About XTreeM

XTreeM is an innovative AI that provides performance, scalability, and interpretability by stochastically mixing local linear models (*1). XTreeM has low prediction error for test data, i.e., it is one of the best models to accurately mimic the generative process of data with complex input-output relationships XTreeM uses a proprietary learning engine to stochastically XTreeM uses a unique learning engine to stochastically partition the data based on stationarity and linearity of the data while maintaining model generalization performance, and learns a (constrained) linear model appropriate for each group. This makes it possible to learn models that are robust to non-stationary data and the “curse of dimensionality. In addition, hyperparameter settings are automated, requiring only a few settings by the user, such as the type of regularization (*2) and data partitioning parameters. This frees the user from complex hyperparameter adjustments and makes it easy for anyone to create high-performance models. XTreeM opens up new possibilities in data science by facilitating optimal model design.

*1) The functions trained by XTreeM are similar to those of Mixtures of Experts (MoE), Gaussian processes or kriging models, and XTreeM is distinguished by its ease of use and high performance. In general, learning non-stationary, non-linear data can be challenging due to the large number of hyperparameter choices, such as the settings for each Expert and number for MoEs and the kernel design for Gaussian processes, making it difficult to find the optimal configuration.

*2) L1, L2、ElasticNet

XTreeM - XAI

XTreeM's advantages go beyond ease of training and predictive performance. eXplainable AI (XAI) is a general term for technologies and methods that make AI predictions easier for humans to understand. eXplainable AI (XAI) is the term used to refer to a set of technologies and methods that make AI predictions easier for humans to understand. a generic term for technologies and methods that make AI predictions easier for humans to understand. The goal of XAI is to increase the transparency of AI and improve the reliability of its predictions. XAI enables users to understand the reasoning and rationale behind predictions. This promotes fairness, accountability, and ethical use of AI. The use of XAI provides transparency and credibility to AI, making it safe and secure to use, ultimately increasing user trust and social acceptance. XTreeM is not only able to compute feature contributions globally, but also to rapidly compute and visualize the contribution of each feature on an instance-by-instance basis. xTreeM-XAI offers a SHAP-like visualization package, which allows users to visualize the contribution of each feature on an instance-by-instance basis, and to visualize the contribution of each feature on a global basis.

XTreeM - BO

XTreeM makes it easy to build highly accurate statistical models for Bayesian optimization. Bayesian optimization is one of the methods for efficiently solving optimization problems for functions that are expensive to evaluate. The objective is to find the optimal solution while minimizing the number of times the function is evaluated. In particular, it is expected to be an effective method for experimental design, material design experiments, and simulations, where direct evaluation of the function would be very time-consuming and costly. However, applying Bayesian optimization to real-world problems requires a high degree of specialized knowledge (*1), and XTreeM allows users to intuitively construct highly accurate statistical models for Bayesian optimization. XTreeM allows you to intuitively build highly accurate statistical models for Bayesian optimization, including Expected Improvement (EI), Probability of Improvement (PI), Upper Confidence Bound (UCB), Thompson Sampling (TS ), and Predictive Entropy Search (PES).

*1) Bayesian optimization uses a stochastic model (usually Gaussian process regression, GPR) to model function uncertainty. This model is not suitable for discontinuous objective functions or functions with abrupt changes. Also, when the objective function is non-stationary (when the properties of the function vary greatly from place to place), it is difficult to design the kernel function of the Gaussian process and set the hyperparameters. In addition, Bayesian optimization can have difficulties with problems of high dimensionality (more than a few dozen dimensions). This is known as the “curse of dimensionality,” where the search space explodes, making model training and exploration difficult. Therefore, it is necessary to devise dimension reduction techniques and design kernels that are suitable for high dimensions.

XTreeM - CF

XTreeM-CF offers two extensions to XTreeM-XAI: Scenario Analysis (What-If Analysis) and Counterfactual Explanation (CE). Scenario analysis simulates how a change in a particular feature affects the results; based on XAI, you can create scenarios involving variables that affect the results and compare how the results change when you run those scenarios. Using XTreeM as a Meta-Learner, a module for estimating causal effects is also available. Counterfactual explanations are explanatory techniques that show “how features should change in order to achieve an objective.

Installation flow

The XTreeM implementation process consists of four main steps. Each step provides professional support from initial goal setting to final system maintenance to ensure business success.

• 1. Project Launch: From the perspective of XTreeM's quality assurance, we will conduct an awareness meeting to gain a deep understanding of your issues and meet your expectations, including project goals and objectives, data overview, acquisition status and acquisition plan, and expectations.

• 2. PoC (Proof of Concept): The PoC may be conducted multiple times, depending on the status of data acquisition and other factors.

• 3. Co-Development: If necessary, we can perform custom development or system development to suit your environment.

• 4. Implementation: We will sign a license agreement and you will be able to use XTreeM.