Syllabus

Spring 2016
Topics In Cognitive Science
CSCI 7772 | EDUC 7775 | LING 7775 | PHIL 7775 | PSYC 7775 | SLHS 7775

Fri 12:00-14:00
Muenzinger D430

Instructor

Professor Michael Mozer
Department of Computer Science
Engineering Center Office Tower 741
Office Hours

Course Objectives

The intent of this course is to expose students to the breadth and depth of current research issues in the field of Cognitive Science. Students will attend presentations of innovative theories and methodologies of Cognitive Science that they will be expected to critically evaluate. Students will participate in the ICS Colloquium Series and also the ICS Distinguished Speakers series that hosts internationally recognized Cognitive Scientists who share and discuss their current research. Following colloquia, students will have the opportunity to engage in analysis and discussion of the work that was presented to further their understanding of the material.

Prerequisites

This course is a requirement for students interested in obtaining either the joint Cognitive Science Ph.D. or a graduate certificate in Cognitive Science.  Others may enroll in the course as space is available.

There are no prerequisite courses. This course is primarily offered for one unit of credit, and students in one of the Cognitive Science academic programs must enroll in this course for two semesters.  Students who have strong need to obtain two units of credit for this course in one semester may potentially do so with the instructor's permission. For each unit of credit, students must attend at least six talks and write commentaries on these talks.

Course requirements

Talk attendance

The primary requirement for the course is to attend and comment on a minimum of 6 (or 12) colloquia in Cognitive Science, including the ICS Distinguished Speaker series. The primary opportunity to attend Cognitive Science colloquia is via the ICS speaker series, the schedule for which can be found at www.colorado.edu/ics/colloquium-schedule.  However, not every talk in the ICS series is approved for this course, and many other talks on campus are approved.  The official list of approved talks for this semester appears on this page.

When I am alerted to talks on campus with significant cognitive science content, but which are not part of the ICS speaker series, I will announce them to the class. If you hear of one that I haven't announced, please forward the time, location, abstract, and title to me, and if I determine it to have significant cognitive science content, I will announce the talk to the class as an option.  If for any reason we have a shortage of talks this semester, I can supplement the ICS schedule by bringing in advanced graduate students and faculty during the ICS colloquium slot.

In some semesters that this course is taught, ICS asks students in this course to sign an attendance sheet.  I do not operate like this.  Instead, the commentary you write (see below) will be the mark of your attendance.  

Background readings

For the ICS colloquia, students are responsible for reading appropriate background materials (e.g., journal articles, book chapters) provided by the speaker to serve as grounding for the colloquium.  These materials will be distributed to students 1-2 weeks prior to the colloquium.

Commentaries

For each talk attended, students are to write a commentary of no more than one page.  For the commentary, I want to get your reactions to the talk.  The commentary can include:
  • a summary of what you think the main or most interesting ideas are behind the work, However, there is no need to provide a complete abstract of the talk. I want to know the student's perspective on what the work had to offer.
  • questions about the material for further discussion, either clarification questions or points of disagreement with the authors (``I don't see how such and such will work as the speaker claims...'').
  • comments on how the talk relates to other talks the student has attended or other papers the student has read.
  • a critique of the work. What are the flaws in the ideas presented? What are the limitations? Did the speaker place their work in the appropriate theoretical context? Did the speaker overstate their results? 
  • If the talk gave you inspiration for new research ideas, describe these ideas.  The ideas might be extensions of the work, new directions to move in, or applications of the work to other subfields of cognitive science.
Think of the commentary as your opportunity to share your thoughts and insights with both me and the entire class. I have created a google group, TopicsInCognitiveScienceSpring2016@googlegroups.com, and your commentaries should be sent to the group for reaction by the professor and other students. To simplify our lives, I'd like the commentaries to be mailed to the group, with a email subject line "[STUDENT LAST NAME] COMMENTARY [SPEAKER]". You are invited to comment on commentaries and to add your reactions to the commentaries, so that we get a group electronic discussion going.

