This paper explores the use of soft decision trees [1] in basic reinforcement applications to examine the efficacy of using passive-expert like networks for optimal Q-Value learning on Artificial Neural Networks (ANN)...This paper explores the use of soft decision trees [1] in basic reinforcement applications to examine the efficacy of using passive-expert like networks for optimal Q-Value learning on Artificial Neural Networks (ANN). The soft decision tree networks were built using the PyTorch machine learning and the OpenAi’s Gym environment frameworks. The conducted research study aimed at assessing the performance of soft decision tree networks on Cartpole as provided in the OpenAi Gym software package. The baseline performance metric that the soft decision tree networks were compared against was a simple Deep Neural Network using several linear layers with ReLU and Softmax activation functions for the input and output layers, respectively. All networks were trained using the Backpropagation algorithm provided generically by PyTorch’sAutograd module.展开更多
Decision tree(DT)plays an important role in pattern recognition and machine learning,which is widely used for regression tasks because of its natural interpretability.Nevertheless,the traditional decision tree is cons...Decision tree(DT)plays an important role in pattern recognition and machine learning,which is widely used for regression tasks because of its natural interpretability.Nevertheless,the traditional decision tree is constructed by recursive Boolean division.The discrete decision-making process in DT makes it non-differentiable,and causes the problem of hard decision boundary.To solve this problem,a probability distribution model—Staired-Sigmoid is proposed in this paper.The Staired-Sigmoid model is used to differentiate the decision-making process,by which the samples can be assigned to two sub-trees more finely.Based on Staired-Sigmoid,we further propose the soft decision tree(SDT)for regression tasks,where the samples are assigned to different sub-nodes according to a continuous probability distribution.This process is differentiable,and all parameters in SDT can be optimized by gradient descent algorithms.Owing to its constructing rules,SDT is more stable than decision tree,and it is easier to overcome the problem of overfitting.We validate SDT on several datasets obtained from UCI.Experiments demonstrate that SDT achieves better performance than decision tree,and it significantly alleviates the overfitting.展开更多
文摘This paper explores the use of soft decision trees [1] in basic reinforcement applications to examine the efficacy of using passive-expert like networks for optimal Q-Value learning on Artificial Neural Networks (ANN). The soft decision tree networks were built using the PyTorch machine learning and the OpenAi’s Gym environment frameworks. The conducted research study aimed at assessing the performance of soft decision tree networks on Cartpole as provided in the OpenAi Gym software package. The baseline performance metric that the soft decision tree networks were compared against was a simple Deep Neural Network using several linear layers with ReLU and Softmax activation functions for the input and output layers, respectively. All networks were trained using the Backpropagation algorithm provided generically by PyTorch’sAutograd module.
文摘Decision tree(DT)plays an important role in pattern recognition and machine learning,which is widely used for regression tasks because of its natural interpretability.Nevertheless,the traditional decision tree is constructed by recursive Boolean division.The discrete decision-making process in DT makes it non-differentiable,and causes the problem of hard decision boundary.To solve this problem,a probability distribution model—Staired-Sigmoid is proposed in this paper.The Staired-Sigmoid model is used to differentiate the decision-making process,by which the samples can be assigned to two sub-trees more finely.Based on Staired-Sigmoid,we further propose the soft decision tree(SDT)for regression tasks,where the samples are assigned to different sub-nodes according to a continuous probability distribution.This process is differentiable,and all parameters in SDT can be optimized by gradient descent algorithms.Owing to its constructing rules,SDT is more stable than decision tree,and it is easier to overcome the problem of overfitting.We validate SDT on several datasets obtained from UCI.Experiments demonstrate that SDT achieves better performance than decision tree,and it significantly alleviates the overfitting.
