WPI Worcester Polytechnic Institute

Computer Science Department
------------------------------------------

AIRG Topics - Fall 2013

Our group meets on Thursdays at 11:00 a.m., FL 246, Beckett Conf. Room.


Dates and topics for this semester are as follows:

Sept 5
AIRG Organizational Meeting (Coordinator: Joe Beck)

Sept 12
CS AI Faculty 1
"AI Research Overview"
    Janice Gobert, Neil Heffernan, Dmitry Berenson, Sonia Chernova

Sept 19
CS AI Faculty 2
"AI Research Overview"
    Carolina Ruiz, Dave Brown, Joe Beck

Sept 26
Location change for this week: Campus Ctr, Morgan rm.
tbd

Oct 3
Location change for this week: FL 141
Hao Wan
"Investigation of Time Series Classification"
    Time series classification methods fall into two categories: global characteristic based methods and local characteristic based methods. In this talk, I will discuss some time series classification methods and their advantages/disadvantages. Then I will introduce my startup project of classifying time series based on both of global and local characteristic, and will provide some initial analysis of our data.

Oct 10
Adam Smith
** Postponed **

Oct 17
Ben Suay
"Learning from Demonstration with Multiple Techniques"
    Learning is an important component for meaningful and efficient human-robot interaction. Previous research has shown that robots that show traits of learning, adaptivity and intelligence are perceived more positively by their human collabo- rators in social environments. Although there are many different learning techniques presented for agents and robots, practicality of these algorithms are not clear in the field. Existing algorithms have different costs and benefits. We are looking to find an answer to the question: Can we interchangeably use outputs of a task learned with fundamentally different algorithms? A general concrete mathematical solution to this question can potentially give robots the possibility to learn the very same task with different modalities, for instance, feedback and direct commands. This new approach could then let robots combine the knowledge they obtained using feedback and direct commands in a complementary manner. To our knowledge this problem has not been investigated for Learning from Demonstration for robots. This short paper presents our research with a highlight of our preliminary results. We believe that our future results will help robots learn and execute a given task with the advantage of different techniques while avoiding disadvantages when possible.

Oct 24
Break

Oct 31
tbd
tbd

Nov 7
Location change for this week: Campus Ctr, Morgan rm.
Adrian Boteanu
    Current recommendation engines attempt to answer the same question: given a user with some activity in the system, which is the next entity, be it a restaurant, a book or a movie, that the user should visit or buy next. The presumption is that the user would favorably review the item being recommended. The goal of our project is to predict how a user would rate an item he/she neverĀ rated, which is a generalization of the task recommendation engines perform. Previous work successfully employs machine learning techniques, particularly statistical methods. However, there are some outlier situations which are more difficult to predict, such as new users. In this paper we present a rating prediction approach targeted for entities for which little prior information exists in the database. We put forward and test a number of hypotheses, exploring recommendations based on nearest neighbor-like methods. We adapt existing common sense topic modeling methods to compute similarity measures between users and then use a relatively small set of key users to predict how the target user will rate a given business. We implemented and tested our system for recommending businesses using the Yelp Academic Dataset. We report initial results for topic-based rating predictions, which perform consistently across a broad range of parameters.

Nov 14
Location change for this week: FL 141
Bo Peng
tbd

Nov 21
Ahmedul Kabir
"Exploratory Analysis of Stroke Patient Data"
    The objective of this research is to perform exploratory analysis that will help build predictive models and recognize important factors related to stroke. The data, collected from nearly one thousand stroke patients, include demographic information, personal and family history of health problems, laboratory test results and recommended treatment. Some of the major goals of the research are to predict patient outcomes after stroke treatment, to explore the different factors that lead to stroke and analyze the effectiveness of different forms of treatment. So far, some common data mining techniques have been applied to the data, their performance has been evaluated, and the resulting models have been analyzed in order to find significant patterns.

Nov 28
Thanksgiving Break (27 Nov - 1 Dec)

Dec 5
tbd
tbd

Dec 12
Andras Kornai
"Digital language death"
    Of the approximately 7,000 languagesĀ spoken today, some 2,500 are generally considered endangered. This consensus figure vastly underestimates the danger of language death in the digital domain, where less than 5% of languages can still possibly survive, and less than 3% are actually likely to. We present evidence of a massive die-off caused by the digital divide, not some future event that could, by some clever policies, be avoided or significantly mitigated -- the deed is already done.

Dec 19
(hold for poss. thesis presentation)

-- Dec 20 -- Last day of Semester


[Return to the WPI Homepage] [Return to the CS Homepage]

AIRG Coordinator / Thu Dec 12 17:15:38 EST 2013