Reasoning

An effective AI approach based on Bayesian Network.

Ardi Machine Reasoning

The ability to reach logical conclusions based on prior information is an essential part of the human cognition. In other words, the human brain’s comprehension function through reasoning is a big part that makes human intelligence unique. With memorized information, humans comprehend and understand the relationship between events, and therefore, are able to solve problems based on these relationships between different pieces of information.

One of the main challenges in building an efficient AI system is the ability to learn and reason under uncertainty.

Ardi’s proven successful approach is based on the framework of Bayesian Network, which offer an expressive visual and quantitative tool for:

  • Learning and representing reasoning procedures
  • Understanding causality among variables
  • With this strong foundation, Ardi's Reasoning may improve risk behavior prediction accuracy by 10 times.

    Image

    Types of Reasoning

    Ardi Reasoning corporates diagnostic, predictive, inter-causal and a combined method to make logical arguments. Users have the ability to incorporate Bayesian network discovery, as well as other behavior prediction methods to determine the causality between data and/or calculate the conditional probability with given conditions.

    Diagnostic: Given evidence about an effect, how does this change the beliefs in this causes?

    Predictive: Given evidence, what are the predicted outcomes?

    Intercausal: Given evidence about a cause and its effect, how does it change the beliefs in other causes?

    Combined: Given evidence about background causes and effects, what are the new beliefs in intermediate nodes?


    Image

    Bayesian Network for Reasoning

    Graph representation of real-world data

  • Conditional independencies and graphical language capture structure of many real-world distributions
  • Graph structure provides more insights and allows in-depth domain knowledge discovery through network construction
  • Expert prior knowledge may often be incorporated when learning the graph structure
  • Learned Bayesian model solves analytical limitations

  • Learned model can be used for various tasks
  • Support all features of probabilistic learning
  • Deal with missing data and hidden variables

  • Image

    Use Case

    Laurent is a financial analyst. As the international economic integration is deepening, when a certain industry or enterprise encounters a crisis, it affects the micro and macro economy and could cause great harm. Therefore, in order to ensure the stability of the economy and the smooth and effective operations of enterprises, it is of great significance to construct financial management theories, establish financial risk early warning mechanisms, and accurately identify crisis signals to predict potential financial risks. Graphen’s Reasoning function makes Laurent’s work much more efficient.

    Ardi uses Bayesian Network to combine financial indicators and non-financial indicators, learning Bayesian network structure with reference to DuPont analysis and expert knowledge, forming a complex topology map, determining the mutual influential relationship between various indicators, that is, the dependence between each routine. It also collects the company’s historical data over the past three years and conducts Bayesian network parameter learning to dynamically reflect and predict the company’s financial status.


    Image

    Request a demo today.