Giulia Pasquale Email & Phone Number

Genoa, It

Genova, Liguria, Italy

Genova, Italia

My current research focuses on automated visual object perception, targeting the detection, recognition and localization of multiple objects in the scene.With respect to classical computer vision settings, robot/machine vision scenarios present critical challenges (wide range of nuisances, scarcity of data, unknown objects, etc...), but also clear opportunities (on-site data collection and annotation, limited setting, spatial and temporal information, etc...), that need be exploited in order to achieve good performance for real-world applications.My research activity builds on the work of my PhD and aims at devising suitable systems for robotic platforms, by leveraging latest computer vision advances based on deep learning. The algorithms are deployed on humanoid robots like iCub and R1, and are made publicly available through GitHub.

Genova, Italia

Massive and extensive investigation of the benefits and limitations of applying deep learning methods to visual object recognition on robotic platforms. I addressed problems like automatic data collection and annotation, robustness to visual nuisances, open-set and online/incremental learning, which are typical of a scenario of a humanoid interacting with people in unconstrained environments.A deployment of an on-the-fly visual object recognition system investigated during my PhD is currently in use on iCub and R1 humanoids (code available as part of the iCub open-source eco-system), remarkably improving the real time visual recognition capabilities of the robots.The datasets that I collected in order to benchmark the systems are part of the iCubWorld project (see https://robotology.github.io/iCubWorld/ for more info).

Genova

Parallel implementation of algorithms for high-performance stochastic simulation of large biological systems. Specifically:− voxel-based methods for the stochastic simulation of Reaction-Diffusion systems (CUDA implementation of the Sτ-DPP algorithm and comparison with the MPI implementation);− CUDA implementation of the Fast Non-dominating Sorting Genetic Algorithm (NSGA-II), including different parallelization schemes for the Fast Non-dominating Sorting procedure, that is the core of the algorithm.

Doctor Of Philosophy - Phd, Bioengineering And Robotics (Humanoid Robotics)

Laurea Magistrale, Bioingegneria, 110/110 E Lode

Study of efficient CUDA parallelization schemes for key building blocks in the computation of visual features (e.g., convolution.

Laurea Triennale, Ingegneria Biomedica, 110/110 E Lode

Extension of a distributed hierarchical architecture modelling the encoding of the binocular disparity map in the dorsal stream of visual.

Diploma Scientifico, 100/100