AI, machine learning and deep learning · Introductory

Data Reduction

Written by SignalsLab

Early edition: these lessons are being reviewed.

Content

Summary

Data reduction techniques minimize dataset volume while preserving analytical integrity through dimensionality, numerosity, and compression methods. The Discrete Wavelet Transform (DWT) is a key compression tool that transforms data into coefficients, allowing for efficient lossy compression by truncating less significant values.

Intuition

Think of data reduction as packing for a trip: you want to bring everything you need (analytical integrity) but fit it into a smaller bag (reduced volume). The goal is efficiency; the time spent packing must not exceed the time saved by traveling lighter. In the context of signals, the Discrete Wavelet Transform acts like a smart filter. It breaks a signal into different frequency components. Unlike other methods that might blur the whole picture, DWT keeps local details sharp. By keeping only the most important 'coefficients' (the high-value items in your bag) and discarding the rest, you get a smaller dataset that still tells the same story.

Key equations

X=(x1,x2,...,xn)
(1)

An n-dimensional data vector representing n measurements from n database attributes.

(x2i,x2i+1)
(2)

Pairs of data points to which the smoothing and difference functions are applied.

Derivation

  1. Data reduction aims to produce a smaller representation that maintains the integrity of the original data.
  2. The Discrete Wavelet Transform (DWT) converts an input vector X into a vector X' of wavelet coefficients.
  3. The input length L must be an integer power of 2, achieved by zero-padding if necessary.
  4. The algorithm uses a hierarchical pyramid approach, halving the data at each iteration.
  5. Smoothing and weighted difference functions are applied to pairs of data points.
  6. This process separates low-frequency and high-frequency content recursively until the data length is 2.
  7. Truncating selected coefficients allows for lossy compression, approximating the original data.
  8. DWT is preferred over DFT for lossy compression because it is localized in space and conserves local detail.

Worked example

Problem

You have a dataset vector of length 8 that needs to be compressed using DWT. How is the initial processing structured?

  1. Verify the length L=8 is an integer power of 2 (2^3).
  2. Apply smoothing and difference functions to pairs (x1,x2), (x3,x4), etc.
  3. This produces two sets of length 4 (low and high frequency).
  4. Recursively apply the process to the low-frequency set until length 2 is reached.

Answer The data is recursively halved using smoothing and difference functions on pairs, starting from length 8 down to 2, creating wavelet coefficients that can be truncated.

Common mistakes

  • Assuming DWT increases data size: While the transformed vector X' is the same length as X, reduction is achieved by truncating coefficients.
  • Confusing DWT with DFT: DWT is preferred for lossy compression because it is localized in space and preserves local detail better than DFT.
  • Ignoring input length constraints: The input vector length L must be an integer power of 2, requiring zero-padding if necessary.

Animation

Animation coming soon

The animated version of this lesson is being prepared for the web. In the Signals AI app, lessons are drawn step by step.

Code

This is a text lesson: it has no animation code yet.

ccBYNC© 2026 SignalsLab. Lesson text shared as is under CC BY-NC 4.0.