Proposal for Qualifying Exam
نویسندگان
چکیده
Convolutional Neural Networks (CNNs) are a type of deep neural network which have performed well at image and audio classification. One approach to understanding the success of CNNs is Mallat’s scattering transform, which formalizes the observation that the filters learned by a CNN have wavelet-like structure. The resulting transform generates a representation that is approximately translation invariant and slowly varying to a wide class of deformations of the input patterns. These features are then used by a linear classifier to perform classification. In this talk, we will explore the application of the scattering transform and a new variant to understanding the classification of objects using sonar. As we have an explicit model with a fast simulator for this problem, it provides a good test problem for understanding the properties of the scattering transform in the object domain. We obtain over 95% classification accuracy on binary discrimintating shape and speed in the case of synthetic data generated by the fast simulator, and a 95% detection rate for unexploded ordinance, with a false positive of 25% in the case of real data. We construct a new variant of the scattering transform, the shearlet scattering transform (shattering transform), using the shearlet frame instead of Morlet wavelets that are often used in the standard scattering transform. This 2D transform has the advantage of sparsifying the signals containing curvilinear singularities rather than point singularities.
منابع مشابه
Qualifying Exam Syllabus Proposal
Exam Committee Exam Logistics Committee Chairperson: Date: Prof. Monica Vazirani Thursday September 3, 2015 Committee Members: Time: Prof. Dan Romik 11AM-2PM Assoc. Fu Liu Location: Prof. Greg Kuperberg MSB 3106 Prof. Francisco J. Samaniego 1 Proposed Research Talk 1.1 The Catalan numbers and q-analogues. A lattice path is a sequence of North N(0, 1) and East E(1, 0) steps in the first quadrant...
متن کاملQualifying Exam
In order to satisfy qualifying exam requirements adhered to by the School of Informatics and Computing, I will provide an overview of global optimization techniques, specifically simulated annealing, genetic algorithms, and tabu seach.
متن کاملNotes on Numerical Analysis
This set of notes covers topics that most commonly show up on the Numerical Analysis qualifying exam in the Mathematics department at UCLA. Each section covers a specific area of numerical analysis. At the end of each section, we provide a list of key theorems. If the proof of the theorem is directly relevant to the qualifying exam, it will be included in this document. Otherwise, we simply inc...
متن کاملEmergency medicine residents’ attitudes and opinions of in-training exam preparation
PURPOSE Emergency Medicine (EM) residents take the American Board of Emergency Medicine (ABEM) in-training exam, and performance on this exam has been shown to correlate to performance on the ABEM qualifying exam. Though many residencies have in-training exam preparation activities, there is little data on the effectiveness of these efforts. This study aimed to elicit resident perspectives abou...
متن کاملNano-Project Qualifying Exam Process: An Intensified Dialogue between Students and Faculty
An effectively designed examination process goes far beyond revealing students’ knowledge or skills. It also serves as a great teaching and learning tool, incentivizing the students to think more deeply and to connect the dots at a higher level. This extends throughout the entire process: pre-exam preparation, the exam itself, and the post-exam period (the aftermath or, more appropriately, afte...
متن کامل