What: DS Theory Seminar
Date: February 21, 2018
Time: 4:00 – 5:00 P.M.
Location: Computer Science Building, Room 151

Nika Haghtalab
Carnegie Mellon University

“Machine learning by the people, for the people”

Abstract: Typical analysis of learning algorithms considers their outcome in isolation from the effects that they may have on the process that generates the data or the entity that is interested in learning. However, current technological trends mean that people and organizations increasingly interact with learning systems, making it necessary to consider these effects, which fundamentally change the nature of learning and the challenges involved. In this talk, I will explore three lines of research from my work on the theoretical aspects of machine learning and algorithmic economics that account for these interactions: learning optimal policies in game-theoretic settings, without an accurate behavioral model, by interacting with people; managing people’s expertise and resources in data-collection and machine learning; and collaborative learning in a setting where multiple learners interact with each other to discover similar underlying concepts.

Bio: Nika Haghtalab is a Ph.D. candidate at the Computer Science Department of Carnegie Mellon University, co-advised by Avrim Blum and Ariel Procaccia. Her research interests include learning theory and algorithmic economics. She is a recipient of the IBM and Microsoft Research Ph.D. fellowships and the Siebel Scholarship.

The Cognitive Bag Lunch on Weds. Feb. 21 will be held in ILC N400 at 12:05, and will feature Kristen Syrett of Rutgers University (http://rci.rutgers.edu/~syrettk/).  To compensate for asking our colleagues from PBS to make the trek across campus, we will serve pizza (starting at 11:50). Title and abstract follow.

Context sensitivity in adjectives and nominals: Evidence from children and adults

Kristen Syrett

Rutgers University – New Brunswick

Part of what it means to become a proficient speaker of a language is to recognize that the context in which we communicate with each other, including what a speaker’s intentions or goals are, affects the way we arrive at certain interpretations. This seems entirely reasonable for context-dependent expressions like pronouns (they) or relative gradable adjectives (big,expensive), but what about seemingly stable expressions, such as count nouns (fork, ball)? Are words like these—words that appear early in child-directed and child-produced speech—also sensitive to context? In collaborative research with Athulya Aravind (MIT), we have asked precisely this question. We start with a curious yet robust puzzle observed in the developmental psychology literature: young children, when presented with a set of partial and whole objects (like forks) and asked to count or quantify them, appear to treat the partial objects as if they were wholes (Shipley & Shepperson 1990, among others). While children’s non-adult-like behavior may be taken to signal a conceptual shift in development, we adopt a different perspective, entertaining the possibility that children are doing something that adults might indeed be willing to do in certain instances, and that their response patterns reveal something interesting about the context sensitivity of nouns, which we argue is similar to that seen with gradable adjectives. Across three tasks, we show that adults and children are more alike than the previous research has revealed, in that members of both age groups both include partial objects and impose limits on their inclusion in a category, depending on the speaker’s intentions or goals, and the perceptual representation of the object. Furthermore, I draw connections between children’s behavior in this domain to their behavior previously observed in their non-adult-like responses to implicit and explicit comparatives, which still permit an adult-like semantics. Thus, we argue there is conceptual and linguistic continuity in this aspect of development, and that experimental data from both children and adults shed light on the semantics of nominal expressions.

Ashwini Deo of Ohio State University will be presenting a talk Friday at 3:30 in N400. The title and abstract are below. All are welcome!

Title: Alternative circumstances of evaluation and the ser/estar distinction in Spanish

Abstract: The Spanish copulas ser and estar have distributional and interpretational patterns that have resisted an adequate analysis. In this talk, I work towards a unified analysis that treats the two copulas as being presuppositional variants that are differentially sensitive to properties of the circumstances at which the truth of the copular sentence is evaluated. On the proposed analysis,  estar presupposes that the prejacent is boundedly true at the evaluation circumstance. The prejacent’s bounded truth at a circumstance i at a given context of use c depends on two conditions:

(a) there are  no-weaker alternative circumstances i? accessible at c where the prejacent is false.

(b)  i  is  a maximal verifying circumstance at c.

Central to the analysis is the notion of a strength ordering over alternative circumstances of evaluation — a circumstantial  counterpart  to the more familiar  ordering over alternative propositions.  Assuming that this content is conventionally associated with estar  allows for an account of  its distinct flavors and readings with a range of predicates.  ser is shown to be associated with its own inferences that derive from its status as the presuppositionally  weaker, neutral member of the pair.

What: DS Seminar
Date: February 13, 2018
Time: 4:00 – 5:00 P.M.
Location: Computer Science Building, Room 151

Cameron Musco
The Power of Simple Algorithms: From Data Science to Biological Systems 

Abstract:  In recent years, very simple randomized methods, such as stochastic iteration, sampling, and hashing, have become dominant computational tools in large-scale machine learning and data science. In this talk, he will discuss his efforts to understand and harness the remarkable power of these methods.

