作者: 不能妥協(xié) 時(shí)間: 2025-3-21 22:22
Radial Basis Function Neural Networks Have Superlinear VC Dimension,rons. As the main result we show that every reasonably sized standard network of radial basis function (RBF) neurons has VC dimension Ω([itW ] log .), where . is the number of parameters and . the number of nodes. This significantly improves the previously known linear bound. We also derive superlin作者: transient-pain 時(shí)間: 2025-3-22 01:04
Tracking a Small Set of Experts by Mixing Past Posteriors,ves predictions from a large set of . experts. Its goal is to predict almost as well as the best sequence of such experts chosen off-line by partitioning the training sequence into .+1 sections and then choosing the best expert for each section. We build on methods developed by Herbster and Warmuth 作者: MODE 時(shí)間: 2025-3-22 05:44
Potential-Based Algorithms in Online Prediction and Game Theory,e and Warmuth’s Weighted Majority), for playing iterated games (including Freund and Schapire’s Hedge and MW, as well as the Λ-strategies of Hart and Mas-Colell), and for boosting (including AdaBoost) are special cases of a general decision strategy based on the notion of potential. By analyzing thi作者: Bricklayer 時(shí)間: 2025-3-22 11:45
A Sequential Approximation Bound for Some Sample-Dependent Convex Optimization Problems with Applicons. This analysis is closely related to the regret bound framework in online learning. However we apply it to batch learning algorithms instead of online stochastic gradient decent methods. Applications of this analysis in some classification and regression problems will be illustrated.作者: 前奏曲 時(shí)間: 2025-3-22 16:11 作者: 前奏曲 時(shí)間: 2025-3-22 19:14
Ultraconservative Online Algorithms for Multiclass Problems,pe vector per class. Given an input instance, a multiclass hypothesis computes a similarity-score between each prototype and the input instance and then sets the predicted label to be the index of the prototype achieving the highest similarity. To design and analyze the learning algorithms in this p作者: visual-cortex 時(shí)間: 2025-3-22 21:23 作者: indemnify 時(shí)間: 2025-3-23 03:05
Adaptive Strategies and Regret Minimization in Arbitrarily Varying Markov Environments,is problem is captured by a two-person stochastic game model involving the reward maximizing agent and a second player, which is free to use an arbitrary (non-stationary and unpredictable) control strategy. While the minimax value of the associated zero-sum game provides a guaranteed performance lev作者: 我就不公正 時(shí)間: 2025-3-23 08:53
,Robust Learning — Rich and Poor,d classes T(.) where T is any general recursive operator, are learnable in the sense .. It was already shown before, see [14,19], that for . (learning in the limit) robust learning is rich in that there are classes being both not contained in any recursively enumerable class of recursive functions a作者: Fretful 時(shí)間: 2025-3-23 13:23
On the Synthesis of Strategies Identifying Recursive Functions, of its output values. Uniform learning is concerned with the design of single programs solving infinitely many classical learning problems. For that purpose the program reads a description of an identification problem and is supposed to construct a technique for solving the particular problem..As c作者: VERT 時(shí)間: 2025-3-23 15:36
Intrinsic Complexity of Learning Geometrical Concepts from Positive Data, strategy learning such geometrical concept can be viewed as a sequence of . strategies. Thus, the length of such a sequence together with complexities of primitive strategies used can be regarded as complexity of learning the concept in question. We obtained best possible lower and upper bounds on 作者: Lasting 時(shí)間: 2025-3-23 18:02 作者: Expurgate 時(shí)間: 2025-3-23 23:05
Discrete Prediction Games with Arbitrary Feedback and Loss (Extended Abstract),on the predicted values. This setting can be seen as a generalization of the classical multi-armed bandit problem and accommodates as a special case a natural bandwidth allocation problem. According to the approach adopted by many authors, we give up any statistical assumption on the sequence to be 作者: 束以馬具 時(shí)間: 2025-3-24 02:44
Rademacher and Gaussian Complexities: Risk Bounds and Structural Results,cision theoretic setting, we prove general risk bounds in terms of these complexities. We consider function classes that can be expressed as combinations of functions from basis classes and show how the Rademacher and gaussian complexities of such a function class can be bounded in terms of the comp作者: FLIP 時(shí)間: 2025-3-24 07:49