You may find that the number of emails associated with the group is too large. If you wish, you may change your group settings so that you receive weekly summaries of the group discussions or that you log into the group to read the discussions.

I have added everyone who is enrolled in the course to the group using the colorado.edu email address. If you prefer to use a different email address, let me know and I'll update yours.

Commentaries are due within two weeks of the talk date. I've taught this course in the past without including a deadline, and some students delayed submitting commentaries until the semester's end and the discussion group became temporally disjointed. It will benefit our discussions if we're all thinking about the same talks at the same time. I would tighten the deadline even further, but some participants mentioned that they needed a bit of time to digest a talk and read some of the opinions of their classmates before having the confidence to broadcast their views.

All commentaries are due by MAY 1 2016

Grading

To receive an "A" grade in the course, the student must submit 6 (or 12) commentaries that reflect an informed opinion on the research discussed, and student must participate in some electronic discussion of others' commentaries and of the talks in general.  Students will fail the course if they do not submit at least 5 commentaries.

Sanctioned colloquia                NOTE: MORE WILL BE ADDED AS SEMESTER GOES ON

Date and Location Speaker Topic Reading
1/15 @ 10:00, ENVD 201 Paul Maglio, UC Merced Service science and human-centered service systems
1/15 @ 12:00, Muenzinger D428 Marie Banich, Psychology and Neuroscience, CU Drug abuse and the brain

In this talk, I will discuss prior work that I have performed with colleagues at CU Denver on alterations in brain systems found in youth and adults who abuse substances. Following that I will discuss a new national project, the Adolescent Brain Cognitive Development (ABCD) project, which is designed to examine the development of brain systems related to substance abuse, including those responsible for cognitive control and reward processing. This project will also involve collecting a host of other data on cognitive, emotional, and social development. The project will be longitudinal in nature, spanning a 10-year time period, with data collected on over 10,000 individuals at close to 20 sites across the country. CU Boulder has been selected as one of the institutions to be involved in this landmark study, and will be doing so through a joint collaboration between the Institute of Cognitive Science and Institute for Behavioral Genetics. In particular, we, along with three other institutions, will be collecting data on twin pairs to disentangle genetic versus environmental contributions to drug abuse. The on-going design and implications of this study will be discussed.
1/21 @ 10:00, ENVD 201 Scott Robertson, University of Hawaii Political decision making in a mixed media world
1/22 @ 12:00, Muenzinger D428 Christine Brennan, SLHS, CU Phonological grain size sensitivity in auditory cortex is related to reading skill

Previous evidence reveals regions within the auditory cortex selectively activate based on phonological grain size (number of speech sounds presented) (DeWitt and Rauschecker, 2012). Here, we employed direct testing of phonological grain size in order to confirm and further delineate this organization. While reading impairment is associated with deficits in phonological skill, including difficulty isolating phonemes (Shaywitz et al., 1998; Temple et al., 2001), it is unknown if selective activation for grain size is related to reading skill. Since isolation of small grain units is problematic in dyslexia, reading skill may be related to selectivity of the auditory cortex, especially for small grain stimuli. In this study, we examined the relationship between grain size selectivity and reading skill. Using functional magnetic resonance imaging (fMRI), we studied 20 typical adults under different grain size conditions. Stimuli included speech with one, two, or four phonemes. Stimuli were presented in blocks of seven trials from the same condition. During scanning, an active listening task was completed to ensure participants were awake and attending to stimuli. A timed measure of reading skill (TOWRE) was completed outside the scanner (Torgesen et al., 1999). Using SPM8, we contrasted the phonological grain size conditions. We also completed ROI analyses to examined brain-behavior correlations between activation strength and reading skill. Specifically, we compared the difference in activation for the small versus the large conditions with reading skill. We found significantly greater activation in bilateral middle superior temporal gyrus (m-STG) for 1-2 speech sounds (small grain size) and greater activation in the left middle temporal gyrus (MTG), right anterior STG, and right posterior STG for 4 speech sounds (large grain size). Brain-behavior analyses revealed significant correlation of reading skill with activation for the small compared to large grain size condition (p < .01). Adults who had higher reading skill activated the mid-STG more than lower skill readers. In addition, adults with higher skill engaged the left MTG less than lower skill readers for the large compared to small grain size condition, although this trend was not statistically significant. These results further delineate the organization of the temporal cortex, revealing not only selective activation related to phonological grain size, but also that selectivity is related to reading skill. Higher reading skill was associated with stronger engagement of the m-STG, suggesting that sensitivity to single phonemes may be linked to better reading. Importantly, these results provide a link between phonological grain size sensitivity in the brain and reading skill, consistent with behavioral studies that have shown strong correlations between phoneme awareness and reading ability. The findings have implications for developmental dyslexia, a condition often associated with deficits in phonology. Future studies should investigate phonological organization in children with and without dyslexia to determine if differences in grain size representation underlie the disability.
Brennan reading
1/29 @ 12:00, Muenzinger D428 Thomas Hills, University of Warwick Control and Representation in Cognitive Search

