A Kernel Method for the Optimization of the Margin Distribution
read more
Citations
Recent Advances in Open Set Recognition: A Survey
The Extreme Value Machine
EasyMKL: a scalable multiple kernel learning algorithm
Large margin distribution machine
Large Margin Distribution Machine
References
Soft Margins for AdaBoost
A fast iterative nearest point algorithm for support vector machine classifier design
How boosting the margin can also boost classifier complexity
A Mathematical Programming Approach to the Kernel Fisher Algorithm
Related Papers (5)
Frequently Asked Questions (9)
Q2. What are the future works in "A kernel method for the optimization of the margin distribution" ?
In future work, the authors would like to study under which conditions ( e. g. conditions related to the data distribution ) their method is to prefer to other state-of-the-art methods. Htm and its optimization could be another direction of their future research. All datasets can be downloaded from: http: //ida. first.
Q3. What is the effect of the margin distribution on the training set?
Despite the AdaBoost ability to optimize the margin distribution on the training set, it has been shown in [1] that in certain cases, it can also increase the complexity of the weak hypotheses, thus possibly leading to overfitting phenomena.
Q4. What is the ch(C) of a set C?
The convex hull ch(C) of a set C = {c1, . . . , cm|ci ∈ Rd}, is the set of all affine combinations of points in C such that the weights γi of the combination are non-negative, i.e. ch(C) = {γ1c1 + · · ·+γmcm|γi ∈ Γm}.
Q5. What is the way to solve the problem?
||v(γ)||2 + λ||γ||2 (2)It can be shown that the optimal vector v(γ̂) which is solution of the problem above, represents the vector joining two points, v+ into the positive norm-restricted convex hull, i.e. v+ ∈ chη(S+), and v− into the negative norm-restricted convex hull, i.e. v− ∈ chη(S−), for opportune η.
Q6. What is the difference between the two plots?
The two plots clearly show that learning with low λ values requires more training time, whereas models for higher λ values are faster to compute.
Q7. What is the purpose of this article?
Much of last-decade theoretical work on learning machines has been devoted to study the aspects of learning methods that control the generalization performance.
Q8. What is the way to train a model?
It is worth noting that, in most cases, even high λ values (for which the models are much faster to train) give anyway good performances, or at least acceptable when the computational time is an issue.
Q9. What is the simplest way to control the pay-off function of the game?
the authors propose to slightly modify the pay-off function of the game in order to have a flexible way to control the optimization w.r.t. the distribution of the margin in the training set.