Further Explanation of the Effectiveness of Voting Methods: The Game between Margins and Weights,ss .. The algorithms of combining simple classifiers into a complex one, such as boosting and bagging, have attracted a lot of attention. We obtain new sharper bounds on the generalization error of combined classifiers that take into account both the empirical distribution of “classification margins作者: 遷移 時(shí)間: 2025-3-24 13:33
Geometric Methods in the Analysis of Glivenko-Cantelli Classes,ko-Cantelli classes for . in terms of the fat-shatteringdimension of the class, which does not depend on the size of the sample. Usingthe new bound, we improve the known sample complexity estimates and bound the size of the Sufficient Statistics needed for Glivenko-Cantelli classes.作者: construct 時(shí)間: 2025-3-24 15:33 作者: ARY 時(shí)間: 2025-3-24 19:21 作者: paleolithic 時(shí)間: 2025-3-24 23:34
The Sequential Analysis of Survival Datalean perceptron that is accurate to within error ε (the fraction of misclassified vectors). This provides a mildly super-polynomial bound on the sample complexity of learning boolean perceptrons in the “restricted focus of attention” setting. In the process we also find some interesting geometrical properties of the vertices of the unit hypercube.作者: 掃興 時(shí)間: 2025-3-25 07:00
Entwicklungen in der Sequentialanalyse some other learning types . are classified as to whether or not they contain rich robustly learnable classes. Moreover, the first results on separating robust learning from uniformly robust learning are derived.作者: neolith 時(shí)間: 2025-3-25 09:15
https://doi.org/10.1007/978-3-642-70093-4s allows uniform solvability of all solvable problems, whereas even the most simple classes of recursive functions are not uniformly learnable without restricting the set of possible descriptions. Furthermore the influence of the hypothesis spaces on uniform learnability is analysed.作者: 吹氣 時(shí)間: 2025-3-25 14:54
Potential-Based Algorithms in Online Prediction and Game Theory,y developed in game theory. By exploiting this connection, we show that certain learning problems are instances of more general game-theoretic problems. In particular, we describe a notion of generalized regret and show its applications in learning theory.作者: languor 時(shí)間: 2025-3-25 16:59
Estimating a Boolean Perceptron from Its Average Satisfying Assignment: A Bound on the Precision Relean perceptron that is accurate to within error ε (the fraction of misclassified vectors). This provides a mildly super-polynomial bound on the sample complexity of learning boolean perceptrons in the “restricted focus of attention” setting. In the process we also find some interesting geometrical properties of the vertices of the unit hypercube.作者: NIB 時(shí)間: 2025-3-25 23:05
,Robust Learning — Rich and Poor, some other learning types . are classified as to whether or not they contain rich robustly learnable classes. Moreover, the first results on separating robust learning from uniformly robust learning are derived.作者: 違法事實(shí) 時(shí)間: 2025-3-26 01:38
On the Synthesis of Strategies Identifying Recursive Functions,s allows uniform solvability of all solvable problems, whereas even the most simple classes of recursive functions are not uniformly learnable without restricting the set of possible descriptions. Furthermore the influence of the hypothesis spaces on uniform learnability is analysed.作者: Arctic 時(shí)間: 2025-3-26 05:14
,Strukturelle Globalit?t auf globaler Ebene,” and the “approximate dimension” of the classifier, which is defined in terms of weights assigned to base classifiers by a voting algorithm. We study the performance of these bounds in several experiments with learning algorithms.作者: HUMP 時(shí)間: 2025-3-26 09:16
,über die Struktur amorpher Polymere, a similar analysis, we improve on sufficient conditions for a class of real-valued functions to be agnostically learnable with a particular relative accuracy; in particular, we improve by a factor of two the scale at which scale-sensitive dimensions must be finite in order to imply learnability.作者: 救護(hù)車 時(shí)間: 2025-3-26 16:16
Ulrich P?tzold,Horst R?per,Helmut Volpers are pruning classifier ensembles using WM and learning general DNF formulas using Winnow. These uses require exponentially many inputs, so we define Markov chains over the inputs to approximate the weighted sums. We state performance guarantees for our algorithms and present preliminary empirical results.作者: 榨取 時(shí)間: 2025-3-26 20:33
