2022

Natural Language Descriptions of Deep Visual Features

Hernandez, Evan, Schwettmann, Sarah, Bau, David et al.

Understand

Some neurons in deep networks specialize in recognizing highly specific perceptual, structural, or semantic features of inputs.

In computer vision, techniques exist for identifying neurons that respond to individual concept categories like colors, textures, and object classes.
But these techniques are limited in scope, labeling only a small subset of neurons and behaviors in any network.
Is a richer characterization of neuron-level computation possible? We introduce a procedure (called MILAN, for mutual-information-guided linguistic annotation of neurons) that automatically labels neurons with open-ended, compositional, natural language descriptions.

Built on

The distribution of the flora in the alpine zone
Paul Jaccard · 1912
Earlier work this paper cites.
Long short-term memory
Sepp Hochreiter and Jürgen Schmidhuber · 1997
Earlier work this paper cites.
Bleu: a method for automatic evaluation of machine translation
Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu · 2002
Earlier work this paper cites.
Modeling annotators: A generative approach to learning from annotator rationales
Omar Zaidan and Jason Eisner · 2008
Earlier work this paper cites.
Imagenet: A large-scale hierarchical image database
Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei · 2009
Earlier work this paper cites.
Visualizing higher-layer features of a deep network
Dumitru Erhan, Yoshua Bengio, Aaron Courville, and Pascal Vincent · 2009
Earlier work this paper cites.
When training and test sets are different: characterizing learning transfer
Amos Storkey · 2009
Earlier work this paper cites.
Collecting image annotations using amazon’s mechanical turk
Cyrus Rashtchian, Peter Young, Micah Hodosh, and Julia Hockenmaier · 2010
Earlier work this paper cites.
Imagenet classification with deep convolutional neural networks
Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton · 2012
Earlier work this paper cites.
Rich feature hierarchies for accurate object detection and semantic segmentation
Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik · 2014
Earlier work this paper cites.
Microsoft coco: Common objects in context
Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick · 2014
Earlier work this paper cites.
Visualizing and understanding convolutional networks
Matthew D Zeiler and Rob Fergus · 2014
Earlier work this paper cites.
Neural machine translation by jointly learning to align and translate
Dzmitry Bahdanau, Kyung Hyun Cho, and Yoshua Bengio · 2015
Earlier work this paper cites.
Long-term recurrent convolutional networks for visual recognition and description
Jeffrey Donahue, Lisa Anne Hendricks, Sergio Guadarrama, Marcus Rohrbach, Subhashini Venugopalan, Kate Saenko, and Trevor Darrell · 2015
Earlier work this paper cites.
Deep residual learning for image recognition, 2015
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun · 2015
Earlier work this paper cites.
Visualizing and understanding recurrent networks
Original
Andrej Karpathy, Justin Johnson, and Li Fei-Fei · 2015
Earlier work this paper cites.
Understanding deep image representations by inverting them
Aravindh Mahendran and Andrea Vedaldi · 2015
Earlier work this paper cites.
Very deep convolutional networks for large-scale image recognition
Karen Simonyan and Andrew Zisserman · 2015
Earlier work this paper cites.
Show, attend and tell: Neural image caption generation with visual attention
Kelvin Xu, Jimmy Lei Ba, Ryan Kiros, Kyunghyun Cho, Aaron Courville, Ruslan Salakhutdinov, Richard S. Zemel, and Yoshua Bengio · 2015
Earlier work this paper cites.
Generating visual explanations
Lisa Anne Hendricks, Zeynep Akata, Marcus Rohrbach, Jeff Donahue, Bernt Schiele, and Trevor Darrell · 2016
Earlier work this paper cites.
Analogs of linguistic structure in deep representations
Jacob Andreas and Dan Klein · 2017
Earlier work this paper cites.
Translating neuralese
Jacob Andreas, Anca D Dragan, and Dan Klein · 2017
Earlier work this paper cites.

Similar

Network dissection: Quantifying interpretability of deep visual representations
David Bau, Bolei Zhou, Aditya Khosla, Aude Oliva, and Antonio Torralba · 2017
Cited alongside, same era.
Feature visualization
Chris Olah, Alexander Mordvintsev, and Ludwig Schubert · 2017
Cited alongside, same era.
Learning to generate reviews and discovering sentiment
Original
Alec Radford, Rafal Jozefowicz, and Ilya Sutskever · 2017
Cited alongside, same era.
Places: A 10 million image database for scene recognition
Bolei Zhou, Agata Lapedriza, Aditya Khosla, Aude Oliva, and Antonio Torralba · 2017
Cited alongside, same era.
Identifying and controlling important neurons in neural machine translation
Anthony Bau, Yonatan Belinkov, Hassan Sajjad, Nadir Durrani, Fahim Dalvi, and James Glass · 2018
Cited alongside, same era.
Large scale gan training for high fidelity natural image synthesis
Andrew Brock, Jeff Donahue, and Karen Simonyan · 2018
Cited alongside, same era.
e-snli: Natural language inference with natural language explanations
Original
Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom · 2018
Cited alongside, same era.
Datasheets for datasets
Original
Timnit Gebru, Jamie Morgenstern, Briana Vecchione, Jennifer Wortman Vaughan, Hanna Wallach, Hal Daumé III, and Kate Crawford · 2018
Cited alongside, same era.
Grounding visual explanations
Lisa Anne Hendricks, Ronghang Hu, Trevor Darrell, and Zeynep Akata · 2018
Cited alongside, same era.
Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (tcav)
Been Kim, Martin Wattenberg, Justin Gilmer, Carrie Cai, James Wexler, Fernanda Viegas, et al · 2018
Cited alongside, same era.
On the importance of single directions for generalization
Ari S Morcos, David GT Barrett, Neil C Rabinowitz, and Matthew Botvinick · 2018
Cited alongside, same era.
The building blocks of interpretability
Chris Olah, Arvind Satyanarayan, Ian Johnson, Shan Carter, Ludwig Schubert, Katherine Ye, and Alexander Mordvintsev · 2018
Cited alongside, same era.

