Measuring Compositional Generalization: A Comprehensive Method on Realistic Data (bibtex)
by Daniel Keysers, Nathanael Schärli, Nathan Scales, Hylke Buisman, Daniel Furrer, Sergii Kashubin, Nikola Momchev, Danila Sinopalnikov, Lukasz Stafiniak, Tibor Tihon, Dmitry Tsarkov, Xiao Wang, Marc van Zee, Olivier Bousquet
Abstract:
State-of-the-art machine learning methods exhibit limited compositional generalization. At the same time, there is a lack of realistic benchmarks that comprehensively measure this ability, which makes it challenging to find and evaluate improvements. We introduce a novel method to systematically construct such benchmarks by maximizing compound divergence while guaranteeing a small atom divergence between train and test sets, and we quantitatively compare this method to other approaches for creating compositional generalization benchmarks. We present a large and realistic natural language question answering dataset that is constructed according to this method, and we use it to analyze the compositional generalization ability of three machine learning architectures. We find that they fail to generalize compositionally and that there is a surprisingly strong negative correlation between compound divergence and accuracy. We also demonstrate how our method can be used to create new compositionality benchmarks on top of the existing SCAN dataset, which confirms these findings.
Reference:
Measuring Compositional Generalization: A Comprehensive Method on Realistic Data (Daniel Keysers, Nathanael Schärli, Nathan Scales, Hylke Buisman, Daniel Furrer, Sergii Kashubin, Nikola Momchev, Danila Sinopalnikov, Lukasz Stafiniak, Tibor Tihon, Dmitry Tsarkov, Xiao Wang, Marc van Zee, Olivier Bousquet), In International Conference on Learning Representations (ICLR), 2019.
Bibtex Entry:
@InProceedings{keysers-etal:iclr2019,
  Title                    = {Measuring Compositional Generalization: A Comprehensive Method on Realistic Data},
  Author                   = {Daniel Keysers and Nathanael Sch{\"a}rli and Nathan Scales and Hylke Buisman and Daniel Furrer and Sergii Kashubin and Nikola Momchev and Danila Sinopalnikov and Lukasz Stafiniak and Tibor Tihon and Dmitry Tsarkov and Xiao Wang and Marc van Zee and Olivier Bousquet},
  Booktitle                = {International Conference on Learning Representations (ICLR)},
  Year                     = {2019},
  Month                    = {October},
  Abstract                 = {State-of-the-art machine learning methods exhibit limited compositional generalization. At the same time, there is a lack of realistic benchmarks that comprehensively measure this ability, which makes it challenging to find and evaluate improvements. We introduce a novel method to systematically construct such benchmarks by maximizing compound divergence while guaranteeing a small atom divergence between train and test sets, and we quantitatively compare this method to other approaches for creating compositional generalization benchmarks. We present a large and realistic natural language question answering dataset that is constructed according to this method, and we use it to analyze the compositional generalization ability of three machine learning architectures. We find that they fail to generalize compositionally and that there is a surprisingly strong negative correlation between compound divergence and accuracy. We also demonstrate how our method can be used to create new compositionality benchmarks on top of the existing SCAN dataset, which confirms these findings.},
  Url                      = {http://www.marcvanzee.nl/publications/2019/iclr-compositional-generalization.pdf}
}