The Next Input updates
Browse every published The Next Input update in a calm card overview with images, dates, and direct access to each article.
The Next Input update
Nonlinear computation in deep linear networks
The Next Input update
Learning to model other minds
Read paper(opens in a new window)(opens in a new window)
The Next Input update
Learning with opponent-learning awareness
Read paper(opens in a new window)
The Next Input update
OpenAI Baselines: ACKTR & A2C
Read code(opens in a new window)Read paper(opens in a new window)
The Next Input update
More on Dota 2
Our Dota 2 result shows that self-play can catapult the performance of machine learning systems from far below human level to superhuman, given sufficient compute. In the span of a month, our system went from barely matching a high-ranked player to beating the top pros and has continued to improve since then. Supervised deep learning systems can only be as good as their training datasets, but in self-play systems, the available data improves automatically as the agent gets better.
The Next Input update
Dota 2
Rewatch live event
The Next Input update
Gathering human feedback
View code(opens in a new window)
The Next Input update
Better exploration with parameter noise
Read code(opens in a new window)Read paper(opens in a new window)
The Next Input update
Proximal Policy Optimization
View code(opens in a new window)Read paper(opens in a new window)
The Next Input update
Robust adversarial inputs
We’ve created images that reliably fool neural network classifiers when viewed from varied scales and perspectives. This challenges a claim from last week that self-driving cars would be hard to trick maliciously since they capture images from multiple scales, angles, perspectives, and the like.
The Next Input update
Hindsight Experience Replay
The Next Input update
Teacher–student curriculum learning
Read paper(opens in a new window)
Showing 1021 to 1032 of 1,127 updates.