Glossary of terms used in articles about AI, machine learning, neural networks and facial recognition

• Dataset — a collection of data. In the case of facial recognition, a dataset is a collection of images used for training a machine learning model, for example, a set of images and marks for them (labels, ground truth, etc.).
• Align faces — the same as face alignment, in a computer vision area, is a process of identifying the geometric structure of a face, such as a face size, eyes, a nose and mouth coordinations, and transforming input images by their rotation, translation and scaling, into output images with aligned faces. The possibilities of face alignment for 2D images are restricted. For example, a model cannot transform a side view of a face into a front view.
• Facial landmarks — key points on a face, such as eyes, a nose and mouth, detected, tracked and localized as coordinates by algorithms of neural networks.
• Fully-connected layer — a type of layer in neural networks where all the inputs or neurons from one layer are connected to every activation unit or neuron of the next layer.
• Top of the network — a part of the convolutional neural network, which consists of fully-connected layers and is intended for classification, meaning head or output layers of a network.
• Train, validation and test sets — for neural networks, