Embedding Vector

Definition

An embedding vector is a series of numbers and can be considered as a matrix with only one row but multiple columns, such as [2,0,1,9,0,6,3,0].

An embedding vector includes information representing the characteristics of an object, such as RGB (red-green-blue) color descriptions. A color can be described by the proportions of red, green, and blue. An embedding vector in RGB could be [R, G, B].

Advantages

Advances in modern computer and machine learning technologies have led to massive amounts of multimedia data in diverse application fields such as security, medicine, education, and online information services. A multimedia object cannot be simply described by alphanumeric data because a multimedia object have multiple dimensions of properties. For example, it is impossible to describe an image of a human face with a few numeric parameters or text strings.

Instead, embedding vectors describe an object in a multi-dimensional, easily analyzable way, and are suitable to represent numeric or symbolic characteristics of multimedia content. For example, an embedding vector of hundreds of dimensions is used to precisely describe a face.

Embedding vectors are important for many different fields of machine learning and pattern recognition. Machine learning algorithms typically require a numerical representation of objects in order for the algorithms to perform statistical analysis.

Scenarios

Embedding vectors, with its effectiveness and practicality of numerically representing objects, are used widely in different fields of machine learning.

  • Image processing

    Features can be gradient magnitudes, colors, grayscale intensities, edges, areas, and more. Embedding vectors are particularly popular in image processing because it is easy to define numeric attributes for images.

  • Speech recognition

    Features can be sound lengths, noise levels, noise ratios, and more.

  • Spam filtering

    Features can be IP addresses, text structures, frequencies of certain words, certain email headers, and more.

Edit
© 2019 - 2020 Milvus. All rights reserved.