AI journey @ JHU

Artificial Intelligence, Master of Science @ JHU Whiting School of Engineering. Start from 2021 summer

Want to know more about JHU EP? Here is a shared collective document on JHU EP 101

Why this repo

• Documenting my master's journey, including information collection, planning, and reflection.
• Offering a reference, checkpoint, or simply a data point to the community.
• Include general and publicly available information, and my personal notes.

Before you pick a program

• Think of your needs, effort you want to spend, and available options.

  • Master's or PhD
  • Engineering or MBA
  • Full-time or part-time
  • A degree, a certificate, or individual (series) courses.

• Alternative choice

  • Data Science/Machine Learning/AI Entry: Begin your journey with Coursera, Kaggle.
  • Tools: pick up the packages such as Scikit-learn, Xgboost, or search a repo in Github.
  • SOTA Reproduction: Github, conferences (such as ACL, ICML, NIPS, AAAI, IJCAI, ICCV, CVPR, KDD...), or Arxiv
  • Single or a series of courses: Check Resources
  • Certificate Programs: Think of taking a certificate, such as in Standford(SCPD) or Berkely. You got chance to study with the full time students on the SOTA.
  • Other programs: Gatech, Berkeley, University of Michigan, UCLA, UCSD ...

• Which area interests you? DS: insights. ML: forecasts. AI: actions. Reference: What's the difference between data science, machine learning, and artificial intelligence?

• Misc: US degree, finance, time, risk and opportunity cost.

• Read this one: Why AI is Harder Than We Think

• Oct 2025: I found this interesting article. I wish I read it earlier: Doing well in your courses

Possible Goals

• AGI(Artificial General Intelligence) is not likely to happen in my generation. But if something interesting comes up, I wish I can understand it.

  • 2024 Update: I am still skepticial regarding AGI realization in the next two decades, the recent development in the LLMs (and LMMs) are really interesting. Let's see.

• Personal interests

  • Math - Awareness. Lay the foundation. Most (95%) of what you learned in school won't be used in career. What we learn in school is not knowledge, but the exercise of how to learn, and the ability to think. For example, we should be able to pick up new math topics at a reasonable timeframe (3 months for a typical graduate level course)
  • Coding - Re-visit data structure and algorithms, leetcode, familiar with various packages. => Make it work, resolve the problem.
  • Academic Training - Discover and pick up the knowledge, be aware of where good taste and standards are, and restructure my knowledge system to support my future improvement.
  • Think about long term. Write, Create and share

• General goals - collected from Internet and friends, keep growing. Please share your thoughts

  • Program - Finish it
  • Program - Finish with the shortest possible timeframe
  • Program - Finish with quality, such as straight A or honor
  • Academic - Join a research project
  • Academic - Publish a paper, and more.
  • Coding - General coding skills like Leetcode
  • Coding - Familar with various languages, such as Python, C++, Julia, R ...
  • Coding - Familar with various packages and framework, such as Scikit-learn, Tanserflow, Pytorch ...
  • Engineering - Scale up, such as cloud; HPC
  • Career - Extend your network for new career opportunity
  • Career - Start a new career, join a startup
  • Certificate - AWS
  • Community - Influence the people around you, make positive impact.
  • Community - Record my journey which might help others.
  • Softskills -
  • Efficiency - Collect tips and tools to improve the efficiency.

Lesson Learn

• Start early, and secure your time.
• If you want to get a better grade, ALWAYS read carefully about the instruction, and stick with the rubic. Very few professor wants to discuss in detail, and most of them do not care how you think and explore. Simply grade by the rule.
• A course with a few assignment and lack of instructer attention, might result in surprise of the final grade.
• Pick the course interests you, and enjoy it.
• If a teacher do not have time to reply your email, or even host a office hours, you won't be graded in time. And you are likely to receive a surprise on your final grade.
• About ChatGPT in education system:
  • The education industry haven't changed over thousands of years. Based on my observations, the key to receive a high grades highly rely on your understanding of the teacher’s grading criteria and show what the teacher wants to see. Show it clearly, check, you get the score. Teachers often do not care if you truly understand the material or about your insights. Excellent teachers are willing to discuss these during office hours, but others may not have the time to talk to you.
  • In this context, tools like ChatGPT, which very likely to change the entire education system, have not been well integrated into our education system yet. The society focus on the result, not whether you completed your work using paper, a typewriter, or a computer. GPT and Google are just tools. We need to learn how to use them, not simply ban them from the school. If schools cannot build a good system to assess your understanding of knowledge and mastery of tools, it is the school’s problem. You should have a clear goal, understand your abilities, know what the tools can and cannot do, and do NOT let the school stop you learning them and unleash your potential.
  • On the school side: If your school ban it, of course you need to stick with the rule for the program. But please remember that it is NOT your ultimate goal.
  • On the GPT side: It knows all areas with 80% in depth, and it could be wrong. On the other side, human typically good at 1 or 2 areas, with 90% or some top experts might reach 95% (assume you won't be able to talk to top frontier researcher). So you need to figure out how two identify what is correct, what is wrong, what is valuable, and what you really need to test and verify. This is an important skills you need to figure out. The best way is through practice and review.

