Carnegie Mellon Portugal Doctoral Students Give a Talk at Carnegie Mellon

Carnegie Mellon Portugal Doctoral Students Give a Talk at Carnegie Mellon
ICTI Student Research Presentation Luncheon [pdf]
Date: Wednesday, November 16, 2011
Place: Singleton Room, Roberts Hall
Time: 12 pm – 1:30pm

12:00 pm Buffet opens (Singleton Room, Roberts Hall)

12:30 pm Presentation 1: ” Effective Vision Processing for Specific Recognition Problems in Two Robot Platforms”

Susana Brandao By: Susana Brandão, a third year PhD student in the Electrical and Computer Engineering in the scope of the Carnegie Mellon Portugal Program. She finished her undergraduate studies in Physics Engineering from Instituto Superior Técnico – Universidade Técnica de Lisboa (IST-UTL) in 2005. For the following four years she held different positions in various industries: scholarship recipient for scientific research in astrophysics at ISTUTL; intern at Deimos Engenharia, a space engineering company; associate consultant at Boston Consulting Group and a volunteer at Cores do Globo, a fair trade association in Lisbon. In 2009, she started her dual degree Ph.D. in the Carnegie Mellon Portugal program at IST-UTL and CMU, and her research interest is focused in vision for autonomous mobile robots.
Abstract: In this presentation I will address concrete challenges presented to computer vision in autonomous mobile robots by exploring examples of object detection and/or recognition in two different robot platforms. For a mobile robot, vision acts as a sensor required for control, hence with specific real-time and accuracy performance requirements. The motion of the robot creates additional challenges as images are noisy and blurred, and the background is not easily segmented from the points of interest for the robot task. Two examples of autonomous robots that require vision as sensors are the humanoid NAO robot, as a soccer robot in the RoboCup Standard Platform League, and the cooperative CoBot service robot, as it navigates around the Gates Hillman Center (GHC). I present our solutions for an effective object detection for the humanoid NAO soccer robot, which combines offline training with fast online processing. I then focus on the CoBot’s need to take the elevators in GHC, and address the problem of camera control and detection and classification of the floor numbers inside an elevator. Both examples are marked by the need to create real-time solutions for object recognition problems in autonomous mobile robots.

12:50 pm Q & A

1:00 pm Presentation 2: “Supervised Topical Key Phrase Extraction of News Stories using Crowdsourcing, Light Filtering and Co- reference Normalization”

By: Luís Marujo, a third year dual degree Ph.D. student in Carnegie Mellon’s School of Computer Science and Instituto Superior Técnico of the Universidade Técnica de Lisboa (IST/UTL), Portugal. He holds C.A.S (2011), MSc. (2009), and BSc. (2007) in Computer Science and Engineering from IST. He was awarded the best poster award at the S3MR 2011 (2nd Summer School on Social Media Retrieval). His research interests include Recommendation Systems, Natural Language Processing, Information Extraction, Information Retrieval, Social Media and Machine Learning. Luís Marujo
Abstract: Fast and effective automated indexing is a critical problem for personalized online news aggregation systems, such as News360, Google News, and Yahoo! News. Key phrases that consist of one or more words and represent the main concepts of the document are often used for the purpose of indexing. The accuracy of current state of the art automated key-phrase extraction systems (AKE) is in the 30-50% range. This makes improvements in AKE an urgent problem. In this work, we followed a fairly traditional approach of training a classifier to select an ordered list of the most likely candidates for key phrases in a given document. We augmented the process with new features, e.g.: the use of signal words, freebase categories, etc. We have also experimented 2 forms of document pre-processing that we call light filtering and co-reference normalization. Light filtering removes sentences from the document, which are judged peripheral to its main content. Co-reference normalization unifies several written forms of the same named entity into a unique form. Finally, we used Amazon’s Mechanical Turk (Mturk) service to label documents for training and testing.

1:20 pm Q & A

1:30 pm Conclude

Please RSVP At:
By Monday, November 14

For more information, please contact Nicole Hillard-Hudson, – 412-268-1728