2020

Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Masks

Csordás, Róbert, van Steenkiste, Sjoerd, Schmidhuber, Jürgen

Understand

Neural networks (NNs) whose subnetworks implement reusable functions are expected to offer numerous advantages, including compositionality through efficient recombination of functional building blocks, interpretability, preventing catastrophic interference, etc.

Understanding if and how NNs are modular could provide insights into how to improve them.
Current inspection methods, however, fail to link modules to their functionality.
In this paper, we present a novel method based on learning binary weight masks to identify individual weights and subnets responsible for specific functions.

Built on

Self-organization of orientation sensitive cells in the striate cortex
Chr. von der Malsburg · 1973
Earlier work this paper cites.
Modular learning in neural networks
Dana H Ballard · 1987
Earlier work this paper cites.
Connectionism and cognitive architecture: A critical analysis
Jerry A Fodor, Zenon W Pylyshyn, et al · 1988
Earlier work this paper cites.
Catastrophic interference in connectionist networks: The sequential learning problem
Michael McCloskey and Neal J Cohen · 1989
Earlier work this paper cites.
Optimal brain damage
Yann LeCun, John S Denker, and Sara A Solla · 1990
Earlier work this paper cites.
Second order derivatives for network pruning: Optimal brain surgeon
Babak Hassibi and David G Stork · 1993
Earlier work this paper cites.
Long short-term memory
Sepp Hochreiter and Jürgen Schmidhuber · 1997
Earlier work this paper cites.
Rethinking eliminative connectionism
Gary F Marcus · 1998
Earlier work this paper cites.
Modularity in animal development and evolution: elements of a conceptual framework for evodevo
G von Dassow and E Munro · 1999
Earlier work this paper cites.
Design rules / the power of modularity
Carliss Young. Baldwin and Kim B. Clark · 2000
Earlier work this paper cites.
Learning multiple layers of features from tiny images
Alex Krizhevsky, Geoffrey Hinton, et al · 2009
Earlier work this paper cites.
Mnist handwritten digit database
Yann LeCun, Corinna Cortes, and CJ Burges · 2010
Earlier work this paper cites.
Rectified linear units improve restricted boltzmann machines
Vinod Nair and Geoffrey E. Hinton · 2010
Earlier work this paper cites.
The emergence of modularity in biological systems
Dirk M. Lorenz, Alice Jeng, and Michael W. Deem · 2011
Earlier work this paper cites.
Neural networks for machine learning
Geoffrey Hinton · 2012
Earlier work this paper cites.
Estimating or propagating gradients through stochastic neurons for conditional computation
Yoshua Bengio, Nicholas Léonard, and Aaron C. Courville · 2013
Earlier work this paper cites.
The evolutionary origins of modularity
Jeff Clune, Jean-Baptiste Mouret, and Hod Lipson · 2013
Earlier work this paper cites.
Deep inside convolutional networks: Visualising image classification models and saliency maps
Original
Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman · 2013
Earlier work this paper cites.
Neural machine translation by jointly learning to align and translate
Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio · 2015
Earlier work this paper cites.
Conditional computation in neural networks for faster models
Original
Emmanuel Bengio, Pierre-Luc Bacon, Joelle Pineau, and Doina Precup · 2015
Earlier work this paper cites.

Similar

Adam: A method for stochastic optimization
Diederik P. Kingma and Jimmy Ba · 2015
Cited alongside, same era.
Striving for simplicity: The all convolutional net
Jost Tobias Springenberg, A. Dosovitskiy, T. Brox, and Martin A. Riedmiller · 2015
Cited alongside, same era.
Neural module networks
Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Dan Klein · 2016
Cited alongside, same era.
Deep residual learning for image recognition
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun · 2016
Cited alongside, same era.
Pathnet: Evolution channels gradient descent in super neural networks
Original
Chrisantha Fernando, Dylan Banarse, Charles Blundell, Yori Zwols, David Ha, Andrei A Rusu, Alexander Pritzel, and Daan Wierstra · 2017
Cited alongside, same era.
Categorical reparametrization with gumbel-softmax
Eric Jang, Shixiang Gu, and Ben Poole · 2017
Cited alongside, same era.
Overcoming catastrophic forgetting in neural networks
James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al · 2017
Cited alongside, same era.
Pruning filters for efficient convnets
Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf · 2017
Cited alongside, same era.
The concrete distribution: A continuous relaxation of discrete random variables
Chris J Maddison, Andriy Mnih, and Yee Whye Teh · 2017
Cited alongside, same era.
Learning important features through propagating activation differences
Avanti Shrikumar, Peyton Greenside, and Anshul Kundaje · 2017
Cited alongside, same era.
Axiomatic attribution for deep networks
Mukund Sundararajan, Ankur Taly, and Qiqi Yan · 2017
Cited alongside, same era.
Attention is all you need
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin · 2017
Cited alongside, same era.

Then

Systematic generalization: What is required and can it be learned?
Dzmitry Bahdanau, Shikhar Murty, Michael Noukhovitch, Thien Huu Nguyen, Harm de Vries, and Aaron Courville · 2019
Later among the works it cites.
Automatically composing representation transformations as a means for generalization
Michael Chang, Abhishek Gupta, Sergey Levine, and Thomas L. Griffiths · 2019
Later among the works it cites.
Universal transformers
Mostafa Dehghani, Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit, and Lukasz Kaiser · 2019
Later among the works it cites.
The lottery ticket hypothesis: Finding sparse, trainable neural networks
Jonathan Frankle and Michael Carbin · 2019
Later among the works it cites.
Weight agnostic neural networks
Adam Gaier and David Ha · 2019
Later among the works it cites.
Reconciling deep learning with symbolic artificial intelligence: representing objects and relations
Marta Garnelo and Murray Shanahan · 2019
Later among the works it cites.
Continual learning via neural pruning
Siavash Golkar, Michael Kagan, and Kyunghyun Cho · 2019
Later among the works it cites.
Attention-based structural-plasticity
Original
Soheil Kolouri, Nicholas Ketz, Xinyun Zou, Jeffrey Krichmar, and Praveen Pilly · 2019
Later among the works it cites.
Pytorch: An imperative style, high-performance deep learning library
Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala · 2019
Later among the works it cites.
Task-driven modular networks for zero-shot compositional learning
Senthil Purushwalkam, Maximilian Nickel, Abhinav Gupta, and Marc’Aurelio Ranzato · 2019
Later among the works it cites.
Routing networks and the challenges of modular and compositional computation
Original
Clemens Rosenbaum, Ignacio Cases, Matthew Riemer, and Tim Klinger · 2019
Later among the works it cites.
Analysing mathematical reasoning abilities of neural models
David Saxton, Edward Grefenstette, Felix Hill, and Pushmeet Kohli · 2019
Later among the works it cites.
Interpreting layered neural networks via hierarchical modular representation
Chihiro Watanabe · 2019
Later among the works it cites.
On network science and mutual information for explaining deep neural networks
Brian Davis, Umang Bhatt, Kartikeya Bhardwaj, Radu Marculescu, and José MF Moura · 2020
Closest in time.
Neural networks are surprisingly modular
Original
Daniel Filan, Shlomi Hod, Cody Wild, Andrew Critch, and Stuart Russell · 2020
Closest in time.
On the binding problem in artificial neural networks
Original
Klaus Greff, Sjoerd van Steenkiste, and Jürgen Schmidhuber · 2020
Closest in time.
Environmental drivers of systematicity and generalization in a situated agent
Felix Hill, Andrew K. Lampinen, Rosalia Schneider, Stephen Clark, Matthew Botvinick, James L. McClelland, and Adam Santoro · 2020
Closest in time.
Compositionality decomposed: How do neural networks generalise?
Dieuwke Hupkes, Verna Dankers, Mathijs Mul, and Elia Bruni · 2020
Closest in time.
Learning to combine top-down and bottom-up signals in recurrent neural networks with attention over modules
Sarthak Mittal, Alex Lamb, Anirudh Goyal, Vikram Voleti, Murray Shanahan, Guillaume Lajoie, Michael Mozer, and Yoshua Bengio · 2020
Closest in time.
Multi-task reinforcement learning with soft modularization
Original
Ruihan Yang, Huazhe Xu, Yi Wu, and Xiaolong Wang · 2020
Closest in time.

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…