A. Blum and P. Chalasani. Learning switching concepts. In Proc. 5th Annual Workshop on Computational Learning Theory, 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... a target function belonging to a particular class of functions [41, 66] relax the assumption of independently generated examples [1, 7] allow for drift in the generating distribution [4, 8, 33] and relax the assumption that there is a xed relationship between measurements and class labels [6, 12, 39]. Typically, the measurement space X is taken to be some subset of # . For the sake of simplicity,we will almost exclusively deal with the problem of binary classi cation with Y = ##1#. The analysis is often signi cantly more elegant for the binary case than for the multi class case (where ....

A. Blum and P. Chalasani. Learning switching concepts. In Proceedings of the 5th Annual Workshop on Computational Learning Theory, pages 231-242. ACM Press, 1992.

.... for speci c cases (cf. 27] It has also been noticed that in many realistic scenarios, the samples from which the learner has to generalize are not fully speci ed [21, 22] The learning models which have been formulated for studying this type of problems usually assume sometimes implicitly [6] that there is a xed set of relevant variables which are invisible to the learner. In such problems, the learner may only attempt to nd a good probabilistic prediction rule with respect to the visible attributes. However, as observed by Ben David and Dichterman [3] there are many cases in which ....

Blum, A. and Chalasani, P. (1992). Learning switching concepts. In Proceedings of the 5th Annual Workshop on Computational Learning Theory, pages 231-242.

....input x and boolean attributes (y 1 ; ym ) which uses (y 1 ; ym ) to select f i from a set of polynomial functions f 1 ; f k and then computes and outputs f i (x) can be learned, as long as the selector function can be learned. Independent of our work, Blum and Chalasani [6] also consider a model of learning from examples where the examples may be classified according to one of several different concepts. In their model an adversary controls the decision of which concept would be used to classify the next example. Under this model they study the task of learning ....

A. Blum and P. Chalasani. Learning Switching Concepts. In Proc. 5th Annual Workshop on Computational Learning Theory, pages 231--242, 1992.

....concerning noisy information like that of Kearns and Li [18] as well as those dealing with probabilistic concepts like the Kearns Schapire p concepts [19] may be viewed as special cases of our scenario. Another related model is the model of switching concepts introduced by Blum and Chalasani [10]. In that model a probabilistic behavior of the unknown concept is generated by switching between several different (deterministic) concepts. By attributing these switches to the values of some hidden variable, one gets yet another example of a scenario that may be viewed as learning under ....

.... regarded as a mixture of the projected (deterministic) concepts; each projected concept c i is associated with a probability p i such that P i p i = 1 (the probabilities are determined by the underlying distribution over the hidden variables) In this model, presented by Blum and Chalasani in [10], an example is classified using the concept c i with probability p i . Knowing the class from which these concepts are chosen, the learner may try to learn the probabilistic behavior of their mixture. Indeed, Blum and Chalasani show how to learn a mixture of monotone disjunctions. They also ....

[Article contains additional citation context not shown here]

Avrim Blum and Prasad Chalasani. Learning switching concepts. In Proceedings of the 5th Annual Workshop on Computational Learning Theory, pages 231--242, 1992.

....weak union. This kind of non interaction is not new. In order to make reasoning in certain first order logic systems tractable, Dalal and Etherington (1992) consider various restrictions to the interaction permitted among the elements of a set of logical sentences. In the propositional setting, Blum and Chalasani (1992) consider switching concepts. Here there is actually a set of targets, and the current target switches from time to time from one element of the set to another. An example is classified as positive or negative according to the current target. Thus no interaction occurs among the elements in ....

Blum, A. & Chalasani, P. (1992). Learning switching concepts. Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory (pp. 231--242). New York: ACM Press.

.... Mathematisches Institut, Universitat Heidelberg, Im Neuenheimer Feld 294, 69120 Heidelberg, Germany, EU, fstephan math.uni heidelberg.de. 1 Introduction In many machine learning situations, the concepts to be learned or the concepts auxiliarily useful to learn may drift with time [2, 3, 5, 6, 8, 10, 17]. As in the just previous references, to sufficiently track drifting concepts to permit learning something of them at all, it is necessary to consider some restrictions on the nature of the drift. For example, Helmbold and Long [8] bound the probability of disagreement between subsequent concepts. ....

....references, to sufficiently track drifting concepts to permit learning something of them at all, it is necessary to consider some restrictions on the nature of the drift. For example, Helmbold and Long [8] bound the probability of disagreement between subsequent concepts. Blum and Chalasani [3] place some constraints on how many different concepts may be used, or the frequency of concept switches. Bartlett, Ben David and Kulkarni [2] consider class of legal function sequences based on some constraints (such as being formed from a walk on a directed graph) Most of the above work which ....

[Article contains additional citation context not shown here]

A. Blum and P. Chalasani. Learning switching concepts. In Proceedings of the fifth Annual Workshop on Computational Learning Theory, Pittsburgh, Pennsylvania, pages 231--242. ACM Press Computer Society Press, 1992.

.... are not fully specified [28, 30] The importance of this focus of attention problem has been noticed since the emergence of the computational learning theory [1] However, the learning models which have been formulated for studying this type of problems usually assume sometimes implicitly [13] that there is a fixed set of relevant variables which are invisible to the learner. In such problems, the learner may only attempt to find a good probabilistic prediction rule with respect to the visible attributes. However, as observed by Ben David and Dichterman [6] there are many cases in ....

....all the variables) is mapped into its projection on the set of visible variables. With respect to the visible variables, a probabilistic behavior is induced, and the learner tries to learn this behavior. Another related model is the model of switching concepts introduced by Blum and Chalasani [13]. In that model a probabilistic behavior of the unknown concept is generated by switching between several different (deterministic) concepts. By attributing these switches to the values of some hidden variable, one gets yet another example of a scenario that may be viewed as learning under ....

