Giulia C. Fanti
Home Research Resume Contact
Current Work

I am currently interested in private media retrieval. Below is a description of the private audio search system I developed with Dr. Matthieu Finiasz and Professor Kannan Ramchandran.

Private audio search demo Short video illustrating the use of our privacy-preserving audio recognition app.

Private face recognition demo Short video illustrating the use of our privacy-preserving facial recognition app.

Private audio search description M.S. Thesis, containing a detailed description of the system.


Here are some brief animations to clarify various aspects of the algorithm. They are intended as supplements to the description above and related reading (e.g. Haitsma and Kalker, "A Highly Robust Audio Fingerprinting System")

Feature generation (originally proposed by Haitsma and Kalker)

Audio search (originally proposed by Haitsma and Kalker). This illustrates how the original, non-private audio search functions. It relies on the feature vectors from the previous animation.

 
 
Prior Work
Uplink Femtocell Modeling, Spring 2012
Course project for EE224B (Wireless Communications) with Kangwook Lee

Current attempts at modeling femtocell capacity are very complex and depend heavily on assumptions about the distribution of users. We formulate a model that is significantly simplified while still capturing salient features of cell behaviour.

Latent Dirichlet Allocation for Geo-Tagging Flickr Photos, Fall 2011
Course project for EE281A (Statistical Learning Theory) with Rashmi K. V.

Significant attention is being paid to the problem of geotagging photos, or determining where they were taken based on content and metadata. We explore an LDA-based approach for estimating the city in which a photograph was taken.

Semidefinite Programming Relaxation for Localization in Non-Line-of-Sight Environments, 2011-2012
Technical Report with V. Ekambaran and K. Ramchandran

Sensor networks often have noisy distance estimates between neighboring entities. We explore an SDP relaxation for resolving these noisy distance estimates into precise location estimates. Specifically, we consider the case in which distance estimates are affected by strong positive bias, e.g. multipath.

Wireless power transfer (2008-2010)
Undergraduate research with Professor J.O. Mur-Miranda at Olin College of Engineering.
  • J.O. Mur-Miranda and G. Fanti, "Peak wireless power transfer using magnetically coupled series resonators", 2010 IEEE International Energy Conference and Exhibition (EnergyCon), (2010)
  • J.O. Mur-Miranda, G. Fanti, Y. Feng, K. Omanakuttan, R. Ongie, A. Setjoadi, and N. Sharpe, "Wireless power transfer using weakly coupled magnetostatic resonators", Energy Conversion Congress and Exposition (ECCE), (2010)