引用davmre · 2015-10
“such as mutual information between the agents' action sequence and future state, e.g. Variational Information Maximisation for Intrinsically Motivated Reinforcement Learning”
举例:以行动序列与未来状态的互信息作为内在奖励
用变分下界和卷积网络,从像素端到端优化赋能
互信息优化多依赖指数复杂度的 Blahut-Arimoto 算法,不适合现代机器学习。论文把变分推断与深度学习结合,提出可扩展的互信息优化方法,并以内在动机学习中的“赋能(empowerment)”为切入点:利用互信息的变分下界与卷积网络处理视觉输入,得到随机优化算法,可直接从像素到动作做基于赋能的推理。
评论者引用它作为基于互信息(如动作序列与未来状态之间)的内在奖励的例子,并称它是 DeepMind 研究者在这一方向上的近期工作。
“such as mutual information between the agents' action sequence and future state, e.g. Variational Information Maximisation for Intrinsically Motivated Reinforcement Learning”
“A recent paper by 2 Google Deepmind researchers on this topic:”
“DeepMind has been tracking this general problem for a while [1]”