2022

Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets

Hacohen, Guy, Dekel, Avihu, Weinshall, Daphna

Understand

Investigating active learning, we focus on the relation between the number of labeled examples (budget size), and suitable querying strategies.

Our theoretical analysis shows a behavior reminiscent of phase transition: typical examples are best queried when the budget is low, while unrepresentative examples are best queried when the budget is large.
Combined evidence shows that a similar phenomenon occurs in common classification models.
Accordingly, we propose TypiClust -- a deep active learning strategy suited for low budgets.

Built on

A sequential algorithm for training text classifiers
Lewis, D. D. and Gale, W. A · 1994
Earlier work this paper cites.
Curriculum learning
Bengio, Y., Louradour, J., Collobert, R., and Weston, J · 2009
Earlier work this paper cites.
Imagenet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L · 2009
Earlier work this paper cites.
Learning multiple layers of features from tiny images
Krizhevsky, A., Hinton, G., et al · 2009
Earlier work this paper cites.
Active learning literature survey
Settles, B · 2009
Earlier work this paper cites.
Why label when you can search? alternatives to active learning for applying human resources to build classification models under extreme class imbalance
Attenberg, J. and Provost, F · 2010
Earlier work this paper cites.
Off to a good start: Using clustering to select the initial training set in active learning
Hu, R., Mac Namee, B., and Delany, S. J · 2010
Earlier work this paper cites.
Self-paced learning for latent variable models
Kumar, M. P., Packer, B., and Koller, D · 2010
Earlier work this paper cites.
A convex optimization framework for active learning
Elhamifar, E., Sapiro, G., Yang, A., and Sasrty, S. S · 2013
Earlier work this paper cites.
Very deep convolutional networks for large-scale image recognition
Original
Simonyan, K. and Zisserman, A · 2014
Earlier work this paper cites.
Tiny imagenet visual recognition challenge
Le, Y. and Yang, X · 2015
Earlier work this paper cites.
Deep neural networks are easily fooled: High confidence predictions for unrecognizable images
Nguyen, A., Yosinski, J., and Clune, J · 2015
Earlier work this paper cites.
An introduction to matrix concentration inequalities
Tropp, J. A · 2015
Earlier work this paper cites.
Multi-class active learning by uncertainty sampling with diversity maximization
Yang, Y., Ma, Z., Nie, F., Chang, X., and Hauptmann, A. G · 2015
Earlier work this paper cites.
Dropout as a bayesian approximation: Representing model uncertainty in deep learning
Gal, Y. and Ghahramani, Z · 2016
Earlier work this paper cites.
A batch-mode active learning framework by querying discriminative and representative samples for hyperspectral image classification
Wang, Z., Du, B., Zhang, L., and Zhang, L · 2016
Earlier work this paper cites.
Deep bayesian active learning with image data
Gal, Y., Islam, R., and Ghahramani, Z · 2017
Earlier work this paper cites.
Deep active learning over the long tail
Original
Geifman, Y. and El-Yaniv, R · 2017
Earlier work this paper cites.
Deep active learning for image classification
Ranganathan, H., Venkateswara, H., Chakraborty, S., and Panchanathan, S · 2017
Earlier work this paper cites.
Deep similarity-based batch mode active learning with exploration-exploitation
Yin, C., Qian, B., Cao, S., Li, X., Wei, J., Zheng, Q., and Davidson, I · 2017
Earlier work this paper cites.

