Pedro Herruzo

Rule detection and generalization in deep learning for spatiotemporal data:
- Video prediction.
- Building world-class datasets for weather/traffic predictions for ML deployment.
- Data and task-driven design of benchmark metrics.

Co-organizer of:
- the Weather4cast Competition (IEEE BigData Cup 2021).
- the Traffic4cast Competition (NeurIPS 2020 & 2021).
- 1st Workshop on Complex Data Challenges in Earth Observation @CIKM 2021 .

Leading Machine Learning related tasks in projects related to future urban mobility:
- Mobility Advisor (presented in MWC 19 together with IBM)
- Socially Responsible Navigation (presented in SCEWC 18 together with Waze and the city council of Barcelona)

Teaching laboratory sessions of "Introduction to Programming", 1st-year subject with Python.

Teaching laboratory sessions of Computer Engineering:
- Data Structures: 1st-year subject programming with C++
- Operating systems I: 2nd-year subject programming with C

Graduate Student Researcher at the CVUB group (http://www.ub.edu/cvub/) under the supervision of professor Petia Radeva: Deep Learning methodologies applied to Egocentric images.

Exploitation of Data Warehouse Commercial Segurcaixa, Vidacaixa: Gathering requirements, functional and technical definition, designing and developing ETL processes with SAP BO, PowerCenter, Oracle and Control-M.

The IARAI Traffic4cast competition at NeurIPS 2019 showed that neural networks can successfully predict future traffic conditions 15 minutes into the future on simply...

Deep Neural Networks models are state-of-the-art solutions in accurately forecasting future video frames in a movie. A successful video prediction model needs...

The increasing complexity of mobility plus the growing population in cities, together with the importance of privacy when sharing data from vehicles or any device...

The objective description of lifestyle patterns from egocentric images captured by wearable cameras is considered the next step on health-tracking applications...

First-person stories can be analyzed by means of egocentric pictures acquired throughout the whole active day with wearable cameras. This manuscript...

Nowadays, we can find several diseases related to the unhealthy diet habits of the population, such as diabetes, obesity, anemia, bulimia, and anorexia. In many...

Brief explanation at ICML 2021 of the Weather4cast competition, a benchmark for satellite imagery multi-channel weather prediction using machine learning (YouTube):

Giving the talk of our paper "Recurrent Autoencoder with Skip Connections and Exogenous Variables for Traffic Forecasting", accepted as an outstanding contribution to the Competition Track 'Traffic4cast' in NeurIPS 2019 (YouTube, SlidesLive min 44:34, english):

Giving the talk of our paper "Towards Objective Description of Eating, Socializing and Sedentary Lifestyle Patterns in Egocentric Images", accepted in BMVC 2019.

Avichai Bakst and me talking about the partnership of Waze, SEAT S.A., and the city council of Barcelona under the Connected Citizens Program of Waze (YouTube, english):

In picture,
1) from SEAT S.A.: i) Luca de Meo (president), ii) Jordi Caus (head of future urban concepts), iii) myself explaining the project of my thesis in SEAT S.A., and UPC.
2) from the city council of Barcelona: i) Roger Torrent (president of the Parliament of Catalonia), ii) Quim Torra (president of the Government of Catalonia), iii) Ada Colau (mayor of the city of Barcelona).

You can find more information about it in news [fleetnews, SEAT, S.A.], and a visual explanation in the following YouTube video:

Giving the talk of the project, a collaboration of SEAT S.A., Waze, the city council of Barcelona and the Polytechnic University of Catalonia.

This congress named our work the best master thesis out of all presented manuscripts. The live presentation can be found in the following link to YouTube (spanish):

In this round-table, I was defending the importance of public-private cooperation in order to improve urban mobility through a common data-layer in the city called Open Data. Link to the slides.

In this talk, I was explaining relevant projects in SEAT S.A. in the transition from being only a car manufacturer towards a mobility provider. Link to related news.

Traffic4cast is a challenge inside the NeurIPS 2019 conference organized by IARAI. The objective is about predicting traffic states (average speed, number of vehicles, and main traffic direction) in three different cities: Moscow, Istanbul, and Berlin with data provided by Here.

Three best scores in the leaderboard got a cash prize, but the jury also selected only 4 more projects they believe had an outstanding contribution, with a reward of a free entrance to NeurIPS 2019 conference, and the publication of an extended abstract. This is what we won.

Links:

• Competition website.
• paper/code website.

The objective of this challenge is to model the weather in the city of Barcelona in the best possible way. Using temperature, precipitation, humidity and other variables in a set of 4 stations in the city, we were asked to predict the real average temperature in the city, together with the best location for extra stations that could benefit the measurements in real life. We also presented different creative use cases that could benefit from installing the proposed physical stations. Link to the website and instagram of the challenge.

This congress named our work the best master thesis out of all presented manuscripts. In our work, we took one step further in the direction of self-quantify lifestyle patterns, analyzing first-person stories by means of egocentric pictures acquired throughout the whole active day with wearable cameras.
Our work use AI in order to quantify when a person is socializing with other people or alone, when is sitting or standing, or when she or he is buying grocery, cooking, eating or neither of that. We introduce the first prototype of a website where the user can upload the pictures of the whole day and get a dashboard with these measurements.

Links:

• Live presentation on Youtube (spanish).
• Conference slides.
• Our data preprocessing code.
• Our model code.
• In news.

We participated in a very demanding contest for startup ideas, and we got recognized together with the winners for a second prize. We presented an App that can create nutrition diaries automatically from only pictures, and propose healthy eating habits from it. Links to related news [1], [2], and [3].

Strengthen African Machine Learning

Machine Learning, Deep Learning, Bayesian Statistics, Optimization, Numerical Linear Algebra, Complex Networks, Probabilistic Graphical Models...

Neural Networks, Computer Graphics, Algebraic Structures, and Geometry and Topology. Bachelor thesis with title: Can a CNN recognize Mediterranean food?