Adaptation in Nonlinear Learning Models for Nonstationary Tasks
نویسندگان
چکیده
The adaptation of individual learning rates is important for many learning tasks, particularly in the case of nonstationary learning environments. Sutton has presented with the Incremental Delta Bar Delta algorithm a versatile method for many tasks. However, this algorithm was formulated only for linear models. A straightforward generalization to nonlinear models is possible, but we show in this work that it poses some obstacles, namely the stability of the learning algorithm. We propose a new self-regulation of the model’s activation which ensures stability. Our algorithm shows better performance than other approaches on a nonstationary benchmark task. Furthermore we show how to derive this algorithm from basic loss functions.
منابع مشابه
Adaptive Online Learning for Nonstationary Problems
An adaptation algorithm for online training is examined. For stationary tasks it can reduce the learning rate to reach the best convergence. Instead of simple annealing, it keeps the learning rate exible, such that it can also adapt to nonstationary tasks. Di erent tasks, abrupt or gradual changes, and di erent guidance measures are discussed.
متن کاملVerification of an Evolutionary-based Wavelet Neural Network Model for Nonlinear Function Approximation
Nonlinear function approximation is one of the most important tasks in system analysis and identification. Several models have been presented to achieve an accurate approximation on nonlinear mathematics functions. However, the majority of the models are specific to certain problems and systems. In this paper, an evolutionary-based wavelet neural network model is proposed for structure definiti...
متن کاملMultiple Model-Based Reinforcement Learning
We propose a modular reinforcement learning architecture for nonlinear, nonstationary control tasks, which we call multiple model-based reinforcement learning (MMRL). The basic idea is to decompose a complex task into multiple domains in space and time based on the predictability of the environmental dynamics. The system is composed of multiple modules, each of which consists of a state predict...
متن کاملActive Learning of Parameterized Skills
We introduce a method for actively learning parameterized skills. Parameterized skills are flexible behaviors that can solve any task drawn from a distribution of parameterized reinforcement learning problems. Approaches to learning such skills have been proposed, but limited attention has been given to identifying which training tasks allow for rapid skill acquisition. We construct a non-param...
متن کاملAn Adaptive Forecasting of Nonlinear Nonstationary Time Series under Short Learning Samples
Methods of nonstationary nonlinear time series forecasting under bounded a priori information provide an interdisciplinary applications area that is concerned with learning and adaptation of solutions from a traditional artificial intelligence point of view. It is extremely difficult to solve this type of problems in its general form, therefore, an approach based on the additive nonlinear auto ...
متن کامل