Alessandro Torcinovich

Passionate machine (meta) learner

Free University of Bozen - F. Eng.

Hey! 👋

I am an assistant professor at the Free University of Bozen-Bolzano - Faculty of Engineering, and an academic researcher at ETH Zürich - Institute for Machine Learning. Prior to that, I was a postdoctoral research fellow at ETH Zürich and Ca' Foscari University of Venice. I am specialized in theoretically grounded methodologies for image understanding. I am currently working at the intersection of several topics including weakly supervised and representation learning, robustness assessment and object assembly/reconstruction.

Interests

Machine Learning
Image Understanding
Representation Learning
Robustness Analysis

Education

Postdoc in Computer Science

ETH Zürich / Ca' Foscari University of Venice
Ph.D. in Computer Science

Ca' Foscari University of Venice

Featured Publications

Measuring Orthogonality in Representations of Generative Models

We present a new methodology to identify and measure orthogonality in representations.

Robin C. Geyer, Alessandro Torcinovich, João B. S. Carvalho, Alexander Meyer, Joachim M. Buhmann

Measuring Orthogonality in Representations of Generative Models

The Group Loss++: A Deeper Look Into Group Loss For Deep Metric Learning

We adjust Group Loss, and propose Group Loss++, providing a unified framework for retrieval and re-identification

Ismail Elezi, Jenny Seidenschwarz, Laurin Wagner, Sebastiano Vascon, Alessandro Torcinovich, Marcello Pelillo, Laura Leal-Taixé

The Group Loss++: A Deeper Look Into Group Loss For Deep Metric Learning

The Group Loss for Deep Metric Embedding

We propose Group Loss, a new loss for deep metric learning tasks, that takes into account all the observations in a mini-batch at the same time

Ismail Elezi, Sebastiano Vascon, Alessandro Torcinovich, Marcello Pelillo, Laura Leal-Taixé

The Group Loss for Deep Metric Embedding

Unsupervised Domain Adaptation Using Graph Transduction Games

We extend a previously proposed graph transduction method to perform unsupervised domain adaptation on image datasets for object recognition

Sebastiano Vascon, Sinem Aslan, Alessandro Torcinovich, Twan Van Laarhoven, Elena Marchiori, Marcello Pelillo

Unsupervised Domain Adaptation Using Graph Transduction Games

Transductive Label Augmentation For Improved Deep Network Learning

We present a technique to automatic label unlabeled images, through the use of a graph transductive algorithm

Ismail Elezi, Alessandro Torcinovich, Sebastiano Vascon, Marcello Pelillo

Transductive Label Augmentation For Improved Deep Network Learning

Recent Publications

Quickly discover relevant content by filtering publications.

Guglielmo Beretta, Alessandro Torcinovich, Marcello Pelillo (2024). On Generalized KKT Points for the Motzkin-Straus Program. In JOGO.

Cite DOI

Sara Ferro, Alessandro Torcinovich, Arianna Traviglia, Marcello Pelillo (2023). Exploiting Context in Handwriting Recognition Using Trainable Relaxation Labeling. In ICPRAM.

Ben Vardi, Alessandro Torcinovich, Marina Khoroshiltseva, Marcello Pelillo, Ohad Ben-Shahar (2023). Multi-Phase Relaxation Labeling for Square Jigsaw Puzzle Solving. In VISAPP.

Antonio Emanuele Cinà, Alessandro Torcinovich, Marcello Pelillo (2022). A Black-Box Adversarial Attack for Poisoning Clustering. In PR.

Cite

Marina Khoroshiltseva, Ben Vardi, Alessandro Torcinovich, Arianna Traviglia, Ohad Ben-Shahar, Marcello Pelillo (2021). Jigsaw Puzzle Solving as a Consistent Labeling Problem. In CAIP.

See all publications