The trade-off between exploration and exploitation is a ubiquitous feature of animal life.  Neuromolecular and behavioral evidence from across species suggest the ability to mediate this trade-off originated approximately 700 million years ago in a spatial foraging behavior called area-restricted search, allowing animals to modulate foraging behavior in response to resource density. Further evidence suggests this architecture was later exapted in vertebrates to modulate attention and search in internal representations: to maintain goals in the absence of external stimuli and to look, so to speak, before we leapt.  In this talk, I will present research from my lab investigating internal search using comparisons of computational models that combine quantitative representations of internal environments with control processes that can navigate these environments. The representations are derived from multiple sources, including unsupervised learning from natural language corpora, social networks, and problem representations based on solution similarity.  The control processes consist of random walks and multi-stage models that include dynamic transitions between representations. Over a series of studies, this work suggests that internal search is a form of area-restricted search consistent with the principles of optimal foraging in space and is governed by individual differences in executive control. This approach offers insights into cognitive control not offered by standard proof-of-principle approaches and helps to develop both an evolutionary and process-based account of cognitive control.  I will conclude by presenting applications and questions posed by this work for age-related cognitive decline and changes in lexical representation across the lifespan.
Hills reading
2/5 @ 12:00, Muenzinger D428 Sidney D'Mello, Notre Dame Between Boredom and Bewilderment: Coordinating Feeling and Thinking to Optimize Learning

We study the complex interplay between cognitive and affective states (e.g., confusion, frustration, mind wandering) during learning and leverage insights to develop technologies that  coordinate what learners think and feel in addition to what they know and do. Our basic research investigates how complex mental states arise and influence learning via an analysis of interactions among the learners themselves, the learning content, and the learning activity. We then use signal processing and machine learning techniques to build computational models of mental states from facial features, body movements, peripheral physiology, eye gaze, and contextual cues in a variety of digital learning environments, both in the lab and in the wild. Finally, we close the loop by embedding our models in affect- and attention- aware technologies that increase engagement and learning by dynamically adapting to cognition and emotion. This talk will discuss our theoretical foundations, summarize key findings, and discuss our vision for the future of cyberlearning.

D'Mello reading
2/8 [Monday] @ 15:30, ATLAS 229 (Schnabel board room) Kyle Rector, University of Washington Enhancing Quality of Life for People who are Blind or Low Vision Using Computing Technology

Activities such as exercise and participating in the community enhance quality of life, but they are often not accessible to people who are blind or low vision. For example, people who are visually impaired are more likely to be obese, less likely to attend art museums, and less likely to be employed than people who are sighted. The goal of my research is to design, develop, and evaluate systems that enhance the quality of life for people who are blind or low vision.