Then

From recognition to cognition: Visual commonsense reasoning
Rowan Zellers, Yonatan Bisk, Ali Farhadi, and Yejin Choi · 2019
Later among the works it cites.
Semantic understanding of scenes through the ade20k dataset
Bolei Zhou, Hang Zhao, Xavier Puig, Tete Xiao, Sanja Fidler, Adela Barriuso, and Antonio Torralba · 2019
Later among the works it cites.
Understanding the role of individual units in a deep neural network
David Bau, Jun-Yan Zhu, Hendrik Strobelt, Agata Lapedriza, Bolei Zhou, and Antonio Torralba · 2020
Later among the works it cites.
Exploring simple siamese representation learning, 2020
Xinlei Chen and Kaiming He · 2020
Later among the works it cites.
Enabling certification of verification-agnostic networks via memory-efficient semidefinite programming
Sumanth Dathathri, Krishnamurthy Dvijotham, Alexey Kurakin, Aditi Raghunathan, Jonathan Uesato, Rudy Bunel, Shreya Shankar, Jacob Steinhardt, Ian Goodfellow, Percy Liang, et al · 2020
Later among the works it cites.
Bootstrap your own latent: A new approach to self-supervised learning, 2020
Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre H. Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Daniel Guo, Mohammad Gheshlaghi Azar, Bilal Piot, Koray Kavukcuoglu, Rémi Munos, and Michal Valko · 2020
Later among the works it cites.
spaCy: Industrial-strength Natural Language Processing in Python, 2020
Matthew Honnibal, Ines Montani, Sofie Van Landeghem, and Adriane Boyd · 2020
Later among the works it cites.
Quantifying learnability and describability of visual concepts emerging in representation learning
Iro Laina, Ruth C. Fong, and Andrea Vedaldi · 2020
Later among the works it cites.
Compositional explanations of neurons
Jesse Mu and Jacob Andreas · 2020
Later among the works it cites.
WT5?! Training text-to-text models to explain their predictions
Original
Sharan Narang, Colin Raffel, Katherine Lee, Adam Roberts, Noah Fiedel, and Karishma Malkan · 2020
Later among the works it cites.
A human-centered agenda for intelligible machine learning
Jennifer Wortman Vaughan and Hanna Wallach · 2020
Later among the works it cites.
Towards unique and informative captioning of images
Zeyu Wang, Berthy Feng, Karthik Narasimhan, and Olga Russakovsky · 2020
Later among the works it cites.
Noise or signal: The role of image backgrounds in object recognition
Original
Kai Xiao, Logan Engstrom, Andrew Ilyas, and Aleksander Madry · 2020
Later among the works it cites.
Bertscore: Evaluating text generation with bert
Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi · 2020
Later among the works it cites.
An interpretability illusion for bert
Original
Tolga Bolukbasi, Adam Pearce, Ann Yuan, Andy Coenen, Emily Reif, Fernanda Viégas, and Martin Wattenberg · 2021
Later among the works it cites.
Curve circuits
Nick Cammarata, Gabriel Goh, Shan Carter, Chelsea Voss, Ludwig Schubert, and Chris Olah · 2021
Later among the works it cites.
Emerging properties in self-supervised vision transformers
Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin · 2021
Later among the works it cites.
An image is worth 16x16 words: Transformers for image recognition at scale
Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al · 2021
Later among the works it cites.
Multimodal neurons in artificial neural networks
Gabriel Goh, Nick Cammarata, Chelsea Voss, Shan Carter, Michael Petrov, Ludwig Schubert, Alec Radford, and Chris Olah · 2021
Later among the works it cites.
Toward a visual concept vocabulary for gan latent space
Sarah Schwettmann, Evan Hernandez, David Bau, Samuel Klein, Jacob Andreas, and Antonio Torralba · 2021
Later among the works it cites.
A study of face obfuscation in imagenet
Original
Kaiyu Yang, Jacqueline Yau, Li Fei-Fei, Jia Deng, and Olga Russakovsky · 2021
Later among the works it cites.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…

Reading the bibliography…