Reinforcement learning with hidden states

By Longxin Lin, Tom Michael Mitchell. 1993

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

Memory-based control with recurrent neural networks (2015, background): “they perform control from high-dimensional pixel inputs) and demonstrated that recurrent Q learning [15] can perform the required information integration to resolve short-term partial observability (e.”
Structure-Adaptable Digital Neural Networks (1999, background): “Moreover, mobile robots have noisy and non-linear sensors and su er from the problem of hidden-state [116] or perceptual aliasing [227] which renders the problem of navigation a non-Markovian task.”
Reinforcement learning by backpropagation through an LSTM model/critic (2007, methodology): “Using the next observation and reward as additional, supervised training information will in many cases help to infer a good approximation to the state (see [9]).”
The Artificial Life Roots of Artificial Intelligence (1993, background): “Proposed solutions all go in the direction of new complexity (in the form of models of return, or in more recent cases world models predicting return [61,118]).”
Cooperative behavior acquisition in multi-mobile robots environment by reinforcement learning based on state vector estimation (1998, background): “Lin [4] compared window-Q based on both the current sensation and the N most recent sensations and actions with recurrent-Q based on a recurrent network, and he showed the latter is superior to the former because a recurrent network can cope with historical features appropriately.”
Learning Without State-Estimation in Partially Observable Markovian Decision Processes (1994, background): “There is some empirical evidence that such atechnique can work well on particular non-Markovianproblems (e.g., Barto et al., 1983).”
A robot that reinforcement-learns to identify and memorize important previous observations (2003, methodology): “the stored sequence of observation-action pairs are used by the RL algorithms as if they happen now: “experience replay” [6], [7].”
Reinforcement Learning with Long Short-Term Memory (2001, methodology): “One way is to let the RNN learn a model of the environment, which learns to predict observations and rewards, and in this way learns to infer the environmental state at each point [6, 11].”