Giovani in un'ora - Ciclo di seminari - Prima parte
- Day - Time: 04 March 2021, h.11:00
- Place: Area della Ricerca CNR di Pisa - Room: Zoom
Speakers
Referent
Abstract
Giulio Ermanno Pibiri - "Efficiency for Real-World Applications"
Abstract: Practical efficiency in Algorithms and Data Structures is essential to deliver better services by using less computing resources. Its impact is far-reaching and implies substantial economic gains. However, it is usually achieved via a non trivial combination of theory, experimentation, tuning, and low-level programming. In this brief talk, I will introduce this fascinating research area and why it matters, along with practical examples of applications, such as: inverted indexing, language modeling, RDF triple indexing, query auto-completion, and minimal perfect hashing.
Luca Ciampi - "Unsupervised Domain Adaptation for Traffic Density Estimation and Counting"
Abstract: Convolutional Neural Networks have produced state-of-the-art results for a multitude of computer vision tasks under supervised learning. However, the crux of these methods is the need for a massive amount of labeled data to guarantee that they generalize well to diverse testing scenarios. In many real-world applications, there is indeed a large Domain Shift between the distributions of the train (source) and test (target) domains, leading to a significant drop in performance at inference time. Unsupervised Domain Adaptation (UDA) is a class of techniques that aims to mitigate this drawback without the need for labeled data in the target domain. This makes it particularly useful for the tasks in which acquiring new labeled data is very expensive, such as for semantic and instance segmentation. In this talk, I will introduce an end-to-end CNN-based UDA algorithm for traffic density estimation and counting, based on adversarial learning in the output space. The density estimation is one of those tasks requiring per-pixel annotated labels and, therefore, needs a lot of human effort. I conducted experiments considering different types of domain shifts and exploiting various datasets. One of them, the Grand Traffic Auto dataset, is a synthetic collection of images automatically annotated, obtained using the graphical engine of the Grand Theft Auto videogame. In this case, the domain shift is represented by the Synthetic2Real difference. Experiments show a significant improvement using our UDA algorithm compared to the model’s performance without domain adaptation. This idea is currently developing together with the Instituto Superior Técnico (IST) of the University of Lisbon. The collaboration with IST is a consequence of a visiting period granted by the GYM.
Luigi Malomo - "Computational design of soft molds for casting 3D objects"
Abstract: The recent developments and economic growth related to digital fabrication techniques and machinery created a strong need for robust computational tools and methods that allow users to fully harness the capabilities and potential offered by these technologies. This demand led to the birth of a novel research domain called Computational Fabrication, where knowledge and results from different research fields are applied to improve the scope of digital fabrication. In this talk we will present some recent contributions in this field related to the automatic generation of soft molds for casting objects with complex free-form geometry. We will show how the use of a flexible material and our algorithms for mold design improved the state of the art on the subject.