近期生活更新 | Recent Life Update
Author Note: Scroll down for english version text
好久没做近期生活更新啦，最近有点忙，所以就一直拖着没写。
这一个月是刚刚开学的一个月，作为一个大三学生，也该继续努力学习，Pick up from where I left off，努力成为伯克利的卷中卷。
开学的一个月一方面是努力投简历求职，另一方面是申请进Lab。我在求职方面屡屡碰壁：投了很多大型的互联网公司但都没什么回音，只收到了亚麻的一个Automatic SDE Assessment，后续也没有任何面试方面的跟进，我想这是因为从履历上看，我即没有什么优秀的小公司实习的经历，也没有学习太多高阶的算法课吧。
但是在科研(申请Lab方面)我非常幸运的收获了我校一个非常好的Lab的Offer(RAIL Lab @ BAIR)，为了这个Lab的面试我被迫读了两篇很新很新的Paper，都是这个Lab大老板的学生发的Paper，一篇是Implicit Q-Learning => 用Expectile Regression + Entropy 去modify增强学习的损失函数达到off-policy learning的效果，另一篇是把很多DL方面大模型揉进机器人里面做Self-Supervised Learning，都很有意思！
进入Lab后很荣幸可以拥有一个很Chill的美女PhD和一个很Chill的帅哥Post-doc来作为我的Supervisor（痴汉脸）！目前我们在做机器狗在增强学习环境中，如何学习在不同地形下快速的Self-Supervise走路，同时保留对之前学习过地形的记忆。
FTC中国那篇文章我有空就更新，但目前看来这个周末都要被学业占满啦~
It’s been a long time since I last updated on my life, and I’ve been busy recently, so I’ve been putting it off.
This month is the first month of school, and as a third-year student, I should continue to study hard, Pick up from where I left off, and strive to become a Berkeley “roll in the roll” - it’s a chinese slang that basically says a person’s super nerdy and studies his ass off.
In the first month of school, on one hand, I worked hard to submit my resume to apply for an internship, and on the other hand, I applied to enter labs. I have repeatedly hit a wall when it comes to job hunting: I have invested in many large Internet companies, but I have no response. I only received an Automatic SDE Assessment from Amazon, and I did not get any follow-up with regard to any interviews. I think this is because from my resume, I don’t have any excellent internship experience in small companies, and I haven’t learned too many advanced algorithm courses @ Berkeley.
But in scientific research (applying for labs), I was very fortunate to get an offer from a very good Lab in our school (RAIL Lab @ BAIR). For the lab interview, I was forced to read two very new papers, both of which are by students of the faculty(Sergey Levine) of the Lab. One is Implicit Q-Learning => Use Expectile Regression + Entropy to modify the loss function of reinforcement learning to off- The effect of policy learning, another article is Making many large DL models into robots for Self-Supervised Learning, I found both of them very interesting!
After entering the Lab, I am honored to have a very chill & beautiful PhD and a very chill & handsome Post-doc as my Supervisor! At present, we are working on how the robot dog can learn to walk in a self-supervised sertting quickly under different terrains in the reinforcement learning environment, while retaining the memory of the previously learned terrain.
I will update the article on FTC China when I have time. However, it seems that this weekend will be full of schoolwork~