View source for 18066270. ENABLING CLASSICAL PLANNING IN INTERACTIVE ENVIRONMENTS WITH INCOMPLETE MODELS simplified abstract (International Business Machines Corporation)

=ENABLING CLASSICAL PLANNING IN INTERACTIVE ENVIRONMENTS WITH INCOMPLETE MODELS=


==Organization Name==

[[:Category:International Business Machines Corporation|International Business Machines Corporation]]


[[Category:International Business Machines Corporation]]

==Inventor(s)==

[[:Category:Don Joven Ravoy Agravante of Tokyo (JP)|Don Joven Ravoy Agravante of Tokyo (JP)]][[Category:Don Joven Ravoy Agravante of Tokyo (JP)]]

[[:Category:Michiaki Tatsubori of Tokyo (JP)|Michiaki Tatsubori of Tokyo (JP)]][[Category:Michiaki Tatsubori of Tokyo (JP)]]

==ENABLING CLASSICAL PLANNING IN INTERACTIVE ENVIRONMENTS WITH INCOMPLETE MODELS - A simplified explanation of the abstract==


This abstract first appeared for US patent application 18066270 titled 'ENABLING CLASSICAL PLANNING IN INTERACTIVE ENVIRONMENTS WITH INCOMPLETE MODELS

'''Simplified Explanation''': 

The patent application describes a method in a model-based reinforcement learning system where an agent can switch between a probabilistic planning mode and an information gathering mode. In the probabilistic planning mode, the agent computes a predictive state representation to score action candidates based on resolved plans to achieve a goal state. In the information gathering mode, the agent scores action candidates based on the expected value of information to be gathered.

* The method involves switching an agent between a probabilistic planning mode and an information gathering mode.
* In the probabilistic planning mode, the agent computes a predictive state representation to score action candidates based on resolved plans to achieve a goal state.
* In the information gathering mode, the agent scores action candidates based on the expected value of information to be gathered.

'''Key Features and Innovation''':

* Switching between probabilistic planning and information gathering modes.
* Computing predictive state representation for scoring action candidates.
* Scoring based on resolved plans and expected value of information.

'''Potential Applications''':

* Robotics for decision-making processes.
* Autonomous vehicles for adaptive planning.
* Game AI for strategic decision-making.

'''Problems Solved''':

* Enhancing decision-making in dynamic environments.
* Improving goal achievement through planning.
* Optimizing information gathering for better outcomes.

'''Benefits''':

* Increased efficiency in decision-making.
* Enhanced goal achievement.
* Adaptive and flexible agent behavior.

'''Commercial Applications''''':

The technology can be applied in various industries such as autonomous vehicles, robotics, and gaming for improved decision-making processes and goal achievement.

'''Questions about Model-Based Reinforcement Learning''':

1. How does the agent switch between the probabilistic planning and information gathering modes?
2. What are the potential limitations of this model-based reinforcement learning system?


==Original Abstract Submitted==

A computer-implemented method in a model-based reinforcement learning (RL) system with logic states includes switching an agent between a first mode and a second mode, the first mode being a probabilistic planning mode and the second mode being an information gathering mode. In response to the agent being in the probabilistic planning mode, the agent computes a predictive state representation, given a history of observations and actions taken, and the agent scores action candidates based on planning with the predictive state representation so that actions with resolved plans with confidence to achieve a goal state are scored higher than actions without resolved plans. In response to the agent being in the information gathering mode, the agent scores action candidates based on a Q function of a value of expected information to be gathered from a given pair of state and action.



[[Category:G06N20/00]]
[[Category:CPC_G06N20/00]]