2026-09-19T19:28:38Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1229962026-06-06T00:54:36Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Antonio Torralba.
Yip, Richard B.,M. Eng.Massachusetts Institute of Technology.
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science.
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
2019-11-22T00:00:51Z
2019-11-22T00:00:51Z
2019
2019
https://hdl.handle.net/1721.1/122996
1127292891
This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019
Cataloged from student-submitted PDF version of thesis.
Includes bibliographical references (page 29).
While recent years have seen significant advances in the capabilities of image recognition and classification neural networks, we still know little about the relationship between the activation of hidden layers and human-understandable concepts. Recent work in network interpretability has provided a framework for analyzing hidden nodes and layers, showing that in many convolutional architectures, there exists a significant correlation between groups of nodes and human-understandable concepts. We use this framework to investigate the encoding of images produced by standard image classification networks. We do this in the context of encoder-decoder image classification networks. These provide a natural way to observe the effect that perturbing node activations has on the image encoding by observing the generated captions, which are inherently understandable by humans and thus convenient and informative to use. We also generate and analyze captions of images modified by inserting small sub-images of single, human-interpretable concepts. These modifications and the resulting captions show the existence of training-triggered correlations between semantically dissimilar words.
by Richard B. Yip.
M.Eng.
M.Eng. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science
29 pages
eng
Massachusetts Institute of Technology
MIT theses are protected by copyright. They may be viewed, downloaded, or printed from this source but further reproduction or distribution in any format is prohibited without written permission.
http://dspace.mit.edu/handle/1721.1/7582
Electrical Engineering and Computer Science.
Understanding what a captioning network doesn't know
Thesis
application/pdf
2019-11-22T00:00:50Z
Publication
Master
EECS
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="6a866d8a-df7a-4044-a2c4-a6076ca99ea7">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Language>eng</Language>
<Title>Understanding what a captioning network doesn't know</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2019</PublicationDate>
<Authors>
<Author>
<DisplayName>Yip, Richard B.,M. Eng.Massachusetts Institute of Technology.</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>http://dspace.mit.edu/handle/1721.1/7582</License>
<Keyword>Electrical Engineering and Computer Science.</Keyword>
<Abstract>While recent years have seen significant advances in the capabilities of image recognition and classification neural networks, we still know little about the relationship between the activation of hidden layers and human-understandable concepts. Recent work in network interpretability has provided a framework for analyzing hidden nodes and layers, showing that in many convolutional architectures, there exists a significant correlation between groups of nodes and human-understandable concepts. We use this framework to investigate the encoding of images produced by standard image classification networks. We do this in the context of encoder-decoder image classification networks. These provide a natural way to observe the effect that perturbing node activations has on the image encoding by observing the generated captions, which are inherently understandable by humans and thus convenient and informative to use. We also generate and analyze captions of images modified by inserting small sub-images of single, human-interpretable concepts. These modifications and the resulting captions show the existence of training-triggered correlations between semantically dissimilar words.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>