Existing distributional RL algorithms parameterize either the probability side or the return value side of the distribution function, leaving the other side uniformly fixed as in C51, QR-DQN or randomly sampled as in IQN. In this paper, we propose fully parameterized quantile function that ...
Existing distributional RL algorithms parameterize either the probability side or the return value side of the distribution function, leaving the other side uniformly fixed as in C51, QR-DQN or randomly sampled as in IQN. In this paper, we propose fully parameterized quantile functio...
Existing distributional RL algorithms parameterize either the probability side or the return value side of the distribution function, leaving the other side uniformly fixed as in C51, QR-DQN or randomly sampled as in IQN. In this paper, we propose fully parameterized quantile function that ...
C51 parameterizes only the probability side, leaving the value side as uniformly fixed classes. With QR-DQN, researchers turned to the quantile function to capture distribution and parameterized only the value side, or thequantile value, uniformly fixing the probab...
Projects Security Insights Additional navigation options master 2Branches0Tags Code README MIT license PyTorch implementation of the state-of-the-art distributional reinforcement learning algorithm Fully Parameterized Quantile Function (FQF). Implementation includesDQN extensionswith which FQF represents the most...
Fully Parameterized Quantile Function for Distribution Reinforcement Learning Derek Yang, Li Zhao, Zichuan Lin, Tao Qin, Jiang Bian, Tie-yan Liu If you use this code in your research, please cite @inproceedings{yang2019fully, title={Fully Parameterized Quantile Function for Distributional Reinforcement...
function, leaving the other side uniformly fixed as in C51, QR-DQN or randomly sampled as in IQN. In this paper, we propose fully parameterized quantile function that parameterizes both the quantile fraction axis (i.e., the x-axis) and the value axis (i.e., y...