In this talk, I will first describe a project with in-depth interviews and surveys that inform the design of future eyes-free exercise technologies. I will then describe a solution that I designed and developed called Eyes-Free Yoga, which is an accessible yoga exergame that provides auditory instructions and feedback. Next I will describe the design and evaluation of Eyes-Free Art, an audio proxemic interface to help explore 2D paintings. Finally, I will discuss the opportunities for developing technologies to enhance quality of life for people who are blind or low vision including exercise in the wild, working with children, and improving experiences with art and community.
2/9 [Tuesday]  @ 12:00, Muenzinger D428 (?) Steven Bethard Parsing the Language of Time

Getting a computer to understand the timeline underlying a written narrative is a critical component of tools for review of patient medical histories, analysis of intelligence reports, and tests of reading comprehension. But human language is rarely explicit in the way that would be most convenient for a computer, and events, times, and temporal relations are often implicit, left to be inferred by the reader. In this talk, I will first present a typical computational methodology for constructing timelines from the explicit and implicit cues of language: a series of supervised machine learning components trained on example texts whose timelines have been annotated manually by humans. Then I will show how we can improve this approach by analyzing big data that has not been annotated by humans but nonetheless reveals patterns in how humans talk about time. Finally, I will present an alternative approach to inferring timelines from text that achieves better generalization through modeling the incremental and compositional nature of the language of time.
Bethard reading
2/10 [Wednesday] @ 14:00, ENVD 201 (the Garage, located on 2nd floor of ENVD through door at top of stairs) Manyan Goel, University of Washington Seattle Teaching Old Sensors New Tricks

A fundamental issue with any new sensing technology is the deployment burden. The new technology often comes with a rigid and perhaps expensive list of pre-requisites; and affects the overall cost-effectiveness, deployability, and accessibility of the solution. For example, a user might find it hard to buy and carry a medical device with them everyday. The primary theme of my Ph.D. research has been to reduce the adoption and deployment barriers of new sensing technologies by using the sensors that are already around us. In this talk, I will present my research on extending the capabilities of the on-board sensors on consumer devices for various applications. First, I will discuss how we can leverage the on-device sensors to enable novel human-computer interactions and make the mobile devices more usable. Next, I will discuss my work on using the on-device sensors for building  health technologies that perform similarly to their clinically-approved counterparts; and how they continue to evolve through deployments in various parts of the world. Finally, I will discuss how my recent work in extending the on-device sensors is shaping my future interests and how I plan to continue to lower the adoption barriers for new sensing technologies."
2/16, 16:00-17:00, Muenzinger E214
John Gabrieli, Professor, Department of Brain and Cognitive Sciences, McGovern Institute for Brain Research, MIT
Prediction as a Humanitarian and Pragmatic Contribution from Human Cognitive Neuroscience

Neuroimaging has greatly enhanced the cognitive neuroscience understanding of the human brain and its variation across individuals (neurodiversity) in both health and disease. Such progress has not yet, however, propelled changes in educational or medical practices that improve people’s lives. We review neuroimaging findings in which initial brain measures (neuromarkers) are correlated with or predict future (1) education, learning, and performance in children and adults; (2) criminality; (3) health-related behaviors; and (4) responses to pharmacological or behavioral treatments. Neuromarkers often provide better predictions (neuroprognosis), alone or in combination with other measures, than traditional behavioral measures. With further advances in study designs and analyses, neuromarkers may offer opportunities to personalize educational and clinical practices that lead to better outcomes for people.
2/19 @ 12:00, Muenzinger D428 Jeremy Reynolds, Senior Data Scientist, Microsoft Stories from Industry: How can Cognitive Science Programs enable students to succeed?
2/23 @ 12:00, ATLAS 206 Dennis Ramirez Making Meaningful Systems and Experiences in Video Games
2/26 @ 12:00, Muenzinger D428 Peter Foltz, ICS Research Professor Automated analyses of language production in clinical tasks

