Critical AI Skill: Linear Algebra

Linear Algebra is a critical skill for AI leadership. It should be mandatory learning as part of high school and undergraduate curriculums. CHROs should emphasize this skill in their emerging leadership programs. Having basic skills in Linear Algebra allows professionals to proactively manage machines. You do not need to perform the calculations yourself. However, understanding the process will make your conversations with the engineering team more effective. Let me walk you through an example of how matrix operations are used in Computer Vision. The infographic below provides an explanation of a Convolutional Neural Network, otherwise known as a CNN.

If you want to be a leader in the Age of AI, take a class on Linear Algebra. This is a critical skill for leading machines, leading people that build machines, and leading organizations that adopt AI.

The network takes an image and develops an input table, filter kernel, and stride number. The filter kernel is a hyperparameter that detects key features in the image using learnable weights. The stride is also a hyperparameter that determines the number of pixels the filter kernel moves across the input table. A process called im2col (Image to Column) translates a section of the image into a column vector, which then completes a convolution using matrix multiplication. Hence, the name of the neural network. The filter kernel slides across the input image using the stride number. Each section is flattened into a column vector, which then forms an im2col matrix.

After that, matrix multiplication is performed by multiplying the filter row by the im2col matrix. This results in a final feature map, which is passed into an activation function (e.g. ReLU) that introduces non-linearity into the model by setting negative values to zero. The dimensions are then reduced by performing max pooling to identify key features from a small window. This makes the model more attuned to small changes in the original image.

The final step is to take the feature map that has passed through convolution and pooling and feed it into the CNN's fully connected layers as a single one dimensional vector. The layers then perform the final calculations to determine the classification of the image (i.e., this is an eagle, star, gravel road). If you want to be a leader in the Age of AI, take a class on Linear Algebra. It will make a big difference, especially if you are in a non-engineering role.

Linear algebra plays a major role in Computer Vision calculations