Machine Learning in Kotlin: Overview of KMath, Smile, KotlinDL, and DeepLearning4J Libraries

Machine Learning in Kotlin: From Mathematics to Deep Neural Networks

Kotlin is a modern, concise, and safe language that is increasingly used not only for Android development but also for server-side projects, data analysis, and machine learning. Thanks to full compatibility with Java and its own library ecosystem, Kotlin allows you to write clean and expressive code for ML tasks. In this article, we will review key libraries for machine learning in Kotlin: from low-level mathematical operations to ready-made neural network frameworks.

1. KMath — The Mathematical Foundation for ML

KMath (Kotlin Mathematics) is a library for scientific computing created by JetBrains. It provides powerful tools for linear algebra, optimization, statistics, and working with tensors. KMath is ideal for implementing your own machine learning algorithms without being tied to Java.

Key features of KMath:

• Working with vectors, matrices, and multidimensional arrays (tensors);
• Linear algebra (LU decomposition, SVD, eigenvalues);
• Statistical distributions and tests;
• Automatic differentiation (for gradient descent);
• Integration with NumPy via export to ndarray.

// Example: Linear regression using KMathimport space.kscience.kmath.operations.*import space.kscience.kmath.structures.*import space.kscience.kmath.linear.*

fun main() {    // Data: x = [1, 2, 3], y = [2, 4, 6] (target function y = 2x)    val x = Matrix.real(3, 2) { i, j -> if (j == 0) 1.0 else (i + 1).toDouble() }    val y = Matrix.real(3, 1) { i, _ -> (2 * (i + 1)).toDouble() }

    // Solving the normal equation: w = (X^T * X)^(-1) * X^T * y    val xt = x.transpose()    val xtx = xt dot x    val invXtx = xtx.inverse() // LU decomposition    val w = invXtx dot (xt dot y)

    println("Coefficients: intercept = ${w[0, 0]}, slope = ${w[1, 0]}")}

KMath is the choice for those who want to control every step of the computation and not depend on Java libraries.

2. Smile — Classic ML Algorithms in Kotlin

Smile (Statistical Machine Intelligence & Learning Engine) is one of the most comprehensive machine learning libraries on the JVM, written in Java but with excellent Kotlin support. It includes dozens of algorithms: from linear regression and SVM to random forests and neural networks.

Key features of Smile:

• Clustering (K-Means, DBSCAN, GMM);
• Classification (SVM, Decision Trees, Random Forest, Naive Bayes);
• Regression (linear, ridge, LASSO);
• Dimensionality reduction (PCA, t-SNE, UMAP);
• Visualization (built-in charts for Swing).

// Example: Classification with Random Forest in Smileimport smile.classification.RandomForestimport smile.data.DataFrameimport smile.data.`type`.DataTypesimport smile.io.Read

fun main() {    // Loading data (example with iris dataset)    val iris = Read.arff("data/iris.arff")

    // Splitting into features and target variable    val x = iris.select("sepal_length", "sepal_width", "petal_length", "petal_width").toArray()    val y = iris.stringVector("class").toArray()

    // Training Random Forest (100 trees)    val model = RandomForest.fit(x, y)

    // Prediction    val testSample = doubleArrayOf(5.1, 3.5, 1.4, 0.2)    val prediction = model.predict(testSample)    println("Predicted class: $prediction")}

Smile is great for rapid prototyping and production tasks where time-tested algorithms are needed.

3. KotlinDL — Deep Learning from JetBrains

KotlinDL is a deep learning library created by JetBrains specifically for Kotlin. It provides a high-level AP