Machine learning(ML) makes machines independent and self-learning component. Researchers applying machine learning algorithms to solve various real word problems in various domains. Nowadays agriculture affects by var...
详细信息
learning algorithms were implemented for the elimination of strong hiss found in old records and of impulse noise affecting transmitted audio signals. The rough-set method was tested with regard to the automatic setti...
详细信息
learning algorithms were implemented for the elimination of strong hiss found in old records and of impulse noise affecting transmitted audio signals. The rough-set method was tested with regard to the automatic setting of the cutoff threshold in the spectral filtration of noisy audio. Applied methods, results of musical signal processing, and conclusions are presented.
This paper is focused on the learning algorithms for dynamic multilayer perceptron neural networks where each neuron synapsis is modelled by an infinite impulse response (IIR) filter (IIR MLP). In particular, the Back...
详细信息
ISBN:
(纸本)0780327683;0780327691
This paper is focused on the learning algorithms for dynamic multilayer perceptron neural networks where each neuron synapsis is modelled by an infinite impulse response (IIR) filter (IIR MLP). In particular, the Backpropagation Through Time (BPTT) algorithm and its less demanding approximated on-line versions are considered. In fact it is known that the BPTT algorithm is not causal and therefore can be implemented only in batch mode, while many real problems require on-line adaptation. In this paper we give the complete BPTT formulation for the IIR MLP, derive an already known on-line learning algorithm as a particular approximation of the BPTT, and propose a new approximated algorithm. Several computer simulations of identification of dynamical systems will also be presented to assess the performance of the approximated algorithms and to compare the IIR MLP with more traditional dynamic networks.
We consider some generalizations of the classical LMS learning algorithm including the Exponentiated Gradient (EG) algorithm. We show how one can develop these algorithms in terms of a prior distribution over the weig...
详细信息
ISBN:
(纸本)0780358007
We consider some generalizations of the classical LMS learning algorithm including the Exponentiated Gradient (EG) algorithm. We show how one can develop these algorithms in terms of a prior distribution over the weight space. Our framework subsumes the notion of "link-functions". Differential geometric methods are used to develop the algorithms as gradient descent with respect to the natural gradient in the Riemannian structure induced by the prior distribution. This provides a Bayesian Riemannian interpretation of the EG and related algorithms. We relate our work to that of Amari and others who used similar tools in a different manner. Simulation experiments illustrating the behaviour of the new algorithms are presented.
We have designed several new lazy learning algorithms for learning problems with many binary features and classes. This particular type of learning task can be found in many machine learning applications but is of spe...
详细信息
ISBN:
(纸本)3540649921
We have designed several new lazy learning algorithms for learning problems with many binary features and classes. This particular type of learning task can be found in many machine learning applications but is of special importance for machine learning of natural language. Besides pure instance-based learning we also consider prototype-based learning, which has the big advantage of a large reduction of the required memory and processing time for classification. As an application for our learning algorithms we have chosen natural language database interfaces. In our interface architecture the machine learning module replaces an elaborate semantic analysis component. The learning task is to select the correct command class based on semantic features extracted from the user input. We use an existing German natural language interface to a production planning and control system as a case study for our evaluation and compare the results achieved by the different lazy learning algorithms.
Generative adversarial imitation learning (GAIL) is a popular inverse reinforcement learning approach for jointly optimizing policy and reward from expert trajectories. A primary question about GAIL is whether applyin...
详细信息
Generative adversarial imitation learning (GAIL) is a popular inverse reinforcement learning approach for jointly optimizing policy and reward from expert trajectories. A primary question about GAIL is whether applying a certain policy gradient algorithm to GAIL attains a global minimizer (i.e., yields the expert policy), for which existing understanding is very limited. Such global convergence has been shown only for the linear (or linear-type) MDP and linear (or linearizable) reward. In this paper, we study GAIL under general MDP and for nonlinear reward function classes (as long as the objective function is strongly concave with respect to the reward parameter). We characterize the global convergence with a sublinear rate for a broad range of commonly used policy gradient algorithms, all of which are implemented in an alternating manner with stochastic gradient ascent for reward update, including projected policy gradient (PPG)-GAIL, Frank-Wolfe policy gradient (FWPG)-GAIL, trust region policy optimization (TRPO)-GAIL and natural policy gradient (NPG)-GAIL. This is the first systematic theoretical study of GAIL for global convergence.
