On the inconsistency of separable losses for structured prediction
Not on the map. Why?
Abstract (arXiv)
In this paper, we prove that separable negative log-likelihood losses for structured prediction are not necessarily Bayes consistent, or, in other words, minimizing these losses may not result in a model that predicts the most probable structure in the data distribution for a given input. This fact opens the question of whether these losses are well-adapted for structured prediction and, if so, why.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 11 of its 35 references that are in the map. First posted on arXiv 25 Jan 2023.
Builds on
11 of 35 references.
References · 11
- 2019Learning with Fenchel-Young Lossescited by 117
- 2018Noise Contrastive Estimation and Negative Sampling for Conditional Models: Consistency and Statistical Efficiencycited by 103
- 2020Efficient Second-Order TreeCRF for Neural Dependency Parsingcited by 48
- 2019Structured Prediction with Projection Oraclescited by 21
- 2016One-vs-Each Approximation to Softmax for Scalable Estimation of Probabilitiescited by 14
- 2019A General Theory for Structured Prediction with Smooth Convex Surrogatescited by 14
- 2016Dependency Parsing as Head Selectioncited by 13
- 2020Span-based discontinuous constituency parsing: a family of exact chart-based algorithms with time complexities from O(n^6) down to O(n^3)cited by 11
- 2020Consistent Structured Prediction with Max-Min Margin Markov Networkscited by 9
- 2021Partially Supervised Named Entity Recognition via the Expected Entity Ratio Losscited by 7
- 2021On the Consistency of Max-Margin Lossescited by 5
Built on by
Citing papers per year: 2023: 2.
2 papers.
- 2023Structural generalization in COGS: Supertagging is (almost) all you need21 Oct 2023
- 2023On graph-based reentrancy-free semantic parsing15 Feb 2023