How Convolution Works

Brandon Rohrer · Beginner ·📐 ML Fundamentals ·5y ago
A guided tour through convolution in two dimensions for convolutional neural networks and image processing End-to-End Machine Learning Course 322: https://e2eml.school/322 Tutorial on convolution in one dimension: https://e2eml.school/convolution_one_d.html Code for the animations: https://gitlab.com/brohrer/convolution-2d-animation
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from Brandon Rohrer · Brandon Rohrer · 57 of 60

← Previous Next →
1 Robot Learning with a Biologically-Inspired Brain (BECCA)
Robot Learning with a Biologically-Inspired Brain (BECCA)
Brandon Rohrer
 2 BECCA talk at AGI 2011
BECCA talk at AGI 2011
Brandon Rohrer
 3 Robot Learning with a Biologically-Inspired Brain (BECCA), The Sequel
Robot Learning with a Biologically-Inspired Brain (BECCA), The Sequel
Brandon Rohrer
 4 BECCA listens to The Hobbit
BECCA listens to The Hobbit
Brandon Rohrer
 5 Learning the building blocks of speech: BECCA extracts a hierarchy of audio features
Learning the building blocks of speech: BECCA extracts a hierarchy of audio features
Brandon Rohrer
 6 BECCA listens for sound effects in The Hobbit
BECCA listens for sound effects in The Hobbit
Brandon Rohrer
 7 BECCA finds movie trailers while watching the Big Bang Theory
BECCA finds movie trailers while watching the Big Bang Theory
Brandon Rohrer
 8 Listening for unexpected sounds: BECCA detects anomalies in audio data
Listening for unexpected sounds: BECCA detects anomalies in audio data
Brandon Rohrer
 9 Learning the building blocks of vision: BECCA extracts a spatio-temporal hierarchy of features
Learning the building blocks of vision: BECCA extracts a spatio-temporal hierarchy of features
Brandon Rohrer
 10 Watching for the unexpected: BECCA detects anomalies in video data
Watching for the unexpected: BECCA detects anomalies in video data
Brandon Rohrer
 11 BECCA finds a stationary target
BECCA finds a stationary target
Brandon Rohrer
 12 BECCA finds a stationary target at 3X speed
BECCA finds a stationary target at 3X speed
Brandon Rohrer
 13 BECCA watches the X-men and Bruce Lee
BECCA watches the X-men and Bruce Lee
Brandon Rohrer
 14 BECCA plays Quidditch
BECCA plays Quidditch
Brandon Rohrer
 15 BECCA chases a ball
BECCA chases a ball
Brandon Rohrer
 16 BECCA chases a ball, part 2
BECCA chases a ball, part 2
Brandon Rohrer
 17 Becca chases a ball, part 3
Becca chases a ball, part 3
Brandon Rohrer
 18 BECCA creates features from MNIST
BECCA creates features from MNIST
Brandon Rohrer
 19 How reinforcement learning works in Becca 7
How reinforcement learning works in Becca 7
Brandon Rohrer
 20 Deep Learning Demystified
Deep Learning Demystified
Brandon Rohrer
 21 How Data Science Works
How Data Science Works
Brandon Rohrer
 22 How Convolutional Neural Networks work
How Convolutional Neural Networks work
Brandon Rohrer
 23 How Bayes Theorem works
How Bayes Theorem works
Brandon Rohrer
 24 How Deep Neural Networks Work
How Deep Neural Networks Work
Brandon Rohrer
 25 Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM)
Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM)
Brandon Rohrer
 26 How Support Vector Machines work / How to open a black box
How Support Vector Machines work / How to open a black box
Brandon Rohrer
 27 How autocorrelation works
How autocorrelation works
Brandon Rohrer
 28 Getting closer to human intelligence through robotics
Getting closer to human intelligence through robotics
Brandon Rohrer
 29 A minimalist's guide to slicing and indexing pandas DataFrames
A minimalist's guide to slicing and indexing pandas DataFrames
Brandon Rohrer
 30 How decision trees work
How decision trees work
Brandon Rohrer
 31 Data scientist archetypes
Data scientist archetypes
Brandon Rohrer
 32 How to use python's datetime package
How to use python's datetime package
Brandon Rohrer
 33 How optimization for machine learning works, part 1
How optimization for machine learning works, part 1
Brandon Rohrer
 34 How optimization for machine learning works, part 2
How optimization for machine learning works, part 2
Brandon Rohrer
 35 How optimization for machine learning works, part 3
How optimization for machine learning works, part 3
Brandon Rohrer
 36 How optimization for machine learning works, part 4
How optimization for machine learning works, part 4
Brandon Rohrer
 37 How convolutional neural networks work, in depth
How convolutional neural networks work, in depth
Brandon Rohrer
 38 How to pick a machine learning model 4: Splitting the data
How to pick a machine learning model 4: Splitting the data
Brandon Rohrer
 39 How to pick a machine learning model 3: Choosing a loss function
How to pick a machine learning model 3: Choosing a loss function
Brandon Rohrer
 40 How to pick a machine learning model 2: Separating signal from noise
How to pick a machine learning model 2: Separating signal from noise
Brandon Rohrer
 41 How to pick a machine learning model 1: Choosing between models
How to pick a machine learning model 1: Choosing between models
Brandon Rohrer
 42 How to pick a machine learning model 5: Navigating assumptions
How to pick a machine learning model 5: Navigating assumptions
Brandon Rohrer
 43 What do neural networks learn?
What do neural networks learn?
Brandon Rohrer
 44 Interview with iRobot's Director of Data Science Angela Bassa
Interview with iRobot's Director of Data Science Angela Bassa
Brandon Rohrer
 45 How Backpropagation Works
How Backpropagation Works
Brandon Rohrer
 46 Evolutionary Powell's method: A discrete optimizer for hyperparameter optimization
Evolutionary Powell's method: A discrete optimizer for hyperparameter optimization
Brandon Rohrer
 47 1D convolution for neural networks, part 1: Sliding dot product
1D convolution for neural networks, part 1: Sliding dot product
Brandon Rohrer
 48 1D convolution for neural networks, part 2: Convolution copies the kernel
1D convolution for neural networks, part 2: Convolution copies the kernel
Brandon Rohrer
 49 1D convolution for neural networks, part 3: Sliding dot product equations longhand
1D convolution for neural networks, part 3: Sliding dot product equations longhand
Brandon Rohrer
 50 1D convolution for neural networks, part 4: Convolution equation
1D convolution for neural networks, part 4: Convolution equation
Brandon Rohrer
 51 1D convolution for neural networks, part 5: Backpropagation
1D convolution for neural networks, part 5: Backpropagation
Brandon Rohrer
 52 1D convolution for neural networks, part 6: Input gradient
1D convolution for neural networks, part 6: Input gradient
Brandon Rohrer
 53 1D convolution for neural networks, part 7: Weight gradient
1D convolution for neural networks, part 7: Weight gradient
Brandon Rohrer
 54 1D convolution for neural networks, part 8: Padding
1D convolution for neural networks, part 8: Padding
Brandon Rohrer
 55 1D convolution for neural networks, part 9: Stride
1D convolution for neural networks, part 9: Stride
Brandon Rohrer
 56 The Four Grand Challenges of Robots in the Home
The Four Grand Challenges of Robots in the Home
Brandon Rohrer
 ▶ How Convolution Works
How Convolution Works
Brandon Rohrer
 58 The Softmax neural network layer
The Softmax neural network layer
Brandon Rohrer
 59 Batch normalization
Batch normalization
Brandon Rohrer
 60 Getting ready to learn Python, Mac edition #1: Files and directories
Getting ready to learn Python, Mac edition #1: Files and directories
Brandon Rohrer
⚡ AI Lesson Summary ✦ V3 skills ⚖ Mixed