Similar

Practical obstacles to deploying active learning
Original
Lowell, D., Lipton, Z. C., and Wallace, B. C · 2018
Cited alongside, same era.
Active learning for convolutional neural networks: A core-set approach
Sener, O. and Savarese, S · 2018
Cited alongside, same era.
The unreasonable effectiveness of deep features as a perceptual metric
Zhang, R., Isola, P., Efros, A. A., Shechtman, E., and Wang, O · 2018
Cited alongside, same era.
Discriminative active learning
Original
Gissin, D. and Shalev-Shwartz, S · 2019
Cited alongside, same era.
On the power of curriculum learning in training deep networks
Hacohen, G. and Weinshall, D · 2019
Cited alongside, same era.
Towards better uncertainty sampling: Active learning with multiple views for deep convolutional neural network
He, T., Jin, X., Ding, G., Yi, L., and Yan, C · 2019
Cited alongside, same era.
Batchbald: Efficient and diverse batch acquisition for deep bayesian active learning
Kirsch, A., Van Amersfoort, J., and Gal, Y · 2019
Cited alongside, same era.
Parting with illusions about deep active learning
Original
Mittal, S., Tatarchenko, M., Çiçek, Ö., and Brox, T · 2019
Cited alongside, same era.
Variational adversarial active learning
Sinha, S., Ebrahimi, S., and Darrell, T · 2019
Cited alongside, same era.
Diverse mini-batch active learning
Original
Zhdanov, F · 2019
Cited alongside, same era.
Addressing the item cold-start problem by attribute-driven active learning
Original
Zhu, Y., Lin, J., He, S., Wang, B., Guan, Z., Liu, H., and Cai, D · 2019
Cited alongside, same era.
Deep batch active learning by diverse, uncertain gradient lower bounds
Ash, J. T., Zhang, C., Krishnamurthy, A., Langford, J., and Agarwal, A · 2020
Cited alongside, same era.

Then

Towards robust and reproducible active learning using neural networks
Original
Munjal, P., Hayat, N., Hayat, M., Sourati, J., and Khan, S · 2020
Later among the works it cites.
A survey of active learning for text classification using deep neural networks
Original
Schröder, C. and Niekler, A · 2020
Later among the works it cites.
The pitfalls of simplicity bias in neural networks
Shah, H., Tamuly, K., Raghunathan, A., Jain, P., and Netrapalli, P · 2020
Later among the works it cites.
Deep active learning: Unified and principled method for query and training
Shui, C., Zhou, F., Gagné, C., and Wang, B · 2020
Later among the works it cites.
Scan: Learning to classify images without labels
Van Gansbeke, W., Vandenhende, S., Georgoulis, S., Proesmans, M., and Van Gool, L · 2020
Later among the works it cites.
Theory of curriculum learning, with convex loss functions
Weinshall, D. and Amir, D · 2020
Later among the works it cites.
Cold-start active learning through self-supervised language modeling
Yuan, M., Lin, H., and Boyd-Graber, J. L · 2020
Later among the works it cites.
Reducing label effort: Self-supervised meets active learning
Bengar, J. Z., van de Weijer, J., Twardowski, B., and Raducanu, B · 2021
Later among the works it cites.
Emerging properties in self-supervised vision transformers
Original
Caron, M., Touvron, H., Misra, I., Jégou, H., Mairal, J., Bojanowski, P., and Joulin, A · 2021
Later among the works it cites.
On the marginal benefit of active learning: Does self-supervision eat its cake?
Chan, Y.-C., Li, M., and Oymak, S · 2021
Later among the works it cites.
On initial pools for deep active learning
Chandra, A. L., Desai, S. V., Devaguptapu, C., and Balasubramanian, V. N · 2021
Later among the works it cites.
The grammar-learning trajectories of neural language models
Original
Choshen, L., Hacohen, G., Weinshall, D., and Abend, O · 2021
Later among the works it cites.
Principal components bias in deep neural networks
Original
Hacohen, G. and Weinshall, D · 2021
Later among the works it cites.
Low budget active learning via wasserstein distance: An integer programming approach
Original
Mahmood, R., Fidler, S., and Law, M. T · 2021
Later among the works it cites.
Improving unsupervised image clustering with robust learning
Park, S., Han, S., Kim, S., Kim, D., Park, S., Hong, S., and Cha, M · 2021
Later among the works it cites.
A simple baseline for low-budget active learning
Original
Pourahmadi, K., Nooralinejad, P., and Pirsiavash, H · 2021
Later among the works it cites.
Rethinking deep active learning: Using unlabeled data at model training
Siméoni, O., Budnik, M., Avrithis, Y., and Gravier, G · 2021
Later among the works it cites.
Flexmatch: Boosting semi-supervised learning with curriculum pseudo labeling
Original
Zhang, B., Wang, Y., Hou, W., Wu, H., Wang, J., Okumura, M., and Shinozaki, T · 2021
Later among the works it cites.
Active learning through a covering lens
Original
Yehuda, O., Dekel, A., Hacohen, G., and Weinshall, D · 2022
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…