Skip to content
Low-Rank Structured Prediction with Theoretical Guarantees
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
datasets
src
.gitignore
README.md
base.py

README.md

Leveraging Low-Rank Relations Between Surrogate Tasks in Structured Prediction - ICML 2019

We study the interplay between surrogate methods for structured prediction and techniques from multitask learning designed to leverage relationships between surrogate outputs. We propose an efficient algorithm based on trace norm regularization which, differently from previous methods, does not require explicit knowledge of the coding/decoding functions of the surrogate framework. As a result, our algorithm can be applied to the broad class of problems in which the surrogate space is large or even infinite dimensional. We study excess risk bounds for trace norm regularized structured prediction, implying the consistency and learning rates for our estimator. We also identify relevant regimes in which our approach can enjoy better generalization performance than previous methods. Numerical experiments on ranking problems indicate that enforcing low-rank relations among surrogate outputs may indeed provide a significant advantage in practice.

Requirements

Installing igraph

We make use of the igraph package for the decoding step of our algorithm. https://igraph.org/python/

pip install python-igraph

Citation

@inproceedings{luise2019leveraging,
  title={Learning-to-Learn Stochastic Gradient Descent with Biased Regularization},
  author={Luise, Giulia and Stamos, Dimitris and Pontil, Massimiliano and Ciliberto, Carlo},
  booktitle={International Conference on Machine Learning},
  year={2019}
}
You can’t perform that action at this time.