Sigmoid-Weighted Linear Units for Neural Network Function Approximation in Reinforcement Learning
Explore this paper's citation graph
Summary
This study proposes two activation functions for neural network function approximation in reinforcement learning: the sigmoid-weighted linear unit (SiLU) and its derivative function (dSiLU), and suggests the more traditional approach of using on-policy learning with eligibility traces, instead of experience replay, and softmax action selection can be competitive with DQN, without the need for a separate target network.
- Type
- preprint
- Published
- 2017-02-10
- Cited by
- 2,692
- References
- 32
- Access
- Open access
- OpenAlex
- https://openalex.org/W2594171637
- Semantic Scholar
- https://api.semanticscholar.org/CorpusID:6940861
Keywords
Reinforcement learning, Softmax function, Sigmoid function, Computer science, Artificial neural network
References
- Issues in Using Function Approximation for Reinforcement Learning
- Improvements on Learning Tetris with Cross Entropy
- Digital selection and analogue amplification coexist in a cortex-inspired silicon circuit
- Massively Parallel Methods for Deep Reinforcement Learning
- Information processing in dynamical systems: foundations of harmony theory
- Approximate modified policy iteration and its application to the game of Tetris
- High-Dimensional Function Approximation for Knowledge-Free Reinforcement Learning: a Case Study in SZ-Tetris
- TD-Gammon, a Self-Teaching Backgammon Program, Achieves Master-Level Play
- Expected energy-based restricted Boltzmann machine for classification
- Neural Network Ensembles in Reinforcement Learning
- Learning to predict by the methods of temporal differences
- ImageNet: A large-scale hierarchical image database
- Training Products of Experts by Minimizing Contrastive Divergence
- Reinforcement Learning: An Introduction
- Approximate Dynamic Programming Finally Performs Well in the Game of Tetris
- Human-level control through deep reinforcement learning
- The Arcade Learning Environment: An Evaluation Platform for General Agents
- Deep Reinforcement Learning with Double Q-Learning
- Unsupervised learning of distributions on binary vectors using two layer networks
- Dueling Network Architectures for Deep Reinforcement Learning
Cited by
- Online meta-learning by parallel algorithm competition
- Searching for Activation Functions
- Rethinking Spatiotemporal Feature Learning For Video Understanding
- The Emergence of Spectral Universality in Deep Networks
- LiDAR Data Classification Using Morphological Profiles and Convolutional Neural Networks
- Metatrace: Online Step-size Tuning by Meta-gradient Descent for Reinforcement Learning Control
- Post-Specialisation: Retrofitting Vectors of Words Unseen in Lexical Resources
- On the Selection of Initialization and Activation Function for Deep Neural Networks
- ARiA: Utilizing Richard's Curve for Controlling the Non-monotonicity of the Activation Function in Deep Neural Nets
- Habitability classification of exoplanets: a machine learning insight
- A New Activation Function for Artificial Neural Net Based Habitability Classification
- q-Neurons: Neuron Activations Based on Stochastic Jackson’s Derivative Operators
- Evolving indirectly encoded convolutional neural networks to play tetris with low-level features
- On decision regions of narrow deep neural networks
- Unbounded Output Networks for Classification
- Flatten-T Swish: a thresholded ReLU-Swish-like activation function for deep learning
- The Quest for the Golden Activation Function
- Collaborative Self-Regression Method With Nonlinear Feature Based on Multi-Task Learning for Image Classification
- Cooperative and Competitive Reinforcement and Imitation Learning for a Mixture of Heterogeneous Learning Modules
- Weighted Sigmoid Gate Unit for an Activation Function of Deep Neural Network
Related papers
- Convolutional neural network for pothole detection in different road and weather conditions
- Evaluation of Sigmoid and ReLU Activation Functions Using Asymptotic Method
- The Generalized Sigmoid Activation Function: Competetive Supervised Learning
- RSigELU: A nonlinear activation function for deep neural networks
- Weighted Sigmoid Gate Unit for an Activation Function of Deep Neural Network
- Improving the Accuracy of Deep Neural Networks Through Developing New Activation Functions
- DIGIT RECOGNITION MODEL USING NEURAL NETWORKS A COMPARISON BETWEEN SIGMOID AND SOFTMAX AS NON-LINEAR ACTIVATION FUNCTIONS