Machine learning is essentially a branch of data science that uses data to find models within it. By applying mathematical and statistical knowledge, these models are refined in a process called “training.”

While many associate machine learning (ML) and artificial intelligence (AI) with complex and futuristic technologies—like those depicted in films where AI rebels against humans—the reality today is far less advanced. Most current AI systems rely on “if-then” programming, responding only to predefined human instructions. Machine learning, however, goes a step further by enabling models to self-correct and improve.

Categories of Machine Learning

Machine learning can be broadly divided into the following types:

1. Supervised Learning

• Relies on labeled datasets for training.

• The system learns to map inputs to outputs based on provided examples.

• Example: Teaching a machine to identify butterflies among insects by providing labeled data with specific characteristics such as leg count, wings, antennae structure, and body proportions.

• After training, the machine can analyze new insect images and determine whether they are butterflies based on its learned model.

2. Unsupervised Learning

• Works without labeled data; the machine identifies patterns and clusters data points based on their features.

• Example: In identifying butterflies, no labels are provided. The machine must independently identify key features like wings, legs, and body structure to differentiate butterflies from other insects.

• Popular unsupervised learning models include Generative Adversarial Networks (GANs) and clustering algorithms.

• Limitations: Unsupervised models are often considered “black boxes,” where the internal processes are not fully transparent to developers.

Popular Machine Learning Algorithms

There are many machine learning algorithms, each tailored for specific applications. Some well-known examples include:

• Neural Networks

• Decision Trees

• Support Vector Machines (SVMs)

• Bayesian Classifiers

• Clustering Methods

One widely used and beginner-friendly algorithm is K-Means Clustering, which groups data points into clusters based on similarity.

1.K-Means Algorithm: A Simple Unsupervised Clustering Method

K-Means is a straightforward and widely used unsupervised clustering algorithm. The core idea of clustering is to group samples based on their distance or similarity, clustering similar (or close) samples together while separating dissimilar (or distant) samples into different groups. The basic concept of K-Means is to iteratively partition a dataset into K clusters such that the mean of each cluster minimizes the Sum of Squared Errors (SSE) for all points within that cluster.

In mathematical language, for a sample set, K-means minimizes the error function of clustering.

The visualization above effectively illustrates the concept: each blue or red point’s variance to its respective cluster center is minimized. For clusters red and blue, the two central points meet the condition of minimal SSE for this partitioning.

Example: K = 2 (Dividing the Samples into Two Clusters)

Visually, if we want to divide the green points in Figure A into two clusters, we could draw a diagonal line from the upper left to the lower right. But how do we instruct a machine to perform this partitioning accurately? The steps are as follows:

1. Random Initialization:

Select two random points (e.g., red and blue) in the coordinate system as the initial cluster centers. For each point in the dataset, calculate its distance to the red and blue points.

• Assign the point to the nearest cluster center: if it’s closer to the blue point, color it blue; if it’s closer to the red point, color it red.

• Iterate through all points until every point is assigned a color, as shown in Step 1.

2. Recalculate Cluster Centers:

Recompute the center of each cluster (mean of all points within a cluster). Adjust the cluster centers to minimize the mean squared error (MSE) of all points within that cluster, as shown in Step 2.

3. Repeat Until Convergence:

Repeat steps 1 and 2 until the cluster centers stabilize and the process converges. This iterative process completes as shown in Steps 3-6.

Blockchain and Machine Learning

Blockchain possesses two key attributes that make it an excellent match for advancing machine learning and artificial intelligence:

1. Privacy: Blockchain enables training on sensitive private data without compromising it.

2. Incentive Mechanism: Blockchain’s unique reward system allows users to earn rewards for sharing data or publishing models on-chain. Anyone can sell their data while maintaining privacy, and developers can publish and train their models on-chain, receiving incentives in return.

The image referenced demonstrates the standardization and commoditization cycle in technology. The era of data monopolization is nearing its end. Blockchain’s accessibility ushers in opportunities for broader participation in the next technological era.

The three critical factors in machine learning are algorithms, computational power, and data. Tech giants like Amazon, Apple, Google, and Facebook dominate computational and data resources, enabling them to maintain a significant lead. Blockchain introduces a chance to disrupt this dominance by decentralizing machine learning. This shift can refocus value creation from data-driven approaches to algorithm-driven innovation. Blockchain’s economic incentives provide a fertile environment for developers to train and optimize models, fostering growth for algorithm developers.

2.1 Project Overview

Integrating machine learning and blockchain is a nascent yet promising field. While many companies are developing tools in this space, the potential remains vast. Below are three notable projects:

2.1.1 DeepBrain Chain

DeepBrain Chain, founded in November 2017, aims to build the world’s largest distributed high-performance computing network powered by blockchain, serving as critical infrastructure for the 5G+AI era.

• Key Components:

1. High-Performance Computing Network: Launched in August 2018.

2. Blockchain Mainnet: Officially launched on May 20, 2021, based on Polkadot’s Substrate framework.

DeepBrain Chain is one of the few blockchain projects to achieve large-scale deployment in high-performance computing. It has made significant progress in usability and commercialization, with applications across blockchain, AI, cloud gaming, visual rendering, biopharmaceuticals, and semiconductor simulations.

Over 50 global companies have deployed high-performance GPU cloud platforms on its network, serving hundreds of businesses and tens of thousands of AI developers.

2.1.2 Numerai

Numerai is a hedge fund leveraging a global community of anonymous data scientists to predict future prices. Combining distributed intelligence, machine learning, blockchain, and tokenization, Numerai creates a new model of fund management driven by collective intelligence.

• Data Distribution:

Data scientists receive anonymized datasets as input for their predictive models. These datasets could include macroeconomic indicators, commodity prices, or exchange rates—proprietary data typically inaccessible to outsiders. Removing metadata allows Numerai to share the data openly.

• Competitions:

Data forms the foundation for competitions to create the most effective predictive models. Historical data with known outcomes helps validate models, while real-time data remains uncertain for critical predictions.

• Scoring and Rewards:

Predictions are scored after being uploaded to Numerai’s “meta-model.” Data scientists are rewarded based on how much their predictions improve the meta-model.

• Intellectual Property:

Participants retain ownership of their models and continue to receive rewards if their models enhance Numerai’s meta-model.

2.1.3 HUMAN Protocol

HUMAN Protocol is a decentralized framework that rewards contributions based on knowledge and skills, bridging AI and machine learning through hybrid models.

Built on blockchain, HUMAN Protocol streamlines management and settlement processes to create an auditable, open-source infrastructure for decentralized labor markets, connecting data seekers with knowledge marketplaces.

• Application in Machine Learning:

The protocol directly addresses labor-intensive tasks in machine learning, such as manual annotation and verification of model inference quality, to make datasets more suitable for training.

• Future Vision:

3.Closing remarks

Beyond its current applications, HUMAN Protocol is designed for the next evolution of machine intelligence, where machines directly query humans for data to refine themselves.

Machine learning and blockchain are two of the most exciting and cutting-edge technologies today. Machine learning serves as the foundation for artificial intelligence and big data, while blockchain has the potential to revolutionize the current financial architecture. Both technologies are data-driven, naturally enabling synergy in certain research directions.

Blockchain offers secure and efficient data sharing and analysis, while machine learning harnesses this data to drive technological advancements. We are already witnessing many developers working on feasible projects in this intersection, and more resources and talent are being funneled into these fields.

The shared and complementary capabilities of these technologies will continue to propel each other forward. Riding the wave of blockchain and navigating the ship of machine learning, we are set to explore the vast, starry ocean of future technology!

Introduction to Machine Learning and Its Applications in Blockchain