Claudia Soares

Invited Assistant Professor at DEEC/IST
Postdoctoral research fellow
SIPG-Signal and Image Processing Group
Institute for Systems and Robotics
Instituto Superior Tecnico


November 2017.

My two MSc students just graduated, both with grade 19 in 20! Congratulations for your excellent work this year! I was also invited to give a talk on the Optimization group of the Math department of FCT-UNL. It was very interesting and we talked about lots of cool new problems. I just signed my contract as an Invited Assistant Professor at DEEC/IST.

October 2017.

I was invited to present at SIAM IS18. My talk will be about Distributed learning in large scale networks: from GPS-denied localization to MAP inference in a graphical model.

July 2017.

I found some great crowd at IFORS! Stefania Bellavia from Italy, with some very interesting work on SDPs, Natasa Krejic, Natasa Krklec Jerinkic and Dusan Jakovetic doing game-changer distributed optimization from Serbia, Ana Luisa Custodio from FCT, Portugal, with an impressive work on derivative free optimization. What a great conference!

June 2017.

We have an accepted paper at OCEANS’17 at anchorage. It is all about distributed maximum likelihood localization of mobile agents. Amazingly, we beat an extended Kalman filter, with a centralized architecture! A big thanks to my co-authors: Pusheng Ji, João Pedro Gomes and António Pascoal!

May 2017.

I successfully submitted Project DIPLOMAT to the FCT call 02/SAICT/2017 as PI! A huge thank you to my co-PI João Pedro Gomes, and the team which shared this journey with us: João Xavier, Anders Lyhne Christensen, Sancho Oliveira, and António Pascoal. Here's a bit on the abstract: “Smart city sensing, sea exploration, robotics and assisted living - four forces promoting the well-being and development of our society - struggle with a problem triggered by technologies that empowered us with devices capable of computing, communication and environmental sensing. How can we process the data collected by this unprecedented number of agents? For this era of distributed Big Data we need distributed processing, where centralized solutions are recreated in an emergent and scalable way by network agents collecting raw data - not by having them flood the network, but by having them exchange well-crafted messages with a distributed algorithm.”

February 2017.

Our abstract on distributed and robust network localization was accepted at IFORS’17, for the distributed optimization track. The conference will be held in Quebec City.

January 2017.

I submitted a very exciting paper on robust distributed localization for networks of mobile agents, jointly with Joao Pedro Gomes, Beatriz Ferreira and Joao Paulo Costeira. Our algorithm is called LocDyn, works like a filter, depending only on a few past estimates of the positions, and at each time step computes distributedly the MAP estimate of the current positions. Also, I presented the last session of the crash course on scientific writing. The slides for both sessions are gathered here.

November 2016.

I've presented the first of two sessions of a short crash course on scientific writing, based on the book The Writer's Diet by Helen Sword and also on my professional past experience as a copywritter. Thanks to Sabina Zejnilovic for calling my attention to a typo on the presentation.

October 2016.

Thanks to the people in CEMAT for inviting me to present on their Probability and Statistics Seminar. My special thanks to the organizer, Ana Henriques, and Professor Manuel Cabral Morais. I presented some of my work on distributed network localization. We also went through some cool proof sketches. The audience was interested and interesting  —  thanks for all the comments and questions!

Joao Pedro Gomes and I submitted a paper on robust and distributed network localization  —  those dealing with real-world systems know how outliers plague data! We developed a soft outlier rejection estimator and convexified it. We presented two methods for dealing with outlier measurements, one working under an synchronous time model, and another for asynchronous distributed computation. Both are really fast: the sync algorithm is a first order method using Nesterov optimal gradient iterations, and the async one is a (provably convergent) coordinate descent method. Surprisingly, the async algorithm is competitive in the number of communications with the synchronous one, despite the requirements on the network operation are much milder.

things I like

  • jazz

  • literature

  • Cycle Chic and the slow bicycle movement

  • natural languages

  • sewing

hacker emblem