Contents

My overarching goal is to follow Paul Graham’s advice to turn myself into ‘the sort of person who can have organic startup ideas’*.

Courses

I don’t know how many of these it is feasible to finish in about 12 weeks but I wish to do as much as possible.

Key:   ⬜ To Do   ✴️ In Progress   ✅ Done   🟨 Target is to finish by end of RC by challenge deadline   🟢 Programming Parts Done   🔵 A self-contained section complete; [ ] Tasks for challenge

Machine learning

(HW1: Autoregressive models considered complete due to creation of this tutorial in addition to the successful use of autoregressive models in other homeworks)

CS

  • MIT Computation Structures AND / OR Nand2tetris

  • Nand2tetris (13 x HW) ✅ ✅ 🟨 🟨 🟨 ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜

  • CUDA programming (Caltech CS179, Oxford)

    UPDATE: going to focus on CS179 as the assignments are more challenging (no criticism of the Oxford course which has good lectures that I have found useful but it has been designed as a crash course rather than a computer science course) Caltech CS179 (6 x labs + 1 x project proposal + 1 x project) ✅ ✅ ✅ ✅ 🟨 🟨 ⬜ ⬜

    Oxford (12 x practicals) ✅ ✅ ✴️ 🟨 🟨 🟨 ⬜ ⬜ ⬜ ⬜ ⬜

  • Caltech CS171: Introduction to Computer Graphics (HW0 + HW1-6 + HW7.1 + HW7.2)

    ✅ ✅ ✅ 🟨 🟨 ⬜ ⬜ ⬜ ⬜

  • MIT Distributed Systems (6.824) (Lab 1 + Lab2A-C + Lab3A-B + Lab4A-B + project proposal + final project)

    ✅ ✅ ✅ ✅ 🟨 🟨 ⬜ ⬜ ⬜ ⬜

  • Cryptography (Stanford course, maybe also Cryptopals)

Stanford course (6 x units + 1 exam) ✅ ✅ ✅ ✅ ✅ 🟨 🟨

Miscellaneous study goals

  • Stanford Convex Optimization (Stanford’s course on the edX platform; 11 units but only 9 assignments so I am doing an equivalent number of exercises for the remaining two)

    ✅ ✅ ✅ ✅ ✅ ✅ ✅ 🟨 🟨 ⬜ ⬜

  • MIT Discrete Mathematics (12 x problem sets but might skip the last 4 as they are about probability which I have covered already)

    ✅ ✅ ✅ ✅ ✅ ⬜ ⬜ ⬜ ❔ ❔ ❔ ❔

  • Geometric Folding Algorithms (5 x HW + 1 x project milestone + 1 x final project; course doesn’t specify milestone but I’m including it analogously to the other ML courses here)

    ✅ 🟨 🟨 🟨 ⬜ ⬜ ⬜

  • MIT Advanced Fluid Mechanics(12 x units) ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
    ✅ ✅ ✅ ✅

  • edX Mastering Quantum Mechanics (14 assignments)

    ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅

Other areas to cover through some means

  • Linear Algebra
    • Review basics more deeply (when I originally studied it I didn’t know much coding and I think I will perceive it differently and more intuitively now that I have a lot of experience manipulating multi-dimensional arrays)
    • ✴️ Eigenvectors and eigenvalues
    • ✴️ Singular Value Decomposition
  • ✴️ Wavelets
  • Causality (list of books discussed here)
  • Quantum Computing and Machine Learning (UTQML101x, this book)
  • Optimal Transport (potential sources to use: A Brief Introduction to Optimal Transport Theory, Computational Optimal Transport)

  • Expectation Maximisation

  • Chemistry AND / OR Biology topic each week (x 12)

    ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜

Other courses that I mean to finish by the end of 2021

Object Detection

  • Read all the references and make a “depth-first” tree for:
    • MaskRCNN
    • DETR
  • Implement
    • MaskRCNN
    • RetinaNet
    • YOLO
    • SSD
    • CenterNet
    • DETR
    • VoxelNet (overlaps with Lyft challenge model (see below)
    • BotNet
    • Others
  • Come up with my own model

Tools and practices

  • Deconstruct an existing version and write my own of the following:
    • Makefile for CUDA
    • Makefile for C++
    • Dockerfile
    • Pip package
    • Bash script
    • requirements.txt

Miscellaneous topics, tasks and tutorials

✴️ Hopfield Networks

Other goals

  • Think of ~10 ideas that are beyond my present level of skills or knowledge i.e. where I have no idea what to do - and then try to come up with a plan to realise them 0️⃣ 3️⃣

✅ ✅ ✅ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜

  • Write >= 12 technical blogposts whose main purpose is to practise articulating technical ideas rather than to be amazing 0️⃣ 3️⃣
  • Read >=101 machine learning papers (I get credit only when I have produced some output related to the paper) 0️⃣ 4️⃣ 8️⃣

✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ 🟨 🟨 🟨 🟨 🟨 🟨 🟨 🟨 🟨 ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜

✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜ ⬜

  • SNAIL
  • MAML for sinusoid data
  • Simple 2D data VAE and DC-GAN
  • Transformer for machine translation (this is a placeholder for now as I can’t remember which other model I implemented and I only want to give myself credit only when it is due)
  • MoCov1v2 (v1 with some v2 features by mistake)
  • MoCov1v2-TPU and MoCov1
  • Various simple Flow Models (counted as 1)

  • Various simple image VAE Models (counted as 1)

  • Implement and train the following Kaggle prize-winning models

    ✴️ Lyft 3D Object Detection for Autonomous Vehicles

    ⬜ OSIC Pulmonary Fibrosis Progression



* How to Get Startup Ideas