[Article contains additional citation context not shown here]

A. Blum and P. Chalasani. Learning switching concepts. In Proc. 5th Annu. Workshop on Comput. Learning Theory, pages 231--242. ACM Press, New York, NY, 1992.

....clustering scenario. Similar segmentation problems have been addressed by Dom [11] in the context of image segPage mentation, Rissanen and Shedler [20] in the context of identifying stretches of production or short lived items in a factory, and Ron and Freund [13] and Blum and Chalasani [7] in the context of learning from a set of distributions. Most of the proposed algorithms are worse than quadratic, and none deal with identifying segments based only on the drift of the relationship between variables, i.e. potentially ignoring drifts that are well explained by drifts in the ....

A. Blum and P. Chalasani. Learning switching concepts. In Proc. fifth annual workshop on cmputational learning theory, 1992.

....efficient algorithms for finding consistent hypotheses are known. For all classes F , both our result and theirs match Bartlett s [Bar92] fl = O(ffl 2 =VCdim(F ) necessary condition up to log factors. Littlestone and Warmuth [LW94] Kuh, Petsche and Rivest [KPR90, KPR91] Blum and Chalisani [BC92], Herbster and Warmuth [HW95] and Auer and Warmuth [AW95] also studied learning in a changing environment, but in frameworks substantially different from that considered here. The main new idea in our proof of the sufficient conditions is in where the assumption that the distributions are close ....

A. Blum and P. Chalasani. Learning switching concepts. Proceedings of the Fifth Annual Workshop on Computational Learning Theory, pages pages 231--242, 1992.

....formula. We will show that p f can be represented as a k probabilistic decision list with increasing probabilities, a class of p concepts for which there is known to exist an efficient algorithm for approximating the Bayes optimal predictor (Kearns Schapire, 1990) A similar technique is used by Blum and Chalasani (1992). TOWARD EFFICIENT AGNOSTIC LEARNING 299 A k probabilistic decision list (k PDL) over variable set V is a sequence of pairs h(d 1 ; r 1 ) d s ; r s )i where each d i is a conjunction of at most k literals from V and each r i 2 [0; 1] We also require that some d i is the constant ....

Blum, A. & Chalasani, P. (1992). Learning switching concepts. Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory (pp. 231--242).

....consecutive functions is a special case of this model. More recently, Freund and Mansour [7] have investigated learning when the distribution changes as a linear function of time. They present algorithms that estimate the error of functions, using knowledge of this linear drift. Blum and Chalasani [4] consider learning switching concepts. The target concept is allowed to switch between concepts in the class, but with some constraint on the total number of concepts visited, or on the frequency of switches. Their most closely related results concentrate on the computational complexity of ....

A. Blum and P. Chalasani. Learning switching concepts. In Proc. 5th Annu. Workshop on Comput. Learning Theory, pages 231--242. ACM Press, New York, NY, 1992.

....in speech analysis and the use of Hidden Markov Model see Rabiner and Juang [7] to generate a hypothesis segmentation and a set of hypothesis distributions which are close to those of the target. This problem is related to the problem of learning switching concepts studied by Blum and Chalasani [3]. However, in their setup the switching entities are concepts, i.e. mappings from some domain to f0; 1g, while in our setup the switching entities are distributions over a single space. In this work we give an efficient algorithm for learning switching distributions. We describe several variants ....

Avrim Blum and Prasad Chalasani. Learning switching concepts. In Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, pages 231--242, July 1992.

....single sequence S of length M , that is generated by the target, and its goal is to generate a hypothesis segmentation and a set of hypothesis distributions which are close to those of the target. This problem is related to the problem of learning switching concepts studied by Blum and Chalasani [3]. However, in their setup the switching entities are concepts, i.e. mappings from some domain to f0; 1g, while in our setup the switching enti 1 For an introduction on HMMs and their use in speech analysis and the use of Hidden Markov Model see Rabiner and Juang [7] ties are distributions ....

Avrim Blum and Prasad Chalasani. Learning switching concepts. In Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, pages 231--242, July 1992.

....a weaker (time averaged) version of this bound. The final result of the paper, when converted to their setting, shows that with this weaker constraint, the allowable drift rate decreases by no more than a log factor, ffl 2 = d log(1=ffl) versus ffl 2 = d log 2 (d=ffl) Blum and Chalasani [1] consider learning switching concepts. The target concept is allowed to switch between concepts in the class, but with some constraint on the total number of concepts visited, or on the frequency of switches. Their most closely related results concentrate on the computational complexity of ....

A. Blum and P. Chalasani. Learning switching concepts. In Proc. 5th Annu. Workshop on Comput. Learning Theory, pages 231--242. ACM Press, New York, NY, 1992.

No context found.

A. Blum and P. Chalasani. Learning switching concepts. In Proc. 5th Annual Workshop on Computational Learning Theory, 1992.

No context found.

Avrim Blum and Prasad Chalasani. Learning switching concepts. Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, pages 231-242, Pittsburgh, Pennsylvania, 27-29 July 1992. ACM Press.

No context found.

Avrim Blum and Prasad Chalasani. Learning switching concepts. Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, pages 231-242, Pittsburgh, Pennsylvania, 27-29 July 1992. ACM Press.

No context found.

A. Blum and P. Chalasani. Learning Switching Concepts. COLT, 1992.

No context found.

A. Blum and P. Chalasani. Learning switching concepts. Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, pages 231-242, Pittsburgh, Pennsylvania, 27-29 July 1992. ACM Press.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC