PhD Seminar Series: “Deep Learning beyond prediction: assessing airport identifiability through their delay profiles”

We will continue with our Seminar Series on Tue, 17 May 2022.

On site: Salón de Grados Leganés

Online: https://media.uc3m.es/live/event/62136ff78f420825398b45a1

For the next event in the Aerospace PhD Seminar Series, we will have the pleasure of hosting Massimiliano Zanin researcher at CSIC-IFISC.

The event will take place in the Salón de Grados on Tuesday 17 May at 1pm, and will be streamed.

Massimiliano Zanin received the Ph.D. degree in electrical and computer engineering from the Universidade Nova deLisboa, Portugal, in 2014. He is currently a researcher at CSIC-IFISC, where he is coordinating the ERC StartingGrant ARCTIC. With more than 160 published peer-reviewed contributions in international conferences and journals,he has a good experience in data mining and machine learning research, both the theory and the application, andhas collaborated with scientists from all over the world. His main topics of interest include Data Science and itsapplication to several real-world problems, including the modelling and understanding of air transport or miningcomplex biomedical data sets. He is a member of the editorial team of Nature Scientific Reports, the EuropeanJournal of Social Behavior, PeerJ, PeerJ Computer Science and Chaos Solitons & Fractals.

“Deep Learning beyond prediction: assessing airport identifiability through their delay profiles”

Abstract: 

Deep Learning, i.e. machine learning models not requiring an a priori definition of features, have found extensiveapplication in all fields of science, including air transport and air traffic management. While most research works arefocusing on prediction tasks, we are here going to explore how Deep Learning can be used to make sense ofhistorical data and extract hidden relationships. More specifically, we will ask whether, given time series representingthe observed dynamics of a set of elements, the score of a classification task aimed at distinguishing them can beused as a metric of their dissimilarity – or, in other words, of their respective “identifiability”.
When this approach is applied to vectors representing the profiles of delays at different airport, Deep Learning modelsare able to recognise airports with high precision, thus suggesting that delays dynamics is an endogenous propertyof airports. This identifiability is nevertheless not homogeneous and, on one hand, increases with connectivity, i.e.large and highly connected airports are more unique; and, on the other hand, reduces with geographical proximity,such that near airports tend to share similar profiles. We will discuss how this relates to the general problem of delaypropagation, and some operational implications of this approach.

The talk is based on the following publications:

  • Ivanoska, I., Pastorino, L., & Zanin, M. (2022). Assessing identifiability in airport delay propagation roles through deeplearning classification. IEEE Access.

The seminars will begin at 13 CEST and will take place in the Auditorium Salón de Grados (Padre Soler) campus of Leganés.
No previous registration is required.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s