Misc @ JHU AI program

• Theoretical or applied approach
• It is common that US can have multiply Master degree. So it leads to 2 options
  • Life long study, with many master degrees
  • Pursue a PhD
• JHU AI program is relevantly new, more courses will be added to the elective courses list, and more paths will define and provide. Pay attention to the new courses.
• Course code. For example: EN.625.250
  • EN: school (Engineering, Medical, etc. )
  • 625: Major (Applied and computational Math 625, CS601 or 605, DS 685 etc. )
  • Next three digit: Course number. The first digit:
    • 2: undergraduate course, not to degree.
    • 6: graduate level course.
    • 7: Advanced graduate level

Journey of applying the schools

It is not easy, so start early. Here you can find my journey, and I hope you will make yours easlier.

• 2018: Plan for my DS master application. It was interrupted by an international assignment.
• 2020: Back to the bay area, resume my application
• School selection: CSRankings - AI. Most of the CS departments only accept full-time.
• TOEFL:
  • Sept 2020: 29/29/23/25. UCLA CS 21/17/24/25, and UW CS all 25+. Focus on speaking
  • Oct 2020: xx/xx/25/21. Focus on writing.
  • Nov 2020: xx/xx/23/27. Maybe I do not belong to UCLA and UW. Move on.
• Application
  • It took me 4-5 months to get all documents verified. WES(world education services) issues me a wrong course-to-course evaluation report in the first place and then leave an obvious issue in the second correction report. After that, they insist something like 2 < 1 and refuse to take any further amendment. Probably, that is life.
  • 3 applications, 3 offers. JHU seems one of the most rigorous programs I can find.
• JHU
  • Provisional admission in DS program @ Jan 2021. Spent Q1 and Q2 on my organization transformation.
  • Interesting to see the breakthrough in mathematic, such as Deep Networks from First Principles, and Geometric Deep Learning, Grids, Groups, Graphs, Geodesics, and Gauges. So I switch to the AI program @ May 2021

Before the first semester

• From the strategic perspective, don’t overestimate. Tactical perspective, don't underestimate.
• Q1 2021: Stabilize the current working environment and secure resource
  • Full-time work: Starting Q3, 50% of my time will focus on ML/AI projects.
  • Personal Time: Secure 19:00-22:00 PM for my study. One weekend (well...we will see)
  • Financial preparation. Company support etc.
• March 2021: Python 3 Programming Specialization. Brush up my python skill.
• March 2021: Precalculus through Data and Modelling Specialization. Familiar with the math terminology in English, and refresh my high school math.
• April 2021: Calculus I with Precalculus: A One-Year Course. I happen to have this book. The US textbook organized logically with a good learning curve. Easy to follow, love it.
• May 2021: Khan: Multivariable calculus. Finished Derivatives of multivariable functions, not easy to get 100%
• May 2021: Advanced Engineering Mathematics, 10th Edition, or you can find it on Amazon, Pre-read a few chapters.

Pre-requisites courses

No matter how good you did 20 years ago, they won't leave too much trace on you. I am not so mad at WES now and plan to take as many pre-courses as possible to refresh my knowledge and prepare for coding:

• 625.250 - Multivariable Calculus and Complex Analysis
• 625.252 - Linear Algebra and Its Applications
• 625.240 - Introduction to Probability and Statistics
• 605.202 - Data Structures

Courses List

• Artificial Intelligence Courses
• Data Science Courses

About coding

Languages

• Python: Of course.
• Java: Widely used in industry and engineering field. Relevantly easy, tons of framework.
• C++: Better performance, closer to hardware, great for CV.

Misc

Source for textbooks

• Amazon

• Abebooks

Resources:

• MIT Open Courseware
• MOOC - Coursera
• Khan (inspired by my classmate).
• Math - Areas of mathematics
• Math - Graduate Texts in Mathematics
• Course - Standford, CS229: Machine Learning
• Course - Standford, CS231n: Convolutional Neural Networks for Visual Recognition
• Course - Berkeley, EECS 208: Computational Principles for High-dimensional Data Analysis,
• Course - Berkeley, CS 285: Deep Reinforcement Learning,
• Course - NYU, DS-GA 1008: Deep Learning
• Papers - Classic papers for beginner

Articles

• Pathways

Tools:

• GitHub Desktop - Maintain this repo
  • Markdown + Latex + Export PDF:
• iPad: iPad 2020 + pencil (or iPad Pro 11 + pencil 2. iPad Pro 12.9 too heavy, 27 inch monitor is better). Good for reading PDF textbooks. It is astonishing to know that the textbook available publicly in PDF(Thanks Author!) but listed $900+ on Amazon as new, $600+ (acceptable) or $200 (good) as second hand. If license is not a issue (assume), then obviously there is something wrong in the US academic publication system.
• Goodnotes 5 - A great tool for taking notes and mark on the PDF textbook
• MindNode: A good mindmap tool. Alternative option: Xmind, SimpleMind, GitMind...
• Screenshot: Snipaste
• Markdown: Typora by far is the best. Alternative option include
  • Mark Text (with bugs)
  • Atom + markdown-preview-enhanced plugin, Heavy. For markdown preview and Latex(KaTex)

Reference:

• Markdown reference - Github Markdown
• KaTeX Reference
• Shorten the URL in GitHub - git.io