GUPT is a privacy preserving data mining platform. GUPT guarantees that the output of data analyses performed using this platform guarantees privacy as defined by differential privacy. In order to use arbitrary binaries for data analysis in GUPT, you will need to invoke the binary from the GuptComputeProvider object, and possibly pipe IO through some form of IPC mechanism.

A detailed description of the GUPT architecture and mechanisms is deferred to the SIGMOD '12 paper.

1. Enables differentially private data mining on black box programs
2. Parallel implementation ensures high performance. By virtue of the algorithm used internally, your program might run even faster than it does natively (outside of the GUPT Runtime).
3. Research platform with many hooks for evaluating various mechanisms for enabling Differential Privacy

1. You will need Python 2.7+ and the NumPy (http://numpy.scipy.org/) library. Additionally you might also want to install SciPy (http://www.scipy.org/) and scikit-learn (http://scikit-learn.org/stable/) for performing additional machine learning.

2. Open the terminal and execute the following command replacing GUPT-DIRECTORY with the path to where you downloaded GUPT to:

export PYTHONPATH=GUPT-DIRECTORY:$PYTHONPATH

3. The samples folder lists various example scripts. Go into any of those directories and execute the python script.

• Prashanth Mohan (http://www.cs.berkeley.edu/~prmohan)
• Abhradeep Guha Thakurta (https://sites.google.com/site/guhathakurtaabhradeep/)
• Elaine Shi (http://www.eecs.berkeley.edu/~elaines/)
• Dawn Song (http://www.cs.berkeley.edu/~dawnsong/)
• David Culler (http://www.cs.berkeley.edu/~culler/)

When citing the project, please use this bib file