Digital Commons Network^™

Search Tool That Utilizes Scientific Metadata Matched Against User-Entered Parameters, Veronika Margaret Megler, David Maier

Computer Science Faculty Publications and Presentations

A method for providing proximate dataset recommendations can begin with the creation of metadata records corresponding to datasets that represent scientific data by a scientific dataset search tool. The metadata records can conform to a standardized structural definition, and may be hierarchical. Values for the data elements of the metadata records can be contained within the datasets. Metadata records with a value that is proximate to a user-entered search parameter can be identified. A proximity score can be calculated for each identified metadata record. The proximity score can express a relevance of the corresponding dataset to the user-entered search parameters. …

The Grace Programming Language Draft Specification Version 0.3.53, Andrew P. Black, Kim B. Bruce, James Noble

Computer Science Faculty Publications and Presentations

This is a specification of the Grace Programming Language. This specification is notably incomplete and everything is subject to change. In particular, this version does not address the library, nested static type system, module system, metadata, immutable data and pure methods, and other areas. For discussion and rationale, see http://gracelang.org.

It is designed for use by university students learning programming in CS1 and CS2 classes that are based on object-oriented programming, faculty and teaching assistants developing materials for 1st and 2nd year programming classes, programming researchers needing an object-oriented programming language as a research vehicle, and programming designers in search …

Data Near Here: Bringing Relevant Data Closer To Scientists, Veronika M. Megler, David Maier

Computer Science Faculty Publications and Presentations

Large scientific repositories run the risk of losing value as their holdings expand, if it means increased effort for a scientist to locate particular datasets of interest. We discuss the challenges that scientists face in locating relevant data, and present our work in applying Information Retrieval techniques to dataset search, as embodied in the Data Near Here application.

On The Role Of Shape Prototypes In Hierarchical Models Of Vision, Michael David Thomure, Melanie Mitchell, Garrett T. Kenyon

Computer Science Faculty Publications and Presentations

We investigate the role of learned shape-prototypes in an influential family of hierarchical neural-network models of vision. Central to these networks’ design is a dictionary of learned shapes, which are meant to respond to discriminative visual patterns in the input. While higher-level features based on such learned prototypes have been cited as key for viewpointinvariant object-recognition in these models [1], [2], we show that high performance on invariant object-recognition tasks can be obtained by using a simple set of unlearned, “shape-free” features. This behavior is robust to the size of the network. These results call into question the roles of …

The Problem Of Semantics In The Metadata Mess, Veronika Margaret Megler, David Maier

Computer Science Faculty Publications and Presentations

This presentation addresses problems related to the volume of available scientific data, and its accessibility or inaccessibility to researchers who seek it. Topics addressed include metadata and reducing semantic diversity, especially as they refer to geospatial and other architectures

Interpreting Individual Classifications Of Hierarchical Networks, Will Landecker, Michael David Thomure, Luis M.A. Bettencourt, Melanie Mitchell, Garrett T. Kenyon, Steven P. Brumby

Computer Science Faculty Publications and Presentations

Hierarchical networks are known to achieve high classification accuracy on difficult machine-learning tasks. For many applications, a clear explanation of why the data was classified a certain way is just as important as the classification itself. However, the complexity of hierarchical networks makes them ill-suited for existing explanation methods. We propose a new method, contribution propagation, that gives per-instance explanations of a trained network's classifications. We give theoretical foundations for the proposed method, and evaluate its correctness empirically. Finally, we use the resulting explanations to reveal unexpected behavior of networks that achieve high accuracy on visual object-recognition tasks using well-known …