Abstract
The Expectation Maximization (EM) algorithm is of key importance for inference in latent variable models including mixture of regressors and experts, missing observations. This paper introduces a novel EM algorithm, called SPIDER-EM, for inference from a training set of size n, n > 1. At the core of our algorithm is an estimator of the full conditional expectation in the E-step, adapted from the stochastic path-integrated differential estimator (SPIDER) technique. We derive finite-time complexity bounds for smooth non-convex likelihood: we show that for convergence to an e-approximate stationary_ point, the complexity scales as KOpt(n, e) = O(e-1) and KCE(n, e) = n + pnO(e-1), where KOpt(n, e) and KCE(n, e) are respectively the number of M-steps and the number of per-sample conditional expectations evaluations. This improves over the state-of-the-art algorithms. Numerical results support our findings.
| Original language | English |
|---|---|
| Journal | Advances in Neural Information Processing Systems |
| Volume | 2020-December |
| Publication status | Published - 1 Jan 2020 |
| Externally published | Yes |
| Event | 34th Conference on Neural Information Processing Systems, NeurIPS 2020 - Virtual, Online Duration: 6 Dec 2020 → 12 Dec 2020 |
Fingerprint
Dive into the research topics of 'A stochastic path-integrated differential estimator expectation maximization algorithm'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver