Compositional Zero-Shot Learning - Dr. Massimiliano Mancini
Details
In compositional zero-shot learning, the goal is to recognize unseen compositions (e.g. old dog) of observed visual primitives states (e.g. old, cute) and objects (e.g. car, dog) in the training set. This is challenging because the same state can for example alter the visual appearance of a dog drastically differently from a car. The first part of the talk will provide an overview of different effective solutions to this problem and their limitations. It will then present a recent approach, called Compositional Graph Embedding that exploits the dependency between states, objects, and their compositions within a graph structure to enforce the relevant knowledge transfer from seen to unseen compositions. Finally, the talk will discuss the limitations of the CZSL problem formulation and benchmarks, describing the first effort to address one of them, i.e. the assumption of knowing the unseen compositions the model will face at test time.
The talk is based on the papers:
Learning Graph Embeddings for Compositional Zero-shot Learning (CVPR 2021)
arxiv: https://arxiv.org/abs/2102.01987
Open World Compositional Zero-Shot Learning (CVPR 2021)
arxiv: https://arxiv.org/abs/2101.12609
Presenter BIO:
Massimiliano Mancini is a postdoc researcher at the Explainable Machine Learning group at the University of Tübingen, led by Prof. Zeynep Akata. He completed his Ph.D. in Engineering in Computer Science at the Sapienza University of Rome, advised by Prof. Barbara Caputo and Prof. Elisa Ricci. During the Ph.D. he has been a member of the ELLIS Ph.D. program, of the Technologies of Vision lab at Fondazione Bruno Kessler, of the Visual Learning and Multimodal Applications Laboratory at Italian Institute of Technology, and a visiting Ph.D. student in the Robotics, Perception, and Learning Laboratory at KTH Royal Institute of Technology in Stockholm. Massimiliano's research interests are on the generalization of deep architectures to new domains and semantic concepts, spanning topics such as Domain Adaptation, Zero-shot Learning, and Incremental Learning.
** ** Please register through the zoom link right after your RSVP. We will send the links to the zoom event via email only to those who have registered through zoom. ** **
-------------------------
Find us at:
All lectures are uploaded to our Youtube channel ➜ https://www.youtube.com/channel/UCHObHaxTXKFyI_EI8HiQ5xw
Newsletter for updates about more events ➜ http://eepurl.com/gJ1t-D
Sub-reddit for discussions ➜ https://www.reddit.com/r/2D3DAI/
Discord server for, well, discord ➜ https://discord.gg/MZuWSjF
Blog ➜ https://2d3d.ai
AI Consultancy -> https://abelians.com