• Fun and interesting ideas to work on as projects?

• NLP

  • Convert novels / fanfic into visual novels? Automatically convert your favourite stories into a visual experience.
    • Automated attribution of quoted speech? https://ojs.aaai.org/index.php/AAAI/article/view/7720
      • Google also: automated quotation attribution
    • i.e. convert into renpy
    • Optional:
      • Find/Train a model that can create art, or obtain pieces of art that resemble the character, facial expression, background?

• Coursework

  • Distributed Systems
  • Datacamp: I've yet to finish that course LMAO
  • Kaggle: Every employer wants to see this
  • Machine Learning course (See markdown/machine_learning section)

• Computer Vision

  • Convert a person (in video) to animated anime character?
    • Automate extraction of person from video frame
    • Automate redrawing of said person
    • Prepare for ebsynth for optical flow
      • https://github.com/jamriska/ebsynth
    • What about... the opposite? Make a person perform something else:,,,,,,,,,,121 deep fake?

• Machine Learning

  • Find/Train a model that can identify speech bubbles (for manga translation)
    • Create a webapp that can perform such modifications on the fly?
    • Create a webapp that allows you to paste raws in, and show you the raws, romanized and english version?

• Reinforcement Learning

  • Implement Rainbow QL, or check github

• Software Engineering

  • Try out that react redux boilerplate you've been looking at

"Module 1"

• [https://aws.amazon.com/about-aws/global-infrastructure/]

• Edge location definitions

• Provides AWS services (compute and resources)

• Split across regions

• Access by Management consoles, CLI and SDKs

• Lmao in the end I didn't take notes

• I recently joined a Discord server on RL because the best way to learn is to interact with like-minded people.

Good learning resources:​

• https://github.com/andyljones/reinforcement-learning-discord-wiki/wiki

Computational resources?​

• https://www.tensorflow.org/tfrc
  • https://discord.com/channels/765294874832273419/765294874832273422/768102671881863178
  • TSFRC

Big Data​

• https://www.quora.com/How-do-you-learn-big-data

NN​

• https://towardsdatascience.com/the-mostly-complete-chart-of-neural-networks-explained-3fb6f2367464
• https://towardsdatascience.com/a-beginners-guide-to-convolutional-neural-networks-cnns-14649dbddce8
• https://machinelearningmastery.com/how-to-configure-the-number-of-layers-and-nodes-in-a-neural-network/
• https://arxiv.org/pdf/1904.07633.pdf
• http://marekromanowicz.com/files/master_thesis.pdf

Fascinating stuff​

• Meta-RL
  • MAML? RL^2? PEARL?
• NAS (Medium)
  • Paper
• OpenAI Gym
• Voice remover
• wav2mid
• NTK
  • https://github.com/damaru2/ntk
• Alpha0 for connect4
• Testing MCTS
• AI safety
  • See also

Glossary

• FLOPS
  • Floating point operations per second

Cute graphs​

• https://timqian.com/chart.xkcd/#introduction

Setting stuff up​

• I want to try using boilerplates to immediately set stuff up quickly.

You and your research​

• http://joschu.net/blog/opinionated-guide-ml-research.html
• http://www.cs.virginia.edu/~robins/YouAndYourResearch.html

I've been trying to figure out what is the best way to consolidate information during my time in NUS. It would be such a waste if I forget what I learn in university, say a few years down the road, and having a good way to consolidate information would also minimize time spent on re-understanding material during the semester.

To achieve that, I have tried using Google Docs, mindmaps using diagrams.net, Anki and Latex cheatsheets as mediums of consolidating information. Some of these methods are better than others in some way or another.

Personally, I think something like mkdocs or in this case, Docusaurus, is kind of similar to taking notes with Google Docs, but the key difference is that it is much more neatly organized and accessible.

• I kind of regret taking notes in Anki, although to be frank it was pretty good at helping me remember concepts.
  • The problem with Anki is that its selling point is also its weakness; I'm too lazy to spend an hour everyday memorizing stuff.
• Diagrams.net is actually pretty good since humans tend to remember images better than documents. The problem is that it is slow in terms of a quick look-up in the future because you no longer remember where everything is positioned.

When my friend Evan introduced me to Docusaurus, I thought I might as well try it out for a few reasons:

1. It's easy to setup and use
2. It supports math equations
3. I like the layout
4. It's got a search feature
5. It can help someone who is also struggling to understand a concept I once struggled to understand

Obviously this means that a substantial amount of content is not in these notes, but if they are important in some way... they will most likely make their way in here eventually