当前位置:网站首页>Interpretation of deepmind's latest paper: the causal reasoning algorithm in discrete probability tree is proposed for the first time

Interpretation of deepmind's latest paper: the causal reasoning algorithm in discrete probability tree is proposed for the first time

2020-11-08 12:56:56 EXP {MM/YYYYY}


translate | Gao Weihua

Produce | AI Technology base

The first figure  | CSDN Pay to download from visual China

At present , Some of the frontiers AI Researchers are looking for clear semantic models to represent context specific causal dependencies , This is necessary for causal induction , stay DeepMind This probability tree model can be seen in the algorithm of .

Probability tree graph is used to represent probability space , A tree diagram can describe a series of independent events or conditional probabilities . 

The nodes on the probability tree represent an event and its probability , The root node represents a probability equal to 1 Specific events of , The set of sibling nodes represents the parent event (the parent event.) A detailed division of .

probability ( A series of events that cause a particular node to occur )= probability ( The node )* probability ( Parent node ).

Probability trees have been around for decades , But it hasn't been ML and AI Too much attention from fans .

new DeepMind The paper  《 Causal reasoning algorithm in probability tree 》 wrote ,“ Probability tree is one of the simplest models of causal generation .”  According to the author , The above algorithm is the first specific algorithm for causal reasoning in discrete probability trees .

Scientists in the field of cognition say , Humans naturally learn to reason by drawing causal relationships from observations , And it's very effective . Although observational data are limited and scarce , But people can quickly understand the causal structure , For example, by observing the co-occurrence frequency of causality and the interaction between physical objects .

Causal induction is a classic problem in machine learning and statistics .  Causal Bayesian Networks (CBN) The model can describe the causal dependency in causal induction , But it can't express context specific independence .

According to the DeepMind The team says , DeepMind Covering the entire causal hierarchy and random propositions 、 Operations on causal events , And the causal reasoning is further extended to “ A very general class of discrete stochastic processes ”.

 DeepMind The team's focus is on finite probability trees , And get the following specific algorithm :

Compute the following to form the minimum representation of any event :

  • Propositional calculus (Propositional calculus)

  • Causal precedent (Causal precedences)

Calculate the basic operations of the following three causal hierarchies  :

  • Conditions

  • Interventions

  • counterfactual

Link to the original text :

https://www.marktechpost.com/2020/10/30/deepmind-research-introduces-algorithms-for-causal-reasoning-in-probability-trees/

source : 

https://syncedreview.com/2020/10/29/deepmind-introduces-algorithms-for-causal-reasoning-in-probability-trees/

The paper :

https : //arxiv.org/pdf/2010.12237.pdf

GitHub:

https://github.com/deepmind/deepmindresearch/tree/master/causal_reasoning

 More highlights 

版权声明
本文为[EXP {MM/YYYYY}]所创,转载请带上原文链接,感谢