properties have long made language difficult to account for from a biological systems perspective, and within models of cognition. I focus on this foundational puzzle – essentially “what does a system need to represent information that is both algebraic and statistical?” - and discuss the computational requirements, including the role of neural oscillations across time, for what I believe is necessary for a system to represent language. I build on examples from cognitive neuroimaging data and computational simulations, and outline a developing theory that integrates basic insights from linguistics and psycholinguistics with the currency of neural computation, in turn demarcating the boundary conditions for artificial systems making contact with human language.
Speaker: Lucas Castillo, PhD student Affiliation: University of Warwick, UK Collaborators: Pablo León-Villagrá, Nick Chater and Adam Sanborn
Break
Implicit representations of meaning in neural language models
Speaker: Jacob Andreas, Assistant Professor Affiliation: MIT, USA
Neural language models are trained on text corpora to place probability distributions over sequences of words. They produce representations of language that have led to dramatic improvements in downstream tasks as diverse as translation, question answering, and image captioning. Language models' usefulness is partly explained by the fact that they robustly encode aspects of linguistic *structure*, including syntactic categories and dependency relations. But the extent to which language modeling induces representations of *meaning*---and the broader question of whether it is even in principle possible to learn about meaning from text alone---have remained a subject of ongoing debate in NLP and linguistics. I'll describe recent work showing that current neural language models build structured representations of meaning that simulate entities and situations as they evolve throughout a discourse. These representations can be linearly decoded into formal descriptions of semantic state analogous to the "file cards" of Heim (1983) and discourse representation structures of Kamp (1984). They can be directly manipulated to produce predictable changes in generated text, and supervised to improve generation quality. Together, these results suggest that the effectiveness of the modern NLP toolkit stems in part from its ability to learn some aspects of meaning with only linguistic form as training data.
Panel discussion
With Andrea Martin, Jacob Andreas, Agnieszka Wykowska, and Tom Griffiths
Best poster award of 250€ kindly sponsored by halocline
Speaker: Sarah Fabi, Postdoctoral researcher Affiliation: University of Tübingen, Germany Collaborators: Sebastian Otte and Martin V. Butz
Understanding human intelligence through human limitations
Speaker: Tom Griffiths, Professor Affiliation: Princeton University, USA
As machines continue to exceed human performance in a range of tasks, it
is natural to ask how we might think about human intelligence in a future populated by super intelligent
machines. One way to do this is to think about the unique computational problems posed by human lives, and
in particular by our finite computational resources and finite lifespan. Thinking in these terms
highlights two problems: making efficient use of our cognitive resources, and being able to learn from
limited amounts of data. It also sets up a third problem: solving computational problems beyond the
scale of any one individual. I will argue that these three problems pick out the key characteristics of
human intelligence, and highlight some recent progress in understanding how human minds solve them.
Session close & Goodbye
Virtual happy hour in Gathertown
Poster sessions
Dear participants, you are free to prepare your poster
presentation in one of the following ways: 1. The “classical” poster format (a single big slide which you
can zoom in and out of) or 2. multiple separate slides (we recommend no more than 5 slides).