SWAPE jointly learns parameters and sparse sample weights
Original title: A Nonconvex Framework for Joint Sparse Sample Weighting and Parameter Estimation
Alternating majorize-minimize updates cover convex and difference-of-convex losses while down-weighting anomalous points.
The paper proposes SWAPE, which optimizes model parameters and a weight for each training point together. It covers convex, weakly convex, and difference-of-convex losses, plus weakly convex sparsity penalties on the weights. An alternating method approximates both subproblems with majorize-minimize steps, and the authors show convergence. Experiments with data anomalies report less impact from undesirable points than benchmark methods.
- Sample weights are learned jointly with parameters rather than fixed before training.
- The paper gives approximation methods for the loss functions it considers.
- Numerical tests compare the framework with benchmark methods on anomaly settings.
From the original article
1 min read at arxiv.orgWe propose a joint sparse sample weighting and parameter estimation (SWAPE) framework that jointly optimizes model parameters and sample weights to train a model while simultaneously assigning an individual weight to each training data point.
By Hassan Nojavan · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.