When two different parties use the same learning rule on their own data, how can we test whether the distributions of the two outcomes are similar? In this paper, we study the similarity of outcomes of learning rules ...
详细信息
When two different parties use the same learning rule on their own data, how can we test whether the distributions of the two outcomes are similar? In this paper, we study the similarity of outcomes of learning rules through the lens of the Total Variation (TV) distance of distributions. We say that a learning rule is TV indistinguishable if the expected TV distance between the posterior distributions of its outputs, executed on two training data sets drawn independently from the same distribution, is small. We first investigate the learnability of hypothesis classes using TV indistinguishable learners. Our main results are information-theoretic equivalences between TV indistinguishability and existing algorithmic stability notions such as replicability and approximate differential privacy. Then, we provide statistical amplification and boosting algorithms for TV indistinguishable learners.
In this paper, we study the generalization properties of Model-Agnostic Meta-learning (MAML) algorithms for supervised learning problems. We focus on the setting in which we train the MAML model over m tasks, each wit...
详细信息
ISBN:
(纸本)9781713845393
In this paper, we study the generalization properties of Model-Agnostic Meta-learning (MAML) algorithms for supervised learning problems. We focus on the setting in which we train the MAML model over m tasks, each with n data points, and characterize its generalization error from two points of view: First, we assume the new task at test time is one of the training tasks, and we show that, for strongly convex objective functions, the expected excess population loss is bounded by O (1/mn). Second, we consider the MAML algorithm's generalization to an unseen task and show that the resulting generalization error depends on the total variation distance between the underlying distributions of the new task and the tasks observed during the training process. Our proof techniques rely on the connections between algorithmic stability and generalization bounds of algorithms. In particular, we propose a new definition of stability for meta-learning algorithms, which allows us to capture the role of both the number of tasks m and number of samples per task n on the generalization error of MAML.
Gradient descent training of sigmoidal feed-forward neural networks on binary mappings often gets stuck with someout puts totally wrong. This is because a sum-squared-error cost function leads to weight updates that d...
详细信息
Gradient descent training of sigmoidal feed-forward neural networks on binary mappings often gets stuck with someout puts totally wrong. This is because a sum-squared-error cost function leads to weight updates that depend on the derivative of the output sigmoid which goes to zero as the output approaches maximal error. Although it is easy to understand the cause, the best remedy is not so obvious. Common solutions involve modifying the training data, deviating from true gradient descent, or changing the cost function. In general, finding the best learning procedures for particular classes of problem is difficult because each usually depends on a number of interacting parameters that need to be set to optimal values for a fair comparison. In this paper I shall use simulated evolution to optimise all the relevant parameters, and come to a clear conclusion concerning the most efficient approach for learning binary mappings. (C) 2003 Elsevier Science Ltd. All rights reserved.
Modelling a dynamical system is a crucial step in the design of a control law. Corresponding to a dynamical system there is a set of models, and the designer chooses one of them. In the case of nonlinear dynamical sys...
详细信息
ISBN:
(纸本)9781467365406
Modelling a dynamical system is a crucial step in the design of a control law. Corresponding to a dynamical system there is a set of models, and the designer chooses one of them. In the case of nonlinear dynamical systems, a possible way to obtain a model that can be used as well for control is an artificial neural network. Accordingly, the performance of the overall system depends on how fast the network can be trained. In this paper, we compare two classes of learning algorithms and discuss their use in the context of identification and control.
暂无评论