A Library for Online Learning Algorithms

What is LIBOL?

LIBOL is an open-source machine learning library that consists of a family of classical and state-of-the-art online learning algorithms for large-scale machine learning and data mining research. It includes two categories of online learning methods: regular linear online learning algorithms and kernel-based online learning algorithms.

Linear methods

LIBOL consists of a large family of state-of-the-art linear onliine learning algorithms in two major categories: (i) first-order online learning, and (ii) second-order online learning. The list of linear algoirthms is summarzed below:

  • First-order Alogirhtms
    • Perceptron
    • Passive-Aggressive (PA) variants (PA, PA-I, PA-II)
    • Online Gradient Descent (OGD)
  • Second-order Alogirhtms
    • Second-order Perceptron (SOP)
    • Confidence-weighted learning (CW)
    • AROW: Adaptive Regularization Of Weights
    • SCW: Soft Confidence Weighted learning

Kernel methods

The list of kernel-based algorithms is summarized below:

  • Unbounded Algorithms
    • Kernel Perceptron
    • Kernel Passive-Aggressive (PA) variants (PA, PA-I, PA-II)
    • Kernel Online Gradient Descent (OGD)
  • Bounded Algorithms
    • Forgetron
    • RBP: Randomized Budget Perceptron
    • Projectron vairants (Projectron and Projectron++)
    • Bounded Online Gradient Descent (BOGD and BOGD++)