Elke van der Meer,Matthias Kolbeons of functions from basis classes and show how the Rademacher and gaussian complexities of such a function class can be bounded in terms of the complexity of the basis classes.We give examples of the application of these techniques in finding data-dependent risk bounds for decision trees, neural networks and support vector machines.作者: GULLY 時(shí)間: 2025-3-26 21:14 作者: Hormones 時(shí)間: 2025-3-27 04:28
Rademacher and Gaussian Complexities: Risk Bounds and Structural Results,ons of functions from basis classes and show how the Rademacher and gaussian complexities of such a function class can be bounded in terms of the complexity of the basis classes.We give examples of the application of these techniques in finding data-dependent risk bounds for decision trees, neural networks and support vector machines.作者: Sputum 時(shí)間: 2025-3-27 06:44
Further Explanation of the Effectiveness of Voting Methods: The Game between Margins and Weights,” and the “approximate dimension” of the classifier, which is defined in terms of weights assigned to base classifiers by a voting algorithm. We study the performance of these bounds in several experiments with learning algorithms.作者: Affirm 時(shí)間: 2025-3-27 11:35 作者: 分貝 時(shí)間: 2025-3-27 14:24
Eine US-amerikanische Zivilgesellschaft?, in both cases turn out to be much lower than those provided by natural learning strategies. Another surprising result is that learning intersections of open semi-hulls (and their complements) turns out to be easier than learning open semi-hulls themselves.作者: Kinetic 時(shí)間: 2025-3-27 18:19
Intrinsic Complexity of Learning Geometrical Concepts from Positive Data, in both cases turn out to be much lower than those provided by natural learning strategies. Another surprising result is that learning intersections of open semi-hulls (and their complements) turns out to be easier than learning open semi-hulls themselves.作者: Asperity 時(shí)間: 2025-3-27 22:39
How Many Queries Are Needed to Learn One Bit of Information?,nd learning by counterexamples (equivalence queries alone). These parameters are finally used to characterize the additional power provided by membership queries (compared to the power of equivalence queries alone). All investigations are purely information-theoretic and ignore computational issues.作者: LAVE 時(shí)間: 2025-3-28 02:18 作者: antiandrogen 時(shí)間: 2025-3-28 08:46
Tracking a Small Set of Experts by Mixing Past Posteriors,for choosing the best expert in each section we first pay log (.) bits in the bounds for identifying the pool of . experts and then logm bits per new section. In the bounds we also pay twice for encoding the boundaries of the sections.作者: Paraplegia 時(shí)間: 2025-3-28 10:47
Ultraconservative Online Algorithms for Multiclass Problems, We then discuss a specific online algorithm that seeks a set of prototypes which have a small norm. The resulting algorithm, which we term MIRA (for Margin Infused Relaxed Algorithm) is ultraconservative as well. We derive mistake bounds for all the algorithms and provide further analysis of MIRA u作者: Malleable 時(shí)間: 2025-3-28 15:02 作者: 古文字學(xué) 時(shí)間: 2025-3-28 20:52 作者: 原始 時(shí)間: 2025-3-29 02:46
Discrete Prediction Games with Arbitrary Feedback and Loss (Extended Abstract),constructively, that is when the loss and feedback functions satisfya certain condition, we present an algorithm that generates predictions with the claimed performance; otherwise we show a sequence that no algorithm can predict without incurring a linear regret with probability at least 1/2.作者: monopoly 時(shí)間: 2025-3-29 04:53 作者: myelography 時(shí)間: 2025-3-29 07:45
https://doi.org/10.1007/978-3-658-03986-8 these networks is superlinear as well, and they yield lower bounds even when the input dimension is fixed. The methods developed here appear suitable for obtaining similar results for other kernel-based function classes.作者: 懶惰民族 時(shí)間: 2025-3-29 14:07 作者: atopic 時(shí)間: 2025-3-29 17:17
Ulrich P?tzold,Horst R?per,Helmut Volpers We then discuss a specific online algorithm that seeks a set of prototypes which have a small norm. The resulting algorithm, which we term MIRA (for Margin Infused Relaxed Algorithm) is ultraconservative as well. We derive mistake bounds for all the algorithms and provide further analysis of MIRA u作者: Phenothiazines 時(shí)間: 2025-3-29 21:56
