Katas for Machine Learning.
Pick a topic, create an exercise using the following suggested template (subject to change):
- what will we get to learn at the end of this exercise?
- how is your environment going to look like
- datasets you are working with
tools, languages etcsince people might want to try different things here, this information can go in the solution readmes
- reference documents, papers, baselines etc.
Work on the solution and keep pushing in this repo. The directory structure can look like this:
README.md # this readme
<some-problem>
README.md # problem specification
<solution-dir-1> # someone's solution
<solution-dir-2> # someone else's solution
Here is an ever increasing dump of randomly selected topics for inspiration.
- optimization
- convex, non convex
- gradient based/free
- population based and other random
- search
- game theory
- constraint satisfaction, reasoning and planning
- propositional and first order logic
- programming
- GPU programming, cuda and opengl
- parallel computing, mpi etc.
- working with memory effectively in large dataset situation
- model quantization and other runtime optimizations
- distributed learning
- computer algebra systems (symbolics)
- auto diff
- tidy data and data wrangling
- metrics
- loss functions
- surrogate loss and calibration
- graphical models
- CRF, HMMs etc.
- variational inference
- learning theory
- complexity and runtime of learning algorithms
- generalization
- regularization
- clustering
- online learning
- linear algebra
- recommendation systems
- neural nets
- convolutional
- recurrencies
- energy based models
- misc architectures
- non differentiables
- architecture search
- MCMC and other sampling based inference techniques
- randomized algorithms
- generalized linear models
- kernel methods
- hypothesis testing
- visualization and interpretation
- semi supervised learning
- active learning
- ensembles, boosting etc.
- knowledge representation
- relational
- gazetteers
- GOFAI, expert systems etc.
- parsing
- reinforcement
- MDPs, POMDPs