Language provides a window into underlying cognitive structures and mechanisms.  In disorders such as schizophrenia, psychosis and bipolar disorder, abnormalities in language production and comprehension are often used as indicators to aid in the diagnosis and in understanding the underlying etiology.  This talk will describe research in which we apply automated language analysis techniques to detect differences between groups of control and clinical populations.  Over the past 15 years, we have collected samples of language from patients, controls, and unaffected siblings performing a range of tasks in clinical trials and neuropsychological research studies.  The tasks include category fluency, logical memory,  story recall, story telling, and answering open-ended questions.  The analyses use corpus-based statistical models of language to examine semantic and statistical properties of responses. The results indicate that the methods can reliably detect differences in regularities in language between groups.  Implications will be discussed for using the approach as a framework for measuring subtle changes in language and understanding the underlying processes that may cause these changes.   Finally, I’ll talk about applying the approach for remote monitoring and treatment of cognitive functioning on mobile devices.
reading
3/1 @ 3:30, Discovery Learning Center (DLC) 170 Chenhao Tao, Cornell ONLINE SOCIAL INTERACTIONS: A LENS ON HUMANS AND A WORLD FOR HUMANS

Online social interactions have become an integral part of people's lives, e.g., presidential candidates use Facebook and Twitter to engage with the public, programmers rely on Stackoverflow to write code, and various communities have been forming online. This unprecedented amount of social interaction offers tremendous opportunities to understand human behavior. Such an understanding can induce significant social impact, ranging from influencing election outcomes to better communication for everyone.

Tan's research leverages newly available massive datasets of social interactions to understand human behavior and predict human decisions. These results can be used to build or improve socio-technical systems. In this talk, he will explain my research at both micro and macro levels. At the micro level, he investigates the effect of wording in message sharing via natural experiments. He develops a classifier that outperforms humans in predicting which tweet will be retweeted more. At the macro level, Tan will examine how users engage with multiple communities and find that, surprisingly, users continually explore new communities on Reddit. Moreover, their exploration patterns in their early ``life'' can be used to predict whether they will eventually abandon Reddit. He will finish with some discussion of future research directions in understanding human behavior.
3/4 @ 12:00, Muenzinger D428 Frank Jaekel, Universität Osnabrück Categorization: From Psychology to Machine Learning and Back
 
Abstract: The ability to categorize is fundamental for cognition—in humans and in machines. Many, if not all, so-called higher cognitive functions, like language or problem-solving, crucially depend on categorization. For this reason, research on categorization plays a central role in cognitive science and artificial intelligence alike. Hence, it is no surprise that many successful machine learning algorithms for categorization were inspired by results and insights from psychology and neuroscience. However, today machine learning is a mature field and more recent methods are usually seen to be grounded in statistics and computer science rather than in cognitive science. Kernel methods, in particular, have gained popularity in machine learning and have proved to be successful in many applied categorization problems. I will describe how similar ideas have developed in psychology and how insights from machine learning can feed back into cognitive science.
3/11 @ 12:00, Muenzinger D428 Joshua Correll, Psychology and Neuroscience, CU Of Kith and Kin: Perceptual Enrichment, Expectancy, and Reciprocal Processing in Face Perception

paper1
paper2
3/11 @ 19:30, Math 100 (building just west of the Engineering center) Stuart Russell, UC Berkeley The long-term future of (Artificial) Intelligence

The news media in recent months have been full of dire warnings about the risk that AI poses to the human race, coming from well-known figures such as Stephen Hawking, Elon Musk and Bill Gates. Should we be concerned? If so, what can we do about it? While some in the mainstream AI community dismiss these concerns, UC Berkeley’s Stuart Russell will argue instead that a fundamental reorientation of the field is required.
3/18 @ 12:00, Muenzinger D428 John Trueswell, University of Pennsylvania The role of cognitive flexibility in language processing and language acquisition

