Abstract
A typical online kernel learning method faces two fundamental issues: the complexity in dealing with a huge number of observed data points (a.k.a the curse of kernelization) and the difficulty in learning kernel parameters, which often assumed to be fixed. Random Fourier feature is a recent and effective approach to address the former by approximating the shift-invariant kernel function via Bocher's theorem, and allows the model to be maintained directly in the random feature space with a fixed dimension, hence the model size remains constant w.r.t. data size. We further introduce in this paper the reparameterized random feature (RRF), a random feature framework for large-scale online kernel learning to address both aforementioned challenges. Our initial intuition comes from the so-called 'reparameterization trick' [Kingma and Welling, 2014] to lift the source of randomness of Fourier components to another space which can be independently sampled, so that stochastic gradient of the kernel parameters can be analytically derived. We develop a well-founded underlying theory for our method, including a general way to reparameterize the kernel, and a new tighter error bound on the approximation quality. This view further inspires a direct application of stochastic gradient descent for updating our model under an online learning setting. We then conducted extensive experiments on several large-scale datasets where we demonstrate that our work achieves state-of-the-art performance in both learning efficacy and efficiency.
Original language | English |
---|---|
Title of host publication | Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI-17) |
Editors | Carles Sierra |
Place of Publication | Marina del Rey CA USA |
Publisher | Association for the Advancement of Artificial Intelligence (AAAI) |
Pages | 2543-2549 |
Number of pages | 7 |
ISBN (Electronic) | 9780999241103 |
ISBN (Print) | 9780999241110 |
DOIs | |
Publication status | Published - 1 Jan 2017 |
Externally published | Yes |
Event | International Joint Conference on Artificial Intelligence 2017 - Melbourne, Australia Duration: 19 Aug 2017 → 25 Aug 2017 Conference number: 26th https://ijcai-17.org/ https://www.ijcai.org/Proceedings/2017/ (Proceedings) |
Conference
Conference | International Joint Conference on Artificial Intelligence 2017 |
---|---|
Abbreviated title | IJCAI 2017 |
Country/Territory | Australia |
City | Melbourne |
Period | 19/08/17 → 25/08/17 |
Internet address |