The Dual PhD Adventure: Great Synergies From Both Sides

Dual Degree PhD :: Filipe Condessa The Dual PhD Adventure: Great Synergies From Both Sides

After five years of doctoral studies in the scope of CMU Portugal Program, the Mikio Takagi best student paper award at the International Geoscience and Remote Sensing Symposium (IGARSS 2015) and a thesis titled – “Robust Image Classification with Context and Rejection”, the Azorean researcher Filipe Condessa successfully concludes his PhD. Congratulations for this great achievement. When asked about this long journey, Filipe Condessa highlights the synergies created with both sides – Portugal and US – and the way this dual degree influenced his studies, professional career and also at a personal level. To know more about Filipe Condessas’s experience read the following interview.

CMU Portugal: What are your research interests?
Filipe Condessa: My main research interests are in the area of robust classification.The core idea of robust classification is to adapt the behavior of classification systems to adapt on problems where errors may appear, thus avoiding errors.

Our work in robust classification was applied on the frontier area of ill-posed image classification problems, with applications both in medical image classification and in remote sensing.

CMU Portugal : How do you describe your PhD experience? Filipe Condessa: My PhD experience was overall very positive! I have met interesting professors and colleagues from both sides of the Atlantic. There were clear synergies from both sides, and I believe I would have not been able to produce such interesting work just by staying in one side of the Atlantic for the five years.

CMU Portugal: What are the main aspects/differences of the CMU Portugal Partnership?
Filipe Condessa: I think the main aspect of the CMU Portugal Partnership is the contact with two different cultures.
Usually, one adapts to the surrounding environment, often without much choice to the adaptation.
A PhD student at IST will adopt many traits associated with IST, some better and some worse, and a PhD student at CMU will adopt many traits associated with CMU, some better and some worse.
By being in contact with two different work cultures (in Portugal and at CMU), I believe I was able to develop a more balanced work culture, trying to capture the best from each side.
On the other hand, there is also the CMU Portugal superstructure supporting continuously the students. This does not happen in one-sided PhD programs in Portugal, the students are left to their own devices to communicate with the funding institution.

CMU Portugal: What were your main achievements? Filipe Condessa: In terms of the achievements of our work, I think we developed a very interesting framework to deal with uncertainty in classification and that we set the corner stone for a significant amount of future work related with cost-efficient improvement of classification performance (at the cost of being able to deal with uncertainty instead of just spending more resources on improving the classifiers). On a personal level, I would say my three main achievements during the PhD program were: (1) being admitted to the PhD program; (2) concluding the PhD program; and (3) being awarded the best student paper award at an Institute of Electrical and Electronics Engineers (IEEE) flagship conference while demonstrating the benefits of the use of robust classification in remote sensing.

CMU Portugal: Would you recommend this opportunity to other students?
Filipe Condessa: I would, I did, and I will keep recommending!

CMU Portugal: Would you like to share with us what your future professional plans will be or if you have any professional offer?
Filipe Condessa: I will be starting, in August 2016, a post-doctoral degree in CMU, advised by Professor Radu Marculescu.
The work will focus on network dynamics. There is a wide area of applications ranging from the now ever so prevalent social networks to biological networks. The core idea is to figure out how can these complex and massive networks be controlled.
This work will have a duration of two years.

PhD Thesis: “Robust Image Classification with Context and Rejection”

Abstract
Classifications systems are ubiquitous; despite efforts going into training and feature selection, misclassifications occur and their effects can be critical. This is particularly true in classification problems where overlapping classes, small or incomplete training sets, and unknown classes occur. In this thesis, we mitigate misclassifications and their effects by adapting the behavior of the classifier on samples with high potential for misclassification through the use of robust classification schemes that combine context and rejection. We thus combine the advantages of using contextual priors in classification with those of classification with rejection. In classification with rejection, we are able to improve classification performance at the expense of not classifying the entire data set. We thus add the following tools to the robust classification toolbox: 1) we derive performance measures for evaluating of classifiers with rejection; 2) we create a family of convex algorithms, SegSALSA, to classify with context; 3) we design architectures for robust classification with context and rejection that encompass interactions between context and rejection. We validate our approach on two different real-world data sets: histopathological and hyperspectral images.