In particular, he will describe his research on developing simple, but principled, sampling methods for learning, estimation, and optimization. He will present a new class of iterative sampling algorithms, which give state-of-the-art theoretical and empirical performance for regression problems, low-rank matrix approximation, and kernel methods. In many cases, the computational improvement offered by these algorithms is quite surprising. For example, our methods can be used to compute a near-optimal low-rank approximation to any positive semidefinite matrix in sublinear time.

In addition to their power in algorithm design, he will discuss his efforts to understand simple, randomized methods through a different lens: by studying how complex behavior emerges from low-level randomized interactions in biological systems. He will demonstrate how many of the same mathematical tools used to study algorithms in data science can be applied to these systems. As an example, he will highlight his research on noisy estimation and decision making in social insect colonies.

Bio:  Cameron Musco is a fifth year Ph.D. student (graduating spring 2018) in the Theory of Computation Group at MIT.
He is advised by Nancy Lynch and supported by an NSF Graduate Fellowship. He studies algorithms, focusing on applications in data science and machine learning. He often works on randomized methods and algorithms that adapt to streaming and distributed computation. He is also interested in understanding randomized computation and algorithmic robustness by studying computational processes in biological systems.

Before MIT, he studied Computer Science and Applied Mathematics at Yale University and worked as a software developer on the Data Team at Redfin.

Cognitive Brown Bag, 2/14/18, 12:00-1:15, Tobin 521B

Michele Fornaciai & Joonkoo Park (PBS)

Serial dependence in numerosity perception

Attractive serial dependence represents an adaptive change in the representation of sensory information, whereby current stimuli appear more similar to previous ones. Here, we characterize the behavioral and neural signatures of serial dependence in numerosity perception, demonstrating that the perceived numerosity of dot-array stimuli in different numerical ranges is biased by a preceding irrelevant stimulus (“inducer”) in an attractive way. Using electroencephalogram and a passive-viewing paradigm, we show that a neural signature of attractive serial dependence emerges even in the absence of an explicit task early in the visual stream, suggesting that serial dependence has a clear perceptual origin independently from a decision process. With a series of follow-up experiments, we further characterize serial dependence in visual number perception. First, we show that this effect has a weak spatial specificity and a relatively broad tuning for numerosity, and that it has a clear cortical origin (rather than subcortical). Second, we show that the attractive effect is strongly modulated by attention, suggesting the involvement of higher level modulatory influences. Our results collectively suggest that serial dependence results from a cortical neural computation starting from an early level of perceptual processing, possibly subserving perceptual stability and influencing downstream cognitive stages. However, these findings also suggest that the integration of past and present stimuli is in turn modulated by higher-level processes, and potentially amplified at later processing stages.

who: Bharath Hariharan, Cornell University
when: 11:45 A.M. 1:15 P.M., Thursday, February 8th
where: Computer Science Building Rm 150

Abstract: The performance of recognition systems has grown by leaps and bounds these last 5 years. However, modern recognition systems still require thousands of examples per class to train. Furthermore, expanding the capabilities of the system by introducing new visual concepts again requires collecting thousands of examples for the new concept. In contrast, humans are known to quickly learn new visual concepts from as few as 1 example, and indeed require very little labeled data to build their powerful visual systems from scratch. The requirement for large training sets also makes it infeasible to use current machine vision systems for rare or hard-to-annotate visual concepts or new imaging modalities.

I will talk about some of our work on reducing this need for large labeled training sets. I will describe novel loss functions for training convolutional network-based feature representations so that new concepts can be learned from a few examples, and ways of hallucinating additional examples for data-starved classes. I will also discuss our attempt to learn feature representations without any labeled data by leveraging motion-based grouping cues. I will end with a discussion of where we are and thoughts on the way forward.

Cognitive Brown Bag, 2/7/18, 12:00-1:15.

NOTE:  This week only, the talk will be in Bartlett 119

Ibrahim Dahlstrom-Hakki (Landmark College)

Teaching Students with Disabilities Online: Language-Based Challenges and Cognitive Access

Many students with Learning Disabilities (LD), Attention Deficit Hyperactivity Disorder (ADHD), and Autism Spectrum Disorder (ASD) struggle in online learning environments. Online courses tend to place high demands on their language processing and executive function skills. In this NSF funded study (DRL-1420198), we look at some of the barriers facing students with disabilities learning statistics concepts through online discussions. We report on the impact of a Social Presence manipulation on their performance and some of the language-based difficulties involved in assessing their knowledge.