The Sequential Analysis of Survival Data opponent’s actions. This paper presents an extension of these ideas to problems with Markovian dynamics, under appropriate recurrence conditions. The Bayes envelope is first defined in a natural way in terms of the observed state action frequencies. As this envelope may not be attained in general, 作者: Ataxia 時(shí)間: 2025-3-30 03:34
Der vergesellschaftete Kapitalismus,es — such as the identification of “hostile” contributors and their data — are brought to light by the Open Mind Initiative, where data is openly contributed over the World Wide Web by non-experts of varying reliabilities. This paper states generalizations of formal results on the relative value of 作者: 不感興趣 時(shí)間: 2025-3-30 07:02
Peter Gerjets,Rainer Westermannconstructively, that is when the loss and feedback functions satisfya certain condition, we present an algorithm that generates predictions with the claimed performance; otherwise we show a sequence that no algorithm can predict without incurring a linear regret with probability at least 1/2.作者: CLEFT 時(shí)間: 2025-3-30 10:06
A Sequential Approximation Bound for Some Sample-Dependent Convex Optimization Problems with Applicons. This analysis is closely related to the regret bound framework in online learning. However we apply it to batch learning algorithms instead of online stochastic gradient decent methods. Applications of this analysis in some classification and regression problems will be illustrated.作者: 自然環(huán)境 時(shí)間: 2025-3-30 14:56
Geometric Methods in the Analysis of Glivenko-Cantelli Classes,ko-Cantelli classes for . in terms of the fat-shatteringdimension of the class, which does not depend on the size of the sample. Usingthe new bound, we improve the known sample complexity estimates and bound the size of the Sufficient Statistics needed for Glivenko-Cantelli classes.作者: 否決 時(shí)間: 2025-3-30 17:06 作者: NIP 時(shí)間: 2025-3-30 20:58
https://doi.org/10.1007/978-3-658-03986-8rons. As the main result we show that every reasonably sized standard network of radial basis function (RBF) neurons has VC dimension Ω([itW ] log .), where . is the number of parameters and . the number of nodes. This significantly improves the previously known linear bound. We also derive superlin作者: 預(yù)示 時(shí)間: 2025-3-31 04:09
, – Strukturen sozialer Unterstützungves predictions from a large set of . experts. Its goal is to predict almost as well as the best sequence of such experts chosen off-line by partitioning the training sequence into .+1 sections and then choosing the best expert for each section. We build on methods developed by Herbster and Warmuth 作者: 沒(méi)血色 時(shí)間: 2025-3-31 07:25
Forschungsgegenstand und Forschungsfragene and Warmuth’s Weighted Majority), for playing iterated games (including Freund and Schapire’s Hedge and MW, as well as the Λ-strategies of Hart and Mas-Colell), and for boosting (including AdaBoost) are special cases of a general decision strategy based on the notion of potential. By analyzing thi作者: blight 時(shí)間: 2025-3-31 11:15 作者: 偶然 時(shí)間: 2025-3-31 14:22 作者: AGATE 時(shí)間: 2025-3-31 19:55 作者: 青春期 時(shí)間: 2025-3-31 22:02
The Sequential Analysis of Survival Data whether the vector’s components satisfy some linear inequality. In 1961, Chow [.] showed that any boolean perceptron is determined by the average or “center of gravity” of its “true” vectors (those that are mapped to 1). Moreover, this average distinguishes the function from any other boolean funct作者: 要素 時(shí)間: 2025-4-1 04:06
The Sequential Analysis of Survival Datais problem is captured by a two-person stochastic game model involving the reward maximizing agent and a second player, which is free to use an arbitrary (non-stationary and unpredictable) control strategy. While the minimax value of the associated zero-sum game provides a guaranteed performance lev作者: dominant 時(shí)間: 2025-4-1 09:30
Entwicklungen in der Sequentialanalysed classes T(.) where T is any general recursive operator, are learnable in the sense .. It was already shown before, see [14,19], that for . (learning in the limit) robust learning is rich in that there are classes being both not contained in any recursively enumerable class of recursive functions a作者: prostate-gland 時(shí)間: 2025-4-1 10:48
https://doi.org/10.1007/978-3-642-70093-4 of its output values. Uniform learning is concerned with the design of single programs solving infinitely many classical learning problems. For that purpose the program reads a description of an identification problem and is supposed to construct a technique for solving the particular problem..As c
