Self-directed study

A self-directed curriculum connecting mathematical foundations, machine learning and quantitative research. All in pursuit of the BIG BUCKS, (and the evolution and perseverance of humanity of course)

StatusIn progress
Category
Learning / personal tools
Tags
  • Mathematics
  • Machine learning
  • Quantitative research
  • Computer Science

Overview

I pick what to learn from what I want to build. Math, machine learning, quant research, and project work to check I actually learned it.

No institution, no credential. It's a map I keep, and it changes whenever I find out something was in the wrong place.

Not everything is worth mastering

I used to treat every subject as something to build from scratch. That doesn't survive a finite number of hours, and it trains the wrong thing. You end up a slower version of the specialist you'd want to hire.

So everything gets a target depth. A few things I need to build, because at the start there's nobody else to build them. Most I need to spec, judge and plug in. Some I only need to recognise well enough to pick the right tool or ask a decent question.

Target depth

  • Mastery / build

    What it means
    I can build it from scratch and explain how.
    What would count
    A working build and a write-up.

  • Direct / evaluate

    What it means
    I can write the spec, judge the build, plug it in, and hold my own in the conversation.
    What would count
    A small build for feel, plus a spec or review.

  • Literacy

    What it means
    I know the words, when it applies, and how to tell good from bad.
    What would count
    A one-page map. What it is, when to use it, which tools are standard.

Where I'm aiming. Not where I am. No progress figures on this page.

Learning map

DiagramIllustrative
Shared foundationsDirectionsWhere it gets testedMathematicalfoundationslinear algebra,probability,information theory,calculusScientificcomputingPython, numericalmethodsSimulation andstochasticsMonte Carlo,dynamical systemsQuantitativeresearchmarket structure,signals, backtestintegrityMemory andcontinual learningassociative memory,replay, forgettingDeep learningfoundationsnetworks, attention,uncertaintyOpen Traderthe quantitative sideMnesisthe memory sideResearch and teamdirectionwork I could not doalone
  • Processing
  • Interface or output

Where I'm pointed, not what I've finished. The arrows show what feeds what.

Subjects

Each one carries a target depth. M for build, D for direct, L for literacy. A target is an intention, not a trophy.

  • Foundational mathematics and scientific computing

    The floor under everything else. Python is the hands-on language.

    • Linear algebraTarget depth M
    • Probability, statistics and information theoryTarget depth M
    • Calculus and automatic differentiationTarget depth M
    • Scientific Python: NumPy, pandas, plottingTarget depth M
    • Numerical methodsTarget depth D
    • Monte Carlo and stochastic simulationTarget depth D
    • Dynamical systems and high-dimensional geometryTarget depth D

  • Quantitative research and market structure

    Tested against Open Trader, not studied on its own.

    • Market microstructureTarget depth M
    • Order book mechanicsTarget depth M
    • Strategy and alpha research processTarget depth M
    • Statistical arbitrage and signalsTarget depth M
    • Backtesting and simulation integrityTarget depth M
    • Portfolio risk, hedging and correlationTarget depth M
    • Market-making modelsTarget depth D
    • Stochastic processes for financeTarget depth D
    • Options theory and stochastic calculusTarget depth D
    • High-frequency data engineeringTarget depth D
    • Low-latency execution in C++Target depth L
    • Trading infrastructure and systems architectureTarget depth L

  • Memory and continual learning

    The reason Mnesis exists.

    • Associative memory and Hopfield networksTarget depth M
    • Sparse distributed memory and vector-symbolic architecturesTarget depth M
    • Attractor networks and neural codingTarget depth M
    • Complementary learning systemsTarget depth M
    • Continual learning theory: forgetting, EWC, replayTarget depth M
    • Memory and replay systemsTarget depth M

  • Deep learning foundations

    Deep enough to judge and plug in. Adopt the frameworks, don't rebuild them.

    • Memory-augmented transformers and differentiable memoryTarget depth M
    • Neural network fundamentalsTarget depth D
    • Attention mechanisms and transformersTarget depth D
    • Optimisation and generalisationTarget depth D
    • Generative modelsTarget depth D
    • Surprise and uncertainty estimationTarget depth D
    • Synaptic plasticity and learning rulesTarget depth D
    • Systems neuroscience of memoryTarget depth D
    • Computational neuron modellingTarget depth L

  • Research and team direction

    How to run work bigger than one person.

    • Capital allocation and risk budgetingTarget depth M
    • Research-program and team orchestrationTarget depth D
    • ML research process and evaluationTarget depth D
    • Compute, infrastructure and MLOps directionTarget depth L
    • Regulatory and compliance literacyTarget depth L
    • Building and directing a research teamTarget depth L

How I keep myself honest

Reading gives you a convincing feeling of understanding that dies the second you try to build something. So the curriculum comes with projects. Open Trader tests the quant half, Mnesis tests the memory half. When a project is hard for a reason I didn't see coming, the map was wrong.

The curriculum lives in a private app of mine and I don't publish progress out of it. Turning an intention into a percentage would make this page claim more than I know.

What's missing

  • Self-taught. No institution, no assessment, no paper at the end.
  • Depth is uneven. Some of this I've worked in. Some I've only mapped.
  • The hard part to arrange alone is correction. I can find the reading. I can't easily find out where I'm shallow.

Next

Not more subjects. People who'll tell me what I've got wrong. That's why I'm asking to audit courses, and why criticism is the thing I actually want out of a conversation.

Think you could pull this apart? I want to hear it.

Email me about Self-directed studyBack to learning