- 31 Downloads
Few-shot learning refers to the machine learning problem of learning a model from very few examples (shots).
Computer vision systems based on machine learning often require the collection of large datasets for their training. This is often a challenging obstacle for their deployment. Moreover, there is evidence that humans are actually able to learn concepts from very few examples [1, 2, 3]. Few-shot learning methods aim to reduce this observed gap between human learning and machine learning. This is achieved by performing a form of transfer learning using data from many previously observed tasks toward new tasks with little data.
Theory and Application
In the 2000s, early research in computer vision on few-shot learning tackled the problem by using hand-designed feature representations and focusing on the exploration of learning and inference algorithms operating on...
- 2.Markman EM (1989) Categorization and naming in children – problems of induction. MIT Press, CambridgeGoogle Scholar
- 4.Miller EG, Matsakis NE, Viola PA (2000) Learning from one example through shared densities on transforms. In: IEEE conference on computer vision and pattern recognition. IEEE Computer Society, pp 464–471Google Scholar
- 5.Fink M (2005) Object classification from a single example utilizing class relevance metrics. In: Advances in neural information processing systems, vol 17, pp 449–456Google Scholar
- 6.Bart E, Ullman S (2005) Cross-generalization: learning novel classes from a single example by feature replacement. In: IEEE conference on computer vision and pattern recognition, pp 672–679. IEEE Computer SocietyGoogle Scholar
- 8.Schmidhuber J (1987) Evolutionary principles in self-referential learning on learning now to learn: The meta-meta-meta…-hook. Diploma thesis, Technische Universitat Munchen, GermanyGoogle Scholar
- 9.Triantafillou E, Zhu T, Dumoulin V, Lamblin P, Evci U, Xu K, Goroshin R, Gelada C, Swersky K, Manzagol P-A, Larochelle H (2020) Meta-dataset: a dataset of datasets for learning to learn from few examples. In: International conference on learning representationsGoogle Scholar
- 10.Ravi S, Larochelle H (2017) Optimization as a model for few-shot learning. In: International conference on learning representationsGoogle Scholar
- 11.Finn C, Abbeel P, Levine S (2017) Model-agnostic meta-learning for fast adaptation of deep networks. In: International conference of machine learningGoogle Scholar
- 12.Bertinetto L, Henriques JF, Valmadre J, Torr P, Vedaldi A (2016) Learning feed-forward one-shot learners. In: Advances in neural information processing systems, vol 29, pp 523–531Google Scholar
- 13.Snell J, Swersky K, Zemel R (2017) Prototypical networks for few-shot learning. In: Advances in neural information processing systems, pp 4077–4087Google Scholar
- 14.Bertinetto L, Henriques JF, Torr P, Vedaldi A (2019) Meta-learning with differentiable closed-form solvers. In: International conference on learning representationsGoogle Scholar
- 16.Vinyals O, Blundell C, Lillicrap T, Wierstra D, et al (2016) Matching networks for one shot learning. In: Advances in neural information processing systems, pp 3630–3638Google Scholar
- 18.Santoro A, Bartunov S, Botvinick M, Wierstra D, Lillicrap T (2016) Meta-learning with memory-augmented neural networks. In: International conference on machine learning, pp 1842–1850Google Scholar
- 19.Graves A, Wayne G, Danihelka I (2014) Neural turing machines. CoRR, abs/1410.5401Google Scholar
- 20.Mishra N, Rohaninejad M, Chen X, Abbeel P (2018) A simple neural attentive meta-learner. In: International conference on learning representationsGoogle Scholar
- 23.Hariharan B, Girshick RB (2017) Low-shot visual recognition by shrinking and hallucinating features. In: IEEE International conference on computer vision, pp 3037–3046Google Scholar