论文标题
Stein自我抑制动力：从过去的样本中受益
Stein Self-Repulsive Dynamics: Benefits From Past Samples
论文作者
论文摘要
我们提出了一种新的Stein自我抑制动力学，用于从棘手的非归一化分布中获取多样化的样品。我们的想法是将Stein变分梯度作为一种排斥力量，以将Langevin Dynamics的样本推向过去的轨迹。这个简单的想法使我们能够显着降低Langevin动力学的自动相关性，从而增加有效的样本量。重要的是，正如我们在理论分析中建立的那样，由于Stein变异梯度的特殊特性，渐近的固定分布即使在增加排斥力的情况下仍然正确。我们对新算法进行广泛的经验研究，表明我们的方法比香草兰格文动力学产生的样本效率更高，不确定性估计更好。
We propose a new Stein self-repulsive dynamics for obtaining diversified samples from intractable un-normalized distributions. Our idea is to introduce Stein variational gradient as a repulsive force to push the samples of Langevin dynamics away from the past trajectories. This simple idea allows us to significantly decrease the auto-correlation in Langevin dynamics and hence increase the effective sample size. Importantly, as we establish in our theoretical analysis, the asymptotic stationary distribution remains correct even with the addition of the repulsive force, thanks to the special properties of the Stein variational gradient. We perform extensive empirical studies of our new algorithm, showing that our method yields much higher sample efficiency and better uncertainty estimation than vanilla Langevin dynamics.
