Measuring Arithmetic Extrapolation Performance

Alexander Rosenberg Johansen, Andreas Madsen

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

67 Downloads (Pure)


The Neural Arithmetic Logic Unit (NALU) is a neural network layer that can learn exact arithmetic operations between the elements of a hidden state. The goal of NALU is to learn perfect extrapolation, which requires learning the exact underlying logic of an unknown arithmetic problem. Evaluating the performance of the NALU is non-trivial as one arithmetic problem might have many solutions. As a consequence, single-instance MSE has been used to evaluate and compare performance between models. However, it can be hard to interpret what magnitude of MSE represents a correct solution and models sensitivity to initialization. We propose using a success-criterion to measure if and when a model converges. Using a success-criterion we can summarize success-rate over many initialization seeds and calculate confidence intervals. We contribute a generalized version of the previous arithmetic benchmark to measure models sensitivity under different conditions. This is, to our knowledge, the first extensive evaluation with respect to convergence of the NALU and its sub-units. Using a success-criterion to summarize 4800 experiments we find that consistently learning arithmetic extrapolation is challenging, in particular for multiplication.
Original languageEnglish
Title of host publicationScience meets Engineering of Deep Learning at 33rd Conference on Neural Information Processing Systems
Number of pages5
Publication date2019
Publication statusPublished - 2019
Event33rd Conference on Neural Information Processing Systems - Vancouver Convention Centre, Vancouver, Canada
Duration: 8 Dec 201914 Dec 2019
Conference number: 33


Conference33rd Conference on Neural Information Processing Systems
LocationVancouver Convention Centre
Internet address

Fingerprint Dive into the research topics of 'Measuring Arithmetic Extrapolation Performance'. Together they form a unique fingerprint.

Cite this