Abstract:We introduce geometric pattern mining, the problem of finding recurring local structure in discrete, geometric matrices. It differs from existing pattern mining problems by identifying complex spatial relations between elements, resulting in arbitrarily shaped patterns. After we formalise this new type of pattern mining, we propose an approach to selecting a set of patterns using the Minimum Description Length principle. We demonstrate the potential of our approach by introducing Vouw, a heuristic algorithm for mining exact geometric patterns. We show that Vouw delivers high-quality results with a synthetic benchmark.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is **finding repeatedly occurring local structures in discrete geometric matrices**, that is, **the geometric pattern mining problem**. Specifically: 1. **Limitations of existing pattern - mining problems**: Traditional pattern mining mainly focuses on patterns of types such as item sets, sub - graphs and sequences, while relatively little research has been done on raster - based data (such as satellite images, texture recognition, etc.), especially for mining complex spatial relationships of non - rectangular shapes. 2. **Introduction of new pattern types**: The paper proposes **geometric patterns of arbitrary shapes**. These patterns are geometrically connected, that is, one element can be reached from another by traversing only the elements within the pattern. In addition, these patterns can contain multiple possible values (including the Boolean case). 3. **Model selection problem**: Since many geometric patterns can be found in a typical matrix, it is necessary to find a set of patterns that are compact and can well describe the data structure. This is formalized as a model - selection problem, where the model is defined by a set of patterns. The paper uses the **Minimum Description Length (MDL) principle** to solve this problem, looking for the laws that describe the data by compressing the data. 4. **Algorithm implementation**: To achieve this goal, the author proposes a heuristic algorithm named **Vouw**, which can provide high - quality results in synthetic benchmark tests and can accurately recover the implanted patterns. In summary, this paper aims to fill the gaps in existing pattern - mining methods when dealing with raster - based data by introducing the geometric pattern - mining problem and its solutions, and to provide an effective pattern - selection method through the MDL principle.

Vouw: Geometric Pattern Mining using the MDL Principle

Spatial Co-Location Pattern Discovery Without Thresholds

Mint: MDL-based approach for Mining INTeresting Numerical Pattern Sets

A Two-Phase Approach for Unexpected Pattern Mining.

KG-MDL: Mining Graph Patterns in Knowledge Graphs with the MDL Principle

Structural Textile Pattern Recognition and Processing Based on Hypergraphs

CODEM: A Novel Spatial Co-location and De-location Patterns Mining Algorithm

NWP-Miner: Nonoverlapping weak-gap sequential pattern mining

GCG: Mining Maximal Complete Graph Patterns from Large Spatial Data

Visual Pattern-Driven Exploration of Big Data

Hierarchical Visual Event Pattern Mining and Its Applications

Mining Frequent Neighborhood Patterns in Large Labeled Graphs

Mining Maximal Pattern-Based Subspace Clusters in High Dimensional Space

Mid-level Deep Pattern Mining

Mining evolutions of complex spatial objects using a single-attributed Directed Acyclic Graph

Efficient Mining Of Minimal Distinguishing Subgraph Patterns From Graph Databases

Pattern Morphing for Efficient Graph Mining

Effective algorithm for maximal pattern-based subspace clustering

A Framework for Pattern Mining and Anomaly Detection in Multi-dimensional Time Series and Event Logs

Mining Frequent Patterns in Process Models

Identification and Uses of Deep Learning Backbones via Pattern Mining