Antonio García-Díaz is a computer engineer, specialised in the fields of computational intelligence (artificial intelligence and machine learning), image processing and analysis, and 3D graphics. Currently he is undertaking his fifth year of PhD studies at the IRIDIA-CoDE department in the Ecole polytechnique de Bruxelles (EPB) from the Université libre de Bruxelles (ULB), Brussels (Belgium), where he obtained his Master's degree in 2018.
For his PhD thesis, "Self-Optimisation of Neural Network Architectures", Antonio is researching, producing and coding novel neural architecture search (NAS) algorithms. These algorithms can automatically generate an artificial neural network architecture that is optimal for a given task: both as simple as possible for the task (in terms of precision) and as small and simple as possible in its scheme. Such networks can be directly embedded into small or edge devices (such as drones, smartphones or tablets), instead of in an external server or cloud, which makes solutions based on these networks more accessible, and allows for an improved control on the processing of data used by these networks.
The main goal of Antonio's research is to enable NAS algorithms to compete with human researchers for the design of neural networks—a long and tedious task for humans. To this aim, Antonio studies various methods for optimising and accelerating the process of NAS. The main focus of his research are self-structuring algorithms, which allow to build or modify an architecture at the same time as it is being trained.
Together with his PhD thesis supervisor, Prof. Hugues Bersini from IRIDIA-CoDE, Antonio is co-author of three papers about the self-structuring algorithms developed within the frame of his doctoral research activities. These articles have been presented in different conferences:
- Self-Optimisation of Dense Neural Network Architectures: An Incremental Approach, published in the proceedings of the IJCNN 2020 conference (DOI: 10.1109/IJCNN48605.2020.9207416).
- DensEMANN: Building A DenseNet From Scratch, Layer by Layer and Kernel by Kernel, published in the proceedings of the IJCNN 2021 conference (DOI: 10.1109/IJCNN52387.2021.9533783).
- DensEMANN + Sparsification: Experiments for Further Shrinking Already Small Automatically Generated DenseNet, published in the proceedings of the ICANN 2022 conference (DOI: 10.1007/978-3-031-15934-3_50).
Antonio joined TRAIL in April 2021. His research is currently funded by the ARIAC project by DigitalWallonia4.ai, and he is invested in projects linked to TRAIL work packages 1 (Human-AI Interaction) and 4 (Optimised Implementations of AI), and to Grand Challenges 7 (Interactions with State-of-the-Art AI to reach Zero-Defect, Zero-Accident, and Zero-Burnout in a Production Environment) and 8 (Using Advanced AI Solutions to Achieve "First Time Right" and "Consistent Product Quality" Throughout the Product Development and Manufacturing Cycle).
In June 2021, Antonio was also proclaimed laureate of the VOCATIO foundation, which gives yearly awards for young talents in Belgium. He was one of the two laureates that year for the "Applied Sciences" category.
Antonio's mother tongue is Spanish (his own nationality), and he also speaks both English and French at native level. He also has a good proficiency in Dutch.