Paper title:
Pegasos: Primal Estimated sub-Gradient Solver for SVM
1. Scope and relevance:
Relevance to the REALM project (1=borderline 5=spot on):
4
Contribution to the community (# citations, work extended by others?):
910 citations
What ML methods are used?
Support vector machine (mainly linear kernel, but there is a discussion of
how to extend the works to nonlinear kernels)
What is the problem class / use case? (anomaly detection, regression, classification etc):
Classification
What is the application domain?
Text classification
What type of data is studied? (time-series, steady-state, static/dynamic):
Static
2. Quality and scientific soundness:
Clarity of the presentation. structure, is the problem well defined?
The paper is well written, and the problem is well defined. However, the
paper assumes a reader has strong backgrounds in convex optimizations.
Are the methods well described/referenced?
Yes
Are the experiments repeatable/extendable?
Yes
Are data sets publicly available?
Yes
Are alternative methods evaluated?
Yes, SVM-Perf, a state-of-the-art SVM solver at the time (2007).