With the further development in the fields of computer vision, network security, natural language processing and so on so forth, deep learning technology gradually exposed certain security risks. The existing deep lea...
详细信息
We analyze the performance of top-down algorithms for decision tree learning, such as those employed by the widely used C4.5 and CART software packages. Our main result is a proof that such algorithms are boosting alg...
详细信息
ISBN:
(纸本)9780897917858
We analyze the performance of top-down algorithms for decision tree learning, such as those employed by the widely used C4.5 and CART software packages. Our main result is a proof that such algorithms are boosting algorithms. By this we mean that if the functions that label the internal nodes of the decision tree can weakly approximate the unknown target function, then the top-down algorithms we study will amplify this weak advantage to build a tree achieving any desired level of accuracy. The bounds we obtain for this amplification show an interesting dependence on the splitting criterion used by the top-down algorithm. More precisely, if the functions used to label the internal nodes have error 1/2 - γ as approximations to the target function, then for the splitting criteria used by CART and C4.5, trees of size (1/Ε)O(1/γ(2)Ε(2)) and (1/Ε)O(log(1/Ε)/γ(2)) (respectively) suffice to drive the error below Ε. Thus (for example), small constant advantage over random guessing is amplified to constant error with trees of constant size. For a new splitting criterion suggested by our analysis, the much stronger bound of (1/Ε)O(1/γ(2)) (which is polynomial in 1/Ε) is obtained. The differing bounds have a natural explanation in terms of concavity properties of the splitting criterion. The primary contribution of this work is in proving that some popular and empirically successful heuristics that are based on first principles meet the criteria of an independently motivated theoretical model.
Improving the approach to the development of algorithms that allow machine agents to act independently within a circumscribed set of goals has garnered more traction, over recent years, than just a phase of observatio...
详细信息
One of the most fascinating challenges in the context of parton density function (PDF) is the determination of the best combined PDF uncertainty from individual PDF sets. Since 2014 multiple methodologies have been de...
详细信息
Using AI to fight fake news efficiently. Using AI to fight misinformation with more modern approaches from machine learning and natural language processing. This paper discusses on extraction, model selection, trainin...
详细信息
Imagination is the fundamental of creating something new. Humans can invent things if they can imagine that. So human imagination is the base of all evolution including science and technology. Humans produce their ima...
详细信息
Phishing is a criminal act in which a Phisher creates almost identical website connections exploiting URL Lexical characteristics to dupe unsuspecting users into exposing sensitive information such as financial data, ...
详细信息
Advancements in technology and computational power have significantly diversified the field of medical sciences, especially in diagnosing human cardiac disorders, which is currently one of the most severe cardiac illn...
详细信息
The article describes the main approaches to the automated determination of authorship of texts based on the analysis of copyright styles. The architectures of a convolutional neural network, a multilayer perceptron, ...
详细信息
Recent theoretical results have shown that improved bounds on generalization error of classifiers can be obtained by explicitly taking the observed margin distribution of the training data into account. Currently, alg...
详细信息
ISBN:
(纸本)1577351894
Recent theoretical results have shown that improved bounds on generalization error of classifiers can be obtained by explicitly taking the observed margin distribution of the training data into account. Currently, algorithms used in practice do not make use of the margin distribution and are driven by optimization with respect to the points that are closest to the hyperplane. This paper enhances earlier theoretical results and derives a practical data-dependent complexity measure for learning. The new complexity measure is a function of the observed margin distribution of the data, and cab be used, as we show, as a model selection criterion. We then present the Margin Distribution Optimization (MDO) learning algorithm, that directly optimizes this complexity measures. Empirical evaluation of MDO demonstrates that it consistently outperforms SVM.
暂无评论