ICTI@CMU Student Research Presentation and Lunch
Date: March 31, 2011 (12:00p – 1:30p)
Place: Newell Simon 3305, Carnegie Mellon University
Agenda
12:30p – 12:50p Weakly Supervised Learning for Tagging Image and Video Collections
 |
Presentation by: Ricardo Cabral (Ph.D. ECE)
Ricardo Cabral is a second year Ph.D. student in the Electrical & Computer Engineering at CMU and IST. His research focuses on Computer Vision and Machine Learning, specifically in object localization and classification, as well as meta-data estimation for images and video. He is also a member of the Human Sensing Lab (http://humansensing.cs.cmu.edu/) and of the PrintART project (http://printart.isr.ist.utl.pt), which aims to provide Artists and Art Historians with Computer Vision tools. Ricardo received his Master’s degree in ECE at IST-Lisbon and a research grant from the Portuguese Science Foundation, in 2009, for work in correspondence methods for structure from motion. |
Abstract: Visual recognition has numerous applications such as autonomous driving, augmented reality, manufacturing and security. Its difficulty essentially arises from large class variabilities of pose and appearances. Most successful approaches have relied on supervised methods, in which the location of the object or action is labeled. However, as we ultimately aim to scale up to a human’s recognition of tens of thousands of categories, this approach becomes problematic: relying solely on manual labeling is both time consuming and prone to subjectiveness. We lessen this burden by exploiting weak labels, denoting the presence of a class in data but not its particular location. While this topic has recently been explored in a mutually exclusive single class setting, several unaddressed issues surface when multiple classes co-occur. We propose a new model for simultaneously localizing different classes in the same media, casting it as an integer optimization problem. Our model’s advantage is twofold: first, we subsume into a single formulation previous single and multi-class localization methods, and get cues to problems not studied before; second, we prove optimal relaxations to the linear domain for some problems, while uncovering the inherent difficulties of others.
12:50p – 1:00p Q & A ICTI Student Research Presentation Luncheon – 31 March 2011 2 of 2
1:00p – 1:20p Architecture-Based Runtime Fault Diagnosis
 |
Presentation by: Paulo Casanova (Ph.D. SE)
Paulo Casanova is a first year Ph.D. student in Software Engineering at CMU and Universidade de Coimbra, Portugal. He has worked in the software IT industry since 2001 as an IT consultant first in Link Consulting and, since 2006, in Novabase. He’s main area of expertise is information system software architecture and design. Lately he has specialized in database and document processing and archival systems. His previous work includes software for telecommunication network organization, public administration workflow software, international organization middleware (in the tourism industry) and banking document archival. His current interest of focus are: heavily loaded, parallel, distributed information processing systems; high availability and reliable software-intensive systems; architectural structures for software information systems. |
Abstract: An important step in achieving robustness to runtime faults is the ability to detect and repair problems when they arise in a running system. Effective fault detection and repair could be greatly enhanced by run-time fault diagnosis and localization. This work describes an approach to runtime fault diagnosis that combines architectural models with spectrum-based reasoning for fault localization (SFL). SFL is a lightweight technique that takes a form of trace abstraction and produces a list (ordered by probability) of likely fault candidates. This technique can be combined with architectural models to support runtime diagnosis that can (a) scale to the size and complexity of modern software systems; (b) accommodate the use of black-box components and proprietary infrastructure for which one has neither a specification nor source code; (c) handle inherent uncertainty about the probable cause of a problem even in the face of transient faults.
1:20p – 1:30p Q & A
1:30p Conclude
