Multimodal representation learning
Lead Research Organisation:
Imperial College London
Department Name: Computing
Abstract
The analysis of massive high-dimensional datasets is becoming ubiquitous in many areas of science and engineering. Such datasets often contain multiple perspectives or modes of the same underlying phenomena and are known as multimodal
datasets.
The majority of machine learning methods to date do not account for the multiple modes of a dataset and are not scalable for very large inputs (particularly those arising from concatenation of multiple modes). The analysis of multimodal data therefore requires specialised and scalable methods that account for correlations between the modes and their complementary nature. In particular there are multiple open problems with such datasets as they often exhibit noise gross errors missing data and missing labels.
Machine learning
datasets.
The majority of machine learning methods to date do not account for the multiple modes of a dataset and are not scalable for very large inputs (particularly those arising from concatenation of multiple modes). The analysis of multimodal data therefore requires specialised and scalable methods that account for correlations between the modes and their complementary nature. In particular there are multiple open problems with such datasets as they often exhibit noise gross errors missing data and missing labels.
Machine learning
Organisations
People |
ORCID iD |
Maja Pantic (Primary Supervisor) | |
Triantafyllos Kefalas (Student) |
Studentship Projects
Project Reference | Relationship | Related To | Start | End | Student Name |
---|---|---|---|---|---|
EP/N509486/1 | 01/10/2016 | 31/03/2022 | |||
2296501 | Studentship | EP/N509486/1 | 01/10/2018 | 31/03/2022 | Triantafyllos Kefalas |