Timezone: »

Generalization and Robustness Implications in Object-Centric Learning
Andrea Dittadi · Samuele Papa · Michele De Vita · Bernhard Schölkopf · Ole Winther · Francesco Locatello

Wed Jul 20 11:20 AM -- 11:25 AM (PDT) @ Hall G

The idea behind object-centric representation learning is that natural scenes can better be modeled as compositions of objects and their relations as opposed to distributed representations. This inductive bias can be injected into neural networks to potentially improve systematic generalization and performance of downstream tasks in scenes with multiple objects. In this paper, we train state-of-the-art unsupervised models on five common multi-object datasets and evaluate segmentation metrics and downstream object property prediction. In addition, we study generalization and robustness by investigating the settings where either a single object is out of distribution -- e.g., having an unseen color, texture, or shape -- or global properties of the scene are altered -- e.g., by occlusions, cropping, or increasing the number of objects. From our experimental study, we find object-centric representations to be useful for downstream tasks and generally robust to most distribution shifts affecting objects. However, when the distribution shift affects the input in a less structured manner, robustness in terms of segmentation and downstream task performance may vary significantly across models and distribution shifts.

Author Information

Andrea Dittadi (Technical University of Denmark)
Samuele Papa (University of Amsterdam)
Samuele Papa

I have a background in information engineering, computer engineering, and artificial intelligence. In 2021 I completed both an MSc (cum laude) in Computer Engineering at the University of Padova and an MSc in Human-Centered Artificial Intelligence at the Technical University of Denmark. During my studies, I focused on fundamental research in the field of Deep Learning, specifically on how to obtain useful representations of images to enable the automation of higher-level cognitive tasks. I am now a PhD candidate under the POP-AART Lab (2021-2024), a collaboration between Elekta, the University of Amsterdam, and the Netherlands Cancer Institute. The aim of the collaboration is personalized online radiotherapy using artificial intelligence methods. The lab is supervised by Jan-Jakob Sonke and Efstratios Gavves. I will focus on using deep generative models to improve the quality of Cone Beam Computed Tomography (CBCT) while enforcing geometric and pathological integrity.

Michele De Vita (DTU)
Bernhard Schölkopf (MPI for Intelligent Systems Tübingen, Germany)

Bernhard Scholkopf received degrees in mathematics (London) and physics (Tubingen), and a doctorate in computer science from the Technical University Berlin. He has researched at AT&T Bell Labs, at GMD FIRST, Berlin, at the Australian National University, Canberra, and at Microsoft Research Cambridge (UK). In 2001, he was appointed scientific member of the Max Planck Society and director at the MPI for Biological Cybernetics; in 2010 he founded the Max Planck Institute for Intelligent Systems. For further information, see www.kyb.tuebingen.mpg.de/~bs.

Ole Winther (DTU and KU)
Francesco Locatello (Amazon Lablet)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors