Derives SMO algorithm for ε-SVR with MAPE loss handling sample-dependent constraints, proves invariance leaving most components unchanged, adds efficiency mechanisms, fixes convergence for odd-symmetry kernels, and shows faster runtimes than standard solvers on benchmarks.
Working set selection using second order information for training support vector machines
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.NA 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Sequential Minimal Optimization for $\varepsilon$-SVR with MAPE Loss and Sample-Dependent Box Constraints
Derives SMO algorithm for ε-SVR with MAPE loss handling sample-dependent constraints, proves invariance leaving most components unchanged, adds efficiency mechanisms, fixes convergence for odd-symmetry kernels, and shows faster runtimes than standard solvers on benchmarks.