Darren LI

Many of these projects are saving time by training on small, highly curated datasets. This suggests there is some flexibility in data scaling laws. The existence of such datasets follows from the line of thinking in Data Doesn't Do What You Think, and they are rapidly becoming the standard way to do training outside Google

无模型方法专注于直接从与环境的交互中找出value function。此类中的算法严重依赖reward 来学习值函数。因此,具有学习reward的功能很重要。 此外,它们通常更易于实现和调整超参数。

According to an advertising agency I consulted, for example, a weekly podcast that generates thirty thousand downloads per episode should be able to reach Kelly’s target of generating a hundred thousand dollars a year in income. Earning a middle-class salary by talking through a digital microphone to a fiercely loyal band of supporters around the... See more

线性资本还在内部组成了一个更为年轻、能够快速反应的Bolt投资三人小组。他们负责在具体投资实践中做到三点——

第一,更灵活的投资额度。线性资本投资的典型目标范围是 300 万至 1000 万美元,对数额更大或更小的支票规模可以灵活调整。

第二,更快的决策速度。通过与创始团队的 3-4 次会议,原则上两周内做出决策,整体执行过程不超过一个月。

第三,更友好的条款。承诺提供匹配市场惯例基础上最简洁的条款。

一言以蔽之,“更轻,更快,更灵活”。

This diversity of acquisitions speaks to Google’s willingness to consider just about any tech company that passes its famous “Toothbrush Test,” an M&A litmus test developed by co-founder Larry Page. Under the test, a target company must produce something that you would use once or twice a day, and that would also make your life better.

整个sim-to-real过程如图4所示,共分为四步:

(1)识别出机器人的物理参数,并对机器人进行刚体运动学/动力学建模;

(2)收集真实的关节电机执行数据,训练一个Actuator Net;

(3)在仿真中,利用Actuator Net建模关节电机,并结合第一步中的刚体运动学/动力学建模,进行强化学习;

(4)将第3步中训练得到的策略部署到真机上。