Brandeis Marshall

Scientific workflow management systems and workflow patterns

Scientific workflow management systems primarily consist of data flow oriented execution models, and consequently, these systems provide a limited number of control flow constructs that are represented in dissimilar ways across different scientific workflow systems. This is a problem, since the exploratory nature of scientific analysis requires the workflows to dynamically adapt to external events and control execution of different workflow components. Hence some degree of control flow is necessary. The lack of standard specifications for specifying control flow constructs in scientific workflow management systems leads to workflows designed using custom developed components with almost no reusability. In this paper, we present a standard set of control flow constructs for scientific workflow management systems using workflow patterns. Firstly we compare the control flow constructs present in three scientific workflow management systems: Kepler, Taverna and Triana. Secondly these patterns are implemented in the form of a template library in Kepler. Finally, we demonstrate the use of this template library to design scientific workflows.

Using the Social Networking Graph for Image Organization

The process of record-keeping and maintenance of digital photographs is an emerging concern. In this paper, we present a tag organization framework for photographs by clustering similar content based photographs. Using a social network graph model, we relate objects and the corresponding relationships amongst them. Each photograph is represented in our framework with five attributes of objects, place, occasion, time and associations. Through experimentation of the MIRFLICKR-25000 image collection, we identity and examine the object popularity and object closeness properties.

Maximizing network security given a limited budget

In order to safeguard an organizations networked assets, a network administrator must decide how to harden the network. To aid the decision-making process, network administrators may use attack graphs, which, through analysis, yield network hardening suggestions. A critical drawback of currently available analyses is the lack of consideration for the network administrators defense budget. We overcome this shortcoming by modeling the problem of choosing security measures given a finite budget as a combinatorial optimization problem. We call this problem the Security Measures Choosing Problem (SMCP). Dynamic programming is used to provide optimal solutions.

Context seeking with social tags

Users intend for their image tags to be helpful in image searching and browsing. However, image tags have added a new layer of complexity, while the evaluation measures remain primarily binary in nature. Obviously, tags are related to each other, but we need to form models that measures the bias between tags.

Taking the Tags with You: Digital Photograph Provenance

Prior to digital photography, the back of the glossy photo paper served as the canvas to write the important photo details. The photos were then transferred to a series of photo albums. The general public creates too many pictures to sustain the physical photo album. One byproduct of taking digital photos is the automatic number scheme embedded in the digital device. The semantic context of any photo can not be represented with this automatic labeling method. Hence, the user is responsible for concisely renaming the digital images. The rise of digital pictures creates a need for more reliant on storage technology for longevity. We present a suite of annotation tags that capture provenance details. We classify consumer photography provenance in terms of photo manipulation, photo clustering and photo semantics.

Applying Aggregation Concepts for Image Search

Through the influx of information content on the Internet, a number of image search methodologies have been presented and implemented to increase the accuracy of image retrieval including keywords, object classification and feature processing. Both keyword and object classification models rely heavily on human subjects, which is time-consuming and error-prone with inconsistency in word agreement. We propose two feature processing methods without human intervention. The feature collage algorithm compares images based on particular features such as color histogram whereas the feature independent algorithm considers each features dimension as independent contributors to the image quality. Using query-by-example, we organize images using rank aggregation methods, previously applied in text information retrieval. We show through empirical experimentation the benefits of our feature processing algorithms over traditional CBIR approaches.

A rank algebra to support multimedia mining applications

Ordering objects with respect to their various relevant properties before and during processing is a basic step in many multimedia mining problems. Examples include mining frequent patterns in sensory data and mining popularity orders in digital television. Designing multimedia and multi-modal mining techniques for complex and adaptive systems, requires the capability of dealing with rankings of diverse collection of inputs and outputs of a complex mining task, in a uniform, declarative manner. In this paper, we present a model and algebra which treat ranks of the media as first class objects to support complex mining tasks. We model each mining task as an algebraic combination of multiple subtasks, thus providing a declarative framework in which the ranked results returned by individual subtasks are combined under appropriate semantics. We also present a novel order distance function, which enables partitioning and aggregation support for mining.

A classification algorithm for finding the optimal rank aggregation method

In this paper, we develop a classification algorithm for finding the optimal rank aggregation algorithm. The input features for the classification are measures of noise and misinformation in the rankers. The optimal ranking algorithm varies greatly with respect to these two factors. We develop two measures to compute noise and misinformation: cluster quality and rank variance. Further, we develop a cost based decision method to find the least risky aggregator for a new set of ranked lists and show that this decision method outperforms any static rank aggregation method by through rigorous experimentation.

Kernel Level Support for Workflow Patterns

In the evolution of computing technology over the decades, file system capabilities have not grown in tandem to processing power. Today, scientific computing is highly data intensive and relies on workflows. Workflows developed are not portable among workflow management system. Also, scientific computation that rely on inherent workflows do not have a kernel support for workflows, and are executed essentially in a batch processing model. A file system that includes native kernel functionalities to support workflow execution would address the issue of parallel processing as well as portability. Such a file system would improve scientific computing performance. This paper describes an approach we developed to add workflow functionality to the Linux kernel and native file system to help simplify the use of workflow management systems for scientific computing.

Aggregating music recommendation Web APIs by artist

Through user accounts, music recommendations are refined by user-supplied genres and artists preferences. Music recommendation is further complicated by multiple genre artists, artist collaborations and artist similarity identification. We focus primarily on artist similarity in which we propose a rank fusion solution. We aggregate the most similar artist ranking from Idiomag, Last.fm and Echo Nest. Through an experimental evaluation of 300 artist queries, we compare five rank fusion algorithms and how each fusion method could impact the retrieval of established, new or cross-genre music artists.