Name
Abstract | ||
|---|---|---|
The state of the art in supervised learning has developed effective models for learning, generalizing, recognizing faces and images, time series prediction, and more. However, most of these powerful models cannot effectively learn incrementally. Infinite Lattice Learner (ILL) is an ensemble model that extends state-of-the-art machine learning methods into incremental learning models. With ILL, even batch models can learn incrementally with exceptional data retention ability. Instead of continually revisiting past instances to retain learned information, ILL allows existing methods to converge on new information without overriding previous knowledge. With ILL, models can efficiently chase a drifting function without continually revisiting a changing dataset. Models wrapped in ILL can operate in continuous real-time environments where millions of unique samples are seen every day. Big datasets too large to fit in memory, or even a single machine, can be learned in portions. ILL utilizes an infinite Cartesian grid of points with an underlying model tiled upon it. Efficient algorithms for discovering nearby points and lazy evaluation make this seemingly impossible task possible. Extensive empirical evaluation reveals impressive retention ability for all ILL models. ILL similarly proves its generalization ability on a variety of datasets from classification and regression to image recognition. |
Year | DOI | Venue |
|---|---|---|
2020 | 10.1007/s00500-019-04330-7 | Soft Computing |
Keywords | DocType | Volume |
Supervised learning, Incremental learning, Ensemble learning, Neural networks | Journal | 24 |
Issue | ISSN | Citations |
9 | 1432-7643 | 0 |
PageRank | References | Authors |
0.34 | 0 | 2 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Justin Lovinger | 1 | 2 | 2.04 |
| Iren Valova | 2 | 136 | 25.44 |