Transfer learning for image classification with sparse prototype representations
To learn a new visual category from few examples, prior knowledge from unlabeled data as well as previous related categories may be useful. We develop a new method for transfer learning which exploits available unlabeled data and an arbitrary kernel function; we form a representation based on kerne...
Main Authors: | , , |
---|---|
Other Authors: | |
Published: |
2008
|
Subjects: | |
Online Access: | http://hdl.handle.net/1721.1/40797 |
_version_ | 1811069458779734016 |
---|---|
author | Quattoni, Ariadna Collins, Michael Darrell, Trevor |
author2 | Trevor Darrell |
author_facet | Trevor Darrell Quattoni, Ariadna Collins, Michael Darrell, Trevor |
author_sort | Quattoni, Ariadna |
collection | MIT |
description | To learn a new visual category from few examples, prior knowledge from unlabeled data as well as previous related categories may be useful. We develop a new method for transfer learning which exploits available unlabeled data and an arbitrary kernel function; we form a representation based on kernel distances to a large set of unlabeled data points. To transfer knowledge from previous related problems we observe that a category might be learnable using only a small subset of reference prototypes. Related problems may share a significant number of relevant prototypes; we find such a reduced representation by performing a joint loss minimization over the training sets of related problems with a shared regularization penalty that minimizes the total number of prototypes involved in the approximation.This optimization problem can be formulated as a linear program thatcan be solved efficiently. We conduct experiments on a news-topic prediction task where the goal is to predict whether an image belongs to a particularnews topic. Our results show that when only few examples are available for training a target topic, leveraging knowledge learnt from other topics can significantly improve performance. |
first_indexed | 2024-09-23T08:10:48Z |
id | mit-1721.1/40797 |
institution | Massachusetts Institute of Technology |
last_indexed | 2024-09-23T08:10:48Z |
publishDate | 2008 |
record_format | dspace |
spelling | mit-1721.1/407972019-04-09T17:06:19Z Transfer learning for image classification with sparse prototype representations Quattoni, Ariadna Collins, Michael Darrell, Trevor Trevor Darrell Vision transfer learning image classification To learn a new visual category from few examples, prior knowledge from unlabeled data as well as previous related categories may be useful. We develop a new method for transfer learning which exploits available unlabeled data and an arbitrary kernel function; we form a representation based on kernel distances to a large set of unlabeled data points. To transfer knowledge from previous related problems we observe that a category might be learnable using only a small subset of reference prototypes. Related problems may share a significant number of relevant prototypes; we find such a reduced representation by performing a joint loss minimization over the training sets of related problems with a shared regularization penalty that minimizes the total number of prototypes involved in the approximation.This optimization problem can be formulated as a linear program thatcan be solved efficiently. We conduct experiments on a news-topic prediction task where the goal is to predict whether an image belongs to a particularnews topic. Our results show that when only few examples are available for training a target topic, leveraging knowledge learnt from other topics can significantly improve performance. 2008-03-03T14:45:13Z 2008-03-03T14:45:13Z 2008-03-03 MIT-CSAIL-TR-2008-012 http://hdl.handle.net/1721.1/40797 Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory 8 p. application/pdf application/postscript |
spellingShingle | transfer learning image classification Quattoni, Ariadna Collins, Michael Darrell, Trevor Transfer learning for image classification with sparse prototype representations |
title | Transfer learning for image classification with sparse prototype representations |
title_full | Transfer learning for image classification with sparse prototype representations |
title_fullStr | Transfer learning for image classification with sparse prototype representations |
title_full_unstemmed | Transfer learning for image classification with sparse prototype representations |
title_short | Transfer learning for image classification with sparse prototype representations |
title_sort | transfer learning for image classification with sparse prototype representations |
topic | transfer learning image classification |
url | http://hdl.handle.net/1721.1/40797 |
work_keys_str_mv | AT quattoniariadna transferlearningforimageclassificationwithsparseprototyperepresentations AT collinsmichael transferlearningforimageclassificationwithsparseprototyperepresentations AT darrelltrevor transferlearningforimageclassificationwithsparseprototyperepresentations |