Damiano Varagnolo

alt text 

Senior Lecturer
Control Engineering Group
Luleå University of Technology
Universitetsområdet A2550
97187 Luleå, Sweden
Phone: +46 720 216 047
Email: damiano.varagnolo@ltu.se
Skype: damianovar
RTM: damiano.varagnolo+f049ea@rmilk.com (instructions)
Calendar
Doodle meet-me
Adobe Connect meetings (admin)
zoom meetings
orcid: 0000-0002-4310-7938
GPG public key

Distributed optimization

alt text 

Optimization is the selection of the best element in a given set, where “best” is with respect to an opportune metric. Optimization is distributed when this selection is performed by several separated entities that interact by means of an opportune cooperation protocol. A practical situation where there is a need of distributed optimization technologies is when a set of vehicles need to cooperatively decide whether to form a platoon or not and, in case, the speed of the platoon. In this example to choose a certain speed corresponds to choose a certain cost for each vehicle (generally different, given that everybody has its own way of monetizing time and fuel consumptions). E.g., the figure considers 4 cars, each with its relation speed (x_i) vs. money (f_i(x_i)).

My interests regard specially a particular class of distributed optimization techniques, called Newton-Raphson Consensus. It distributedly mimics the behavior of Newton-Raphson algorithms. The research efforts are towards extending the applicability of this method to more and more general practical scenarios.

Graph discovering

alt text 

Agents of distributed systems may have uncertain knowledge on the communication graph. E.g., agents may not be aware of the diameter of the network (maximum number of hops), or even of how many agents are in the network. This may be induced by the fact that the network evolves in time, i.e., agents join, agents leave, agents move, etc. Unfortunately having knowledge of these quantities is often useful and important. E.g., in distributed optimization methods this information may be used to tune the parameters to achieve higher convergence speed. It is natural thus to implement distributed graph discovering algorithms and thus let agents infer what they need.

My interests regard a particular class of distributed graph discovering techniques, that are based on anonymous aggregation techniques. I.e., they are tailored for situations where privacy matters, and where the estimation must be performed fast. The efforts are thus towards improving the estimation techniques and characterize their theoretical performances.

Distributed Estimation

alt text 

To estimate means to obtain indications on some quantity of interest from some measurements containing some randomness. Distributed estimation algorithms are thus procedures that let agents of a network cooperatively infer what they need. An example is given by wave energy conversion systems, where smart buoys (the agents) may exchange local information to obtain an estimate of the current profile of the waves in the wave farm.

The field of distributed estimation is extremely vast, and can be divided in several subfields. My interests are mainly in nonparametric consensus-based regression strategies and distributed Support Vector Classification algorithms.

Control of Heating, Venting and Air Conditioning systems

alt text 

To improve the efficiency of HVAC systems means to have a more energy-efficient (and thus a cleaner) world. Among the various ways of improving this efficiency, one is to make these systems smarter. E.g., one might make them able to infer when rooms will be occupied or not, and then pre-heat / cool only that rooms that will be used. Notice that there are several ways of achieving these goals, and that the most direct and simple one is to instrument the buildings (with cameras, more temperature sensors, etc.).

My interests are in understanding the limits of using the existing information, i.e., how much “smarter” a building can be made without adding new sensors. An example is how to use the existing measurement systems to construct buildings occupancy models.