Because children and adults interpret speech in real-time, rapidly making commitments to interpretation essentially on a word-by-word basis, they must learn to deal flexibly with temporary ambiguities that arise in the input. In this talk, I'll present a series of experiments that examine the relationship between language processing, language learning, and cognitive flexibility. It is found that individual differences in how well children flexibly respond to representational conflict during executive function tasks predict how well they deal with temporary syntactic ambiguity during real-time comprehension. I explore how the processing challenges associated with real-time comprehension constrain grammar learning, and might even constrain the types of grammars that arise in languages of the world. The results reveal some of the ways in which the cognitive abilities of the individual shape the linguistic-communicative system of the group.
4/1 @ 12:00, Muenzinger D428 Derek Lomas, UCSD Large-Scale Online Experiments Can Accelerate the Production of Interaction Design Theory

Each day, companies run thousands of design experiments to optimize software designs (e.g., A/B tests). What if just a fraction of these experiments were used to test generalizable theories about how and why interaction designs affect user behavior? How would this impact the field of interaction design?

To illustrate the opportunities for online theory testing, I will share several experiments investigating a classic theory in game design, that games should be neither too hard or too easy. This has been formalized as the hypothesis that “moderate difficulty will produce optimal motivation”. To test this hypothesis, I deployed thousands of variations of an online educational game to >50,000 users. Surprisingly, the results showed that low levels of difficulty consistently produced maximum motivation. Further experiments indicated that the factor of "novelty" plays a critical and rarely recognized role in maintaining player motivation. These findings show how online experiments can be used to build generalizable theories about how designs impact users.


More broadly, these results illustrate how the proliferation of large-scale online theoretical experiments could rapidly produce a large body of empirically-validated interaction design theory. How can we prepare for a "big science" of interaction design? To explore this, I share some additional experimental work with "multi-armed bandits", which suggest how artificial intelligence might participate in the future of large-scale scientific inquiry. I conclude by discussing several of the opportunities, limitations and risks of scientific theory-making in the field of design.

4/8 @ 12:00, Muenzinger D428 Anu Sharma, SLHS, CU Cross-modal brain changes in hearing loss across the age spectrum
 
A basic tenet of neuroplasticity is that the brain will re-organize following sensory deprivation.  Auditory deprivation appears to tax the brain by changing its normal resource allocation.  Compensation for the deleterious effects of hearing loss may include recruitment of alternative or additional brain networks to perform auditory tasks. Our high-density EEG experiments suggest that age-related hearing loss results in significant changes in neural resource allocation, reflecting patterns of increased listening effort, decreased cognitive reserve, which may be associated with dementia-related cognitive decline.  Cross-modal plasticity is another form of cortical re-organization associated with deafness. Cross-modal plasticity occurs when an intact sensory modality recruits cortical resources from a deprived sensory modality to increase its processing capabilities as compensation for the effects of sensory deprivation. Our results suggest evidence of recruitment of higher-order auditory cortical areas by visual and somatosensory modalities in hearing loss and deafness.  Cross-modal cortical re-organization is evident both in congenital deafness and in age-related mild-moderate hearing loss and shows a strong negative correlation with speech perception performance. Overall, our results suggest that compensatory cortical plasticity secondary to sensory deprivation has important neurological consequences and influences outcomes in children and adults with hearing loss.
reading
4/22 @ 12:00, Muenzinger D428 Bob Sturm, Queen Mary University of London Clever Hans, Clever Algorithms: Are your machine learnings learning what you think?

In machine learning, generalisation is the aim, and overfitting is the bane; but just because one avoids the latter does not guarantee the former. Of particular importance in some applications of machine learning is the “sanity" of the models learnt. In this talk I discuss one discipline in which model sanity is essential -- machine music listening — and how several hundreds of research publications may have unknowingly built, tuned, tested, compared and advertised “horses” instead of solutions. The true cautionary tale of the horse-genius Clever Hans provides the most appropriate illustration, but also ways forward.

Reference: B. L. Sturm, “A simple method to determine if a music information retrieval system is a “horse”,” IEEE Trans. Multimedia, vol. 16, no. 6, pp. 1636-1644, 2014.

Additional information for students (click to read)