Machine Learning

Models that learn from data, and the architectures, training methods and theory modern AI is built on.

Sub-fields

ArchitecturesTransformers, state-space models, mixtures of experts and the building blocks networks are assembled from.4 entries · 4 open problems
Training & OptimizationOptimizers, initialisation, normalisation and why gradient descent on huge networks works as well as it does.3 entries · 4 open problems
Scaling LawsHow loss and capability change with parameters, data and compute, and how labs plan training runs from them.2 entries · 2 open problems
Reinforcement LearningAgents that learn from reward, from game-playing systems to the fine-tuning of language models.7 entries · 11 open problems
InterpretabilityReverse-engineering what trained networks compute, from single features to whole circuits.2 entries · 3 open problems

Open problems, most leverage first

All open problems →

#VotesProblemWork mixDVLStatus

010Does RLVR create new reasoning ability or only sharpen the base model? A pass@k protocol that separates the twoReproductionRL with verifiable rewards and GRPO for reasoning models1 link3 bottlenecks
506052Open

Research

Filter in Research →

ArchitecturesAlgorithmActive

The transformer architecture

The attention-based neural network design behind modern language models, and the quadratic cost of attention that drives work on exact, approximate and recurrent alternatives.

2 problems15 dependents0 maintainers

Machine LearningLibraryMature

PyTorch

The dominant open-source deep learning framework: tensors, autograd and a compiler stack.

0 problems4 dependents0 maintainers

Machine Learning

Sub-fields

Open problems, most leverage first

Research

The transformer architecture

PyTorch

Machine Learning

Sub-fields

Open problems, most leverage first

Research

The transformer architecture

PyTorch

Reverse-mode automatic differentiation

RL benchmarks and evaluation: Atari 100k, ProcGen, rliable

Offline reinforcement learning

Model-based RL and world models (MuZero, Dreamer)

Monte Carlo tree search and AlphaZero

RL with verifiable rewards and GRPO for reasoning models

RLHF and preference optimization (DPO)

Policy gradient methods and PPO

Maximal update parametrization (μP) and hyperparameter transfer

Adam, AdamW and the optimizer zoo

Language-model evaluation methodology

Neural scaling laws and compute-optimal training

State-space sequence models (S4, Mamba, Mamba-2)

Mixture-of-experts routing

Normalization layers (BatchNorm, LayerNorm, RMSNorm)

Mechanistic circuits and causal interventions

Sparse autoencoders for interpretability