This video teaches how convolution works in two dimensions for convolutional neural networks and image processing. It covers the basics of convolution, including how to reverse the kernel, move it over the image, and calculate the feature map. By the end of this video, you will understand how to apply convolution to images and extract features from them.

Key Takeaways
  1. Reverse the kernel by flipping it top to bottom and left to right
  2. Move the reversed kernel over the image, element by element, multiplying corresponding pixel values and adding up the results
  3. Repeat the process for all rows and columns of the image to produce a feature map or a scaled-down version of the original image
  4. Use zero padding to make the result the same size as the original image, if desired
  5. Implement convolutional neural network and calculate gradients for back propagation
💡 The convolution operation can be used to extract features from an image, such as edges, lines, or shapes, and can be used in a variety of applications, including image classification, object detection, and image segmentation.

Related AI Lessons

⚡
ML Compilers Aren’t All the Same — Here’s Why
ML compilers like PyTorch, TensorRT, and CoreML have different architectures and design choices, impacting their performance and compatibility across hardware and workloads.
Medium · Programming
 ⚡
ML Compilers Aren’t All the Same — Here’s Why
Learn why ML compilers differ in architecture and design, and how these differences impact model deployment and performance
Medium · Deep Learning
 ⚡
ML Compilers Aren’t All the Same — Here’s Why
Learn why ML compilers differ in architecture and design, and how these differences impact model deployment and performance
Medium · LLM
 ⚡
How Machines Draw the Line: A Beginner’s Guide to Regression
Learn the basics of regression in machine learning and how it's used to make predictions, with examples and explanations for beginners
Medium · Machine Learning
Up next
Abstracting with Functions
Coursera
Watch →