Human Learning in Atari

By Pedro Tsividis, Thomas Pouncy, Jaqueline L. Xu, Josh Tenenbaum, Samuel J. Gershman. 2017

109 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

Citations in the corpus

Reconciling deep learning with symbolic artificial intelligence: representing objects and relations (2019, background): “…reinforcement learning (DRL) system that attains superhuman scores at a video game sees many millions of frames, while a human (drawing on years of other experience) requires only a few hundred frames to grasp the idea of such a game (the rest being largely a matter of honing a motor skill) [13].”
On the Binding Problem in Artificial Neural Networks (2020, background): “, 2018; Gamrian and Goldberg, 2019) and require substantially more training data than humans (Tsividis et al., 2017).”
Unmasking Clever Hans predictors and assessing what machines really learn (2019, background): “Although learning machines have become increasingly successful, they often behave very differently from humans [37, 38].”
A Survey of Reinforcement Learning Informed by Natural Language (2019, background): “Humans are able to learn quickly in new environments due to a rich set of commonsense priors about the world [Spelke and Kinzler, 2007], some of which are reﬂected in natural language [Shusterman et al. , 2011; Tsividis et al. , 2017].”
Strategic Object Oriented Reinforcement Learning (2018, background): “However, when compared with humans [28, 16], these methods require orders of magnitude more samples [28].”
Human Replay Spontaneously Reorganizes Experience (2019, background): “Although artificial intelligence (AI) is making impressive strides, humans still learn orders of magnitude faster (Tsividis et al., 2017).”
Reinforcement Learning, Fast and Slow. (2019, unknown): “data than human experts themselves [22].”
If deep learning is the answer, what is the question? (2020, background): “For example, deep reinforcement learning systems exceed human performance on Atari video games but require many times more training than a human player to acquire this proficiency 93 .”
Investigating Human Priors for Playing Video Games (2018, result): “This simple motivating experiment highlights the importance of prior knowledge that humans draw upon to quickly solve tasks given to them, as was also pointed out by several earlier studies (Diuk et al., 2008; Wasser, 2010; Lake et al., 2016; Tsividis et al., 2017).”
A Perspective on Objects and Systematic Generalization in Model-Based RL (2019, background): “For example, reading the manual of a game before playing greatly increases ﬁrst-episode performance (Tsividis et al., 2017).”
Do recent advancements in model-based deep reinforcement learning really improve data efficiency? (2019, background): “However, human players can achieve the same within minutes (Tsividis et al., 2017).”
Investigating Simple Object Representations in Model-Free Deep Reinforcement Learning (2020, background): “a computational framework to model the interactions between an agent and an environment. At each timestep t, the agent receives an observation St ∈ S from the environment and takes an action At ∈ A , where S and A are the spaces of possible states and actions respectively. The agent then receives a ”