2016

Understanding intermediate layers using linear classifier probes

Alain, Guillaume, Bengio, Yoshua

Understand

Neural network models have a reputation for being black boxes.

We propose to monitor the features at every layer of a model and measure how suitable they are for classification.
We use linear classifiers, which we refer to as "probes", trained entirely independently of the model itself.
This helps us better understand the roles and dynamics of the intermediate layers.

Built on

What is the best multi-stage architecture for object recognition?
Jarrett, K., Kavukcuoglu, K., Lecun, Y., et al · 2009
Earlier work this paper cites.
Kernel analysis of deep networks
Montavon, G., Braun, M. L., and Müller, K.-R. (2011) · 2011
Earlier work this paper cites.
Evasion attacks against machine learning at test time
Biggio, B., Corona, I., Maiorca, D., Nelson, B., Šrndić, N., Laskov, P., Giacinto, G., and Roli, F. (2013) · 2013
Earlier work this paper cites.
Intriguing properties of neural networks
Original
Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., and Fergus, R. (2013) · 2013
Earlier work this paper cites.
Decaf: A deep convolutional activation feature for generic visual recognition
Donahue, J., Jia, Y., Vinyals, O., Hoffman, J., Zhang, N., Tzeng, E., and Darrell, T. (2014) · 2014
Earlier work this paper cites.
Generative adversarial nets
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. (2014) · 2014
Earlier work this paper cites.
How transferable are features in deep neural networks?
Yosinski, J., Clune, J., Bengio, Y., and Lipson, H. (2014) · 2014
Earlier work this paper cites.
Visualizing and understanding convolutional networks
Zeiler, M. D. and Fergus, R. (2014) · 2014
Earlier work this paper cites.
On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation
Bach, S., Binder, A., Montavon, G., Klauschen, F., Müller, K.-R., and Samek, W. (2015) · 2015
Earlier work this paper cites.

Similar

Understanding deep image representations by inverting them
Mahendran, A. and Vedaldi, A. (2015) · 2015
Cited alongside, same era.
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., Berg, A. C., and Fei-Fei, L. (2015) · 2015
Cited alongside, same era.
Going deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., and Rabinovich, A. (2015) · 2015
Cited alongside, same era.
Understanding intermediate layers using linear classifier probes
Original
Alain, G. and Bengio, Y. (2016) · 2016
Cited alongside, same era.
Layer-wise relevance propagation for neural networks with local renormalization layers
Binder, A., Montavon, G., Lapuschkin, S., Müller, K.-R., and Samek, W. (2016) · 2016
Cited alongside, same era.
Inverting visual representations with convolutional networks
Dosovitskiy, A. and Brox, T. (2016) · 2016
Cited alongside, same era.
Deep residual learning for image recognition
He, K., Zhang, X., Ren, S., and Sun, J. (2016) · 2016
Cited alongside, same era.
Analyzing classifiers: Fisher vectors and deep neural networks
Lapuschkin, S., Binder, A., Montavon, G., Müller, K.-R., and Samek, W. (2016) · 2016
Cited alongside, same era.
Bottom up or top down? dynamics of deep representations via canonical correlation analysis
Raghu, M., Yosinski, J., and Sohl-Dickstein, J. (2017a)
Cited in the paper.
Svcca: Singular vector canonical correlation analysis for deep understanding and improvement
Original
Raghu, M., Gilmer, J., Yosinski, J., and Sohl-Dickstein, J. (2017b)
Cited in the paper.

Then

Fractalnet: Ultra-deep neural networks without residuals
Original
Larsson, G., Maire, M., and Shakhnarovich, G. (2016) · 2016
Closest in time.
Visualizing deep convolutional neural networks using natural pre-images
Mahendran, A. and Vedaldi, A. (2016) · 2016
Closest in time.
Swapout: Learning an ensemble of deep architectures
Singh, S., Hoiem, D., and Forsyth, D. (2016) · 2016
Closest in time.
Residual networks behave like ensembles of relatively shallow networks
Veit, A., Wilber, M. J., and Belongie, S. (2016) · 2016
Closest in time.
Understanding deep learning requires rethinking generalization
Original
Zhang, C., Bengio, S., Hardt, M., Recht, B., and Vinyals, O. (2016) · 2016
Closest in time.
Explaining recurrent neural network predictions in sentiment analysis
Original
Arras, L., Montavon, G., Müller, K.-R., and Samek, W. (2017) · 2017
Closest in time.
Residual connections encourage iterative inference
Original
Jastrzebski, S., Arpit, D., Ballas, N., Verma, V., Che, T., and Bengio, Y. (2017) · 2017
Closest in time.
Show, attend and tell: Neural image caption generation with visual attention
Xu, K., Ba, J., Kiros, R., Cho, K., Courville, A., Salakhudinov, R., Zemel, R., and Bengio, Y. (2015) · 2057
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…