23#36 OKR of 2023 Academic Semester and September Summary
I have established OKRs for this semester and this month, which I plan to review monthly in my progress summary. Based on my progress and the objectives of this semester, I will propose new OKRs for the following month.
OKR of 2023 Academic Semester
Deadline: @2024/02/10 09:00 (~ KDD 2024 Deadline)
Objective 1: spend most of my time on research.
Key Result 1
Have a well-prepared 1 on 1 meeting with professor at least every two weeks.
Key Result 2
Read and summarize at least 2 papers every day.
Daily Paper Reading
Key Result 3
Give a monthly summary of the research.
Key Result 4
Finish 1+0.5 to 2 research projects before deadline and submit to KDD 2024.
Key Result 5
Give an in-group sharing about my research.
Objective 2: try to discover specific future research interest.
Key Result 1
Review essential mathematic tools (linear algebra, probability, graph theory) and establish a knowledge system expressed in English.
Key Result 2
Select topics that interest me, participate in paper sharing activities and give a summary talk of selected topic.
Key Result 3
Develop a habit of checking arXiv every day.
Objective 3: speak English proficiently. (fluency > pronunciation)
Key Result 1
Get familiar with common English words and research words.
Key Result 2
Learn some English pronunciation courses and practice.
Key Result 3
Give at least one academic talk in English.
Objective 4: stay healthy.
Key Result 1
Get enough sleep: ~7 hours every night and 40 minutes at noon.
Key Result 2
Exercise every day.
Key Result 3
Take a rest every month.
OKR of September 2023
Reminder: @2023/09/30 07:00
Objective 1: spend most of my time on research.
Key Result 1
Realize all baselines.
Try some attacks methods and think about how to design attack methods on trajectory and perform subtle backdoor attack.
Key Result 2
Reread and summarize baseline papers and relevant attack papers.
Read and summarize at least 2 papers every day.
Key Result 3
Have a well-prepared 1 on 1 meeting with professor at least every two weeks.
Key Result 4
Enough time: 7 - 11:30, 13:30 - 17:30, 20 - 23:30.
Objective 2: study essential knowledge/tools/methods in spare time.
By mastering these basic tools, I can greatly improve my research efficiency.
Key Result 1
Finish CS224W notes, experiments and homework, ahead of AIAA5028.
Key Result 2
Introduction to Linear Algebra Page 22 - Page 180.
Key Result 3
Review English grammar after finishing current pronunciation courses.
OKR of 2-Year M.Phil. Study
Objective 1: be/on the way to being a professional, independent researcher
Key Result 1
An independent researcher
Key Result 2
Excellent soft skills including presentation skills, social skills and proposal skills?
Key Result 3
Follow-through
事事有回应,件件有着落。成为靠谱的人
Summary of the Past Few Months
1. The First Lesson I Learned from Prof. Liu
Here are some notes I wrote during and after the first lecture given by Prof. Liu on September 5, 2023. To be honest, I was completely shocked by the professor's values and learned a lot from this lesson. It is my honor to join the USAIL group and be supervised by Prof. Liu.
In order to remind myself, I set a reminder to review today's notes every month. Next time: @2023/10/05 21:30
Notes: The First Lesson to My Ph.D. / M.Phil. Students
2. 上个月发现的一些问题
精力分散,没有完全专注在科研上
客观上,开学前的诸多事务占用了部分时间;但主观上,积极性的确没有上上个月强,把太多时间分配到了较低效和低优先级的任务如数学、CS224W、迎新等。时间是很有限的,要可持续地形成良性循环,我只能保证 11.5 - 13 个小时的学习科研时间,如果不能尽量提高专注程度和效率,把太多时间分配到杂事上,那么很难保证科研进度。
解决方案:尽可能让所有时间都为科研服务
- 提高专注程度,减少分配在低优先级任务的时间
- 学会跳岛,避免系统学习的倾向。除了必要的基础知识和工具,其他的东西在应用中学习和掌握
- 调整学习方法,自学时以 notes & hw 为主,视频为辅
- 学会复用,rbm 的项目必须与科研方向对齐,没有多余的精力和时间花在项目上
时间管理存在误区
虽然很早我就开始践行 GTD 时间管理,但一直存在任务拆解不够明确、任务量不够细化、没有按优先级管理等问题。在听“开学第一课”时,我意识到应该按照任务的优先级来调整现有的时间管理方法。
解决方案:尝试新的时间管理和日常反馈方案
- 所有任务按照 3 级优先级规划,减少不必要的日常任务提醒
- 在原先每周总结、每月总结的基础上,加入对当月 OKR 完成情况的回顾
- 基于回顾中发现的问题,制定下个月的 OKR
目前闭门造车严重
闭门造车的原因有两点:一是过于追求完美心态,怕犯错。拿实现 baseline 注入攻击来说,总是担心设计的方案不够好,没法兼容后面的方法和新 baseline,导致经常修改设计思路,以至于拖慢了既定进度。二是对目前做的工作了解不够深入,没法用很简单的几句讲清楚。
解决方案:
- 以结果为导向,优先实现功能,不要怕犯错,后面再迭代
- 每天吃饭/散步时找同学尽可能简洁地介绍自己的方法和进展
- 回顾相关的 attack 工作,及时总结他们的思想和攻击领域的论文行文结构
忽视效率工具的使用
目前对以 ChatGPT 为代表的一些效率工具使用还不够熟练,上周写完脚本后尝试让 GPT 帮忙改进,发现效果很好。虽然不能依赖,但要掌握效率工具的使用。
解决方案:
- 多多尝试 GPT、Claude 等生成工具,用他们去改进,不要直接生成,同时要仔细检查
- 目前申请了 Copilot 资格,下周研究在 VSCode 中如何使用