I am currently a Post-doctoral Researcher at the IBM T.J. Watson Research Center in Yorktown Heights, New York, and an associate in Electrical Engineering at Harvard's John A. Paulson School of Engineering and Applied Sciences. I will join Harvard as a faculty in the Fall of 2017. I received my Ph.D. in Electrical Engineering and Computer Science Department at MIT in the Summer of 2015. My main research interests are information theory, inference and statistics, with applications to security, privacy, machine learning and content distribution.
My CV can be found here (current as of August 2015).
Currently, my research focuses on understanding the complementary relationship between estimation and security/privacy. On the one hand, I investigate the fundamental limits of what can be learned from noisy data, and design practical estimation algorithms that approach these limits. On the other hand, I use the insight gained by characterizing these limits to create and analyze communication schemes that are secure against very powerful adversaries (e.g. eavesdroppers with access to unlimited computational resources and time).
I consider myself a scientist who is an engineer at heart, so I enjoy doing fundamental research that serves as a design driver for practical applications. I have a broad set of interests which include information theory, statistics, communications and optimization. You can find more details on specific research projects below.
What can and cannot be estimated from noisy data? Can we prove that, given a noisy observation of a hidden variable, certain functions of the hidden variable cannot be estimated reliably?
We study these fundamental problems using tools from statistics and information theory. In particular, we derive converse (impossibility) results for estimation based on simple statistics of a hidden and an observed variable, namely maximal correlation and the principal inertia components.
Sample Publications:
Main Collaborators: Mayank Varia, Mark Christiansen, Ken Duffy, Yury Polyanskiy, Yihong Wu.
We live in a time where the availability of computing power and storage is ever increasing, and the amount of data collected is unprecedented. The use and analysis of this data will have a fundamental impact on our society, potentially changing, for example, the way we do healthcare and how we educate our children.
However, with Big Data also comes big responsibility. The increased collection and storage of user data creates new privacy and security risks for the average citizen. The massive and widespread collection of data also demands new security and privacy practices.
In our work, we investigate the tradeoff that exists between privacy, security and utility, and devise privacy-preserving mechanisms that can operate at different points of this tradeoff curve. We also analyze security schemes through the information-theoretic lens in order to characterize their fundamental security properties. This, in turn, allows the design of security mechanisms (such as symmetric-key encryption and key management systems) that provide security guarantees against very strong adversaries.
We also have some exciting work on information-theoretic metrics for security that go beyond perfect-secrecy such as symbol secrecy and guesswork.
Sample Publications:
Main Collaborators: Mayank Varia, Mark Christiansen, Ken Duffy, Nadia Fawaz, Salman Salamatian.
Random linear network coding is a powerful technique that can dramatically improve efficiency in data centers and significantly increase throughput in wireless networks. We investigate and quantify the benefits of network coding in different applications, including distributed content caching, wireless multicast and multi-path TCP.
Publications:
Main Collaborators: Weifei Zeng, Arman Rezaee, Jason Cloud.
Wireless communications are inherently limited by noise and interference. In light of this, what is the maximum transmission rate that can be achieved in through wireless network? Is it possible to abstract the noise and find an equivalent, noiseless network model where the same rates can be achieved?
We investigate these questions using the theory of network equivalence. In particular, we derive a set of tools that can be used to find a noiseless network (i.e. a network where all the links have a fixed transmission rate) that achieves the same set of rates as a given noisy wireless network. This “equivalent” noiseless network can then be used to analyze the capacity of the original network. The procedure is similar to the one in circuit theory: Different components of the network (e.g. a MAC or a broadcast channel) are substituted by a bounding, noiseless model. The resulting noiseless network can then be analyzed using standard combinatorial techniques.
During my Master's at Unicamp, I worked at the physical layer and studied statistical properties of the wireless channel. In particular, I characterized the outage probability in the presence of multiple interferers and fading, and studied the statistical properties of different diversity combining schemes.
Publications:
Main Collaborators: Michel Yacoub, Michelle Effros.
At MIT, I worked as a Teaching Assistant for Principles of Digital Communications (6.450), where I received a 6.6/7 rating by an anonymous student survey. I also worked as a grader for Discrete Stochastic Processes (6.262).
In the Spring of 2014, I was a part of the MIT teaching certificate program for graduate students, a semester long course taught by the MIT Teaching and Learning Laboratory for developing and improving teaching skills.
Jan. 2015
Email: flavio (at) seas (dot) harvard (dot) edu
Office:
IBM T.J. Watson Research Center
Office 07-154B
1101 Kitchawan Road, Route 134
PO Box 218
Yorktown Heights, NY
10598