We generalize a data summarization matrix to produce one or multiple summaries, which benefits a broader class of models. Our solution works well in popular languages, like R and Python, on a single machine, and shared-nothing architecture in parallel. A key innovation is evaluating a demanding vector-vector multiplication in C++ code, in a simple function call from a high-level programming language. Our method in the R language is faster and simpler than competing big data analytic systems computing the same models, including Spark (using MLlib, calling Scala functions) and a parallel DBMS (computing data summaries with SQL queries calling UDFs)

Graph analytics remains one of the most computationally intensive tasks in bigdata analytics, mainly due to large graph sizes, the structure of the graphs, andpatterns presented in the graphs. We design and optimize graph analysis algorithms on distributed database systems, specifically, analyzing graph metrics and graph properties such as betweenness centrality, transitive closure, triangle counting, etc. We also design and build scalable graph analysis algorithms from scratch on distributed systems using the C++ language.

The rise of platforms such as the Raspberry Pis has allowed testbeds to be built at larger scales and in many places around the world. While these testbeds have made much progress, we observe a general lack of standard and flexible data analysis capability to fully take advantage of the resource provided by these testbeds. As such, we leverage the common model used by these testbeds and leverage existing technologies effective in other domains. We model the network instumentation as streaming data processing; formulate different streaming data processing using SQL; and implement and evaluate a light-weight DB and SQL-based network instrumentation system on an edge device in a distributed architecture.

Given Python's simplicity and popularity in data analysis, it becomes more enticing to use Python. The downside, however, comes from the fact that Python is not as efficient as languages such as C/C++. As such, we look to combine both Python and C/C++ by wrapping more intensive code in C/C++ using Swig to use in Python. We look to take advantage of the many tools provided by Swig to create a seamless integration between C/C++ and Python, detailing an easy way to maintain the wrapped code and employing the C/C++ code using Python's intuitive syntax.

Finding an efficient and error-proof method for isolating factors that significantly cause a measured value to change is an ongoing problem faced by statisticians. Therefore, we are creating a tool that utilizes the exploratory power of OLAP cubes along with statistical tests to create a procedure that produces statistically sound results. The easy-to-use interface allows users to discover significant metric differences between highly similar populations without need for educated guesses. A GUI is implemented to create a user-friendly environment that allows for easy input, visualization of progress, partial result viewing, and pause functionality.

We present a data mining tool, which implements a Bayesian Classifier in SQL, based on class decomposition with the K-means clustering algorithm. The equations required by the Bayesian Classifier are transformed to portable SQL queries. Several query optimizations are introduced which make the classifier more scalable and robust. Since the classifier is based on clustering which takes several iterations to converge, a GUI is used to pause and resume the iterations while building the model. The picture on the right shows the percentage of time allocated for each step for a single iteration.

Customarily hard drives have been used to store documents. However as the speed of networks and the need for secure centralized repositories of information continues to grow, so does the amount of text data that is stored in relational databases, and the need for tools that can process the data inside the database. The main difference between this application and previous research is the hypothesis that the DBMS can be exploited to effectively process documents. . The objective of the application is to implement an information retrieval engine in a DBMS, allowing fast document storage and retrieval with DBMS.

Association rules are a data mining technique used to discover frequent patterns in a data set. Real world application of this technique is very broad and can include fields such as medical and commerce. We can automatically generate the SQL for creating association rules within a relational DBMS. We are focusing on developing several optimizations that will significantly increase the efficiency of item set and rule generation.

We present a system that can analyze high-dimensional data sets stored in a DBMS. Such data sets are challenging to analyze given their very high dimensionality. We focus on  computing descriptive statistics, the correlation matrix and statistical models derived from it. Computations are based on automatically generated SQL queries and User-Defined Functions. Our proposal enables fast statistical analysis  inside the DBMS, avoiding the need to export data sets.

A federated database consists of several loosely integrated databases, where each database may contain hundreds of tables and thousands of columns, interrelated by complex foreign key relationships. In general, there exists a lots of semi-structured data elements outside the database represented by documents and files. Our research involves building a system that creates metadata models for a federated database using a relational database as a central object type and repository. Our goal is to allow quick and easy querying that deals with the relationship between users/files and data within databases