Open-world learning across domains and modalities
Event details
| Date | 29.02.2024 |
| Hour | 16:30 › 18:30 |
| Speaker | Shuo Wen |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Pascal Frossard
Thesis advisor: Prof. Maria Brbic
Co-examiner: Prof. Robert West
Abstract
Open-world learning is a paradigm in machine
learning that aims to mimic human cognition by allowing models
to continuously learn from new data and adapt to previously
unseen classes or concepts. However, current approaches to openworld
learning rely on strong assumptions, thereby restricting
their applicability to a limited range of open-world scenarios.
For example, many approaches assume that labeled and unlabeled
data originate from the same domain, which is not always true
in real-world scenarios. Overlooking crucial challenges such as
domain shift and modality shift can significantly affect model
performance when encountering data from different distributions
or types. In this report, we propose to build an ideal open-world
learning model by dispelling all those assumptions. Specifically,
the model should be capable of perpetually learning new knowledge
from diverse domains and modalities, with applicability
extending to various fields such as biomedicine.
Background papers
Exam president: Prof. Pascal Frossard
Thesis advisor: Prof. Maria Brbic
Co-examiner: Prof. Robert West
Abstract
Open-world learning is a paradigm in machine
learning that aims to mimic human cognition by allowing models
to continuously learn from new data and adapt to previously
unseen classes or concepts. However, current approaches to openworld
learning rely on strong assumptions, thereby restricting
their applicability to a limited range of open-world scenarios.
For example, many approaches assume that labeled and unlabeled
data originate from the same domain, which is not always true
in real-world scenarios. Overlooking crucial challenges such as
domain shift and modality shift can significantly affect model
performance when encountering data from different distributions
or types. In this report, we propose to build an ideal open-world
learning model by dispelling all those assumptions. Specifically,
the model should be capable of perpetually learning new knowledge
from diverse domains and modalities, with applicability
extending to various fields such as biomedicine.
Background papers
Practical information
- General public
- Free