We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Rapid Adaptation with Conditionally Shifted Neurons

Abstract: We describe a mechanism by which artificial neural networks can learn rapid adaptation - the ability to adapt on the fly, with little data, to new tasks - that we call conditionally shifted neurons. We apply this mechanism in the framework of metalearning, where the aim is to replicate some of the flexibility of human learning in machines. Conditionally shifted neurons modify their activation values with task-specific shifts retrieved from a memory module, which is populated rapidly based on limited task experience. On metalearning benchmarks from the vision and language domains, models augmented with conditionally shifted neurons achieve state-of-the-art results.
Comments: ICML 2018; Added: additional ablation and speed comparison with MetaNet
Subjects: Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE); Machine Learning (stat.ML)
Cite as: arXiv:1712.09926 [cs.LG]
  (or arXiv:1712.09926v3 [cs.LG] for this version)

Submission history

From: Tsendsuren Munkhdalai [view email]
[v1] Thu, 28 Dec 2017 16:47:13 GMT (117kb,D)
[v2] Wed, 7 Mar 2018 15:27:42 GMT (975kb,D)
[v3] Tue, 3 Jul 2018 18:04:34 GMT (1082kb,D)

Link back to: arXiv, form interface, contact.