Human-Level Control Through Deep Reinforcement Learning.¶
Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., et al. (2015). Human-Level Control Through Deep Reinforcement Learning. Nature, 518(7540), 529-533.
Cited by¶
5 citations across 5 artifacts.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Markov Decision Processes (MDPs)
- (2016) on AlphaGo, AlphaZero (2017), MuZero (2019), and many others — demonstrated RL at scales that were inconceivable a decade earlier.
This sourceIntroduced Deep Q-Networks (DQN) achieving human-level performance on Atari games via deep-neural-network function approximation in the MDP framework; landmark deep-RL result.
- (2016) on AlphaGo, AlphaZero (2017), MuZero (2019), and many others — demonstrated RL at scales that were inconceivable a decade earlier.
- Memory Consolidation
- In machine learning, experience replay re-presents past transitions to stabilize what online updates would otherwise catastrophically forget, and continual-learning methods explicitly engineer a consolidation stage that protects past learning while incorporating new.
This sourceIntroduces experience replay—storing and re-presenting transitions to stabilize learning against catastrophic interference, a direct port of hippocampal replay.
- In machine learning, experience replay re-presents past transitions to stabilize what online updates would otherwise catastrophically forget, and continual-learning methods explicitly engineer a consolidation stage that protects past learning while incorporating new.
- Replay
- In reinforcement learning, experience-replay buffers store transitions during interaction and resample them off-policy during training, the trick that made deep RL stable.
This sourceIntroduces deep Q-learning with an experience-replay buffer that decorrelates transitions and stabilizes training.
- In reinforcement learning, experience-replay buffers store transitions during interaction and resample them off-policy during training, the trick that made deep RL stable.
Domain-specific¶
- Bellman Equation
- DeepMind's Deep Q-Network (Mnih et al., Nature, 2015) applied the same equation to learn Atari video games directly from screen pixels
This sourceNature paper in which the same algorithm, network architecture and hyperparameters, applied game by game, learned to play 49 classic Atari games from pixels at a level comparable to a professional human games tester.
Supported in partVerified against the work's full text
“that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning”
- DeepMind's Deep Q-Network (Mnih et al., Nature, 2015) applied the same equation to learn Atari video games directly from screen pixels
Mechanisms¶
- Experience Replay Buffer
- Its strength is turning a perishable, correlated stream into a reusable, decorrelated dataset: it is the reason a single expensive interaction can drive many updates, and the reason off-policy learning is stable enough to work at all.
This sourceShows that replay memory reuses stored transitions while random sampling reduces correlations in training data.
- Its strength is turning a perishable, correlated stream into a reusable, decorrelated dataset: it is the reason a single expensive interaction can drive many updates, and the reason off-policy learning is stable enough to work at all.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Read against the text for 1 of 5 citations: 1 supported in part. Each verdict is shown under its citation below, with what in the work backs the sentence.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Registry ID ref:68e0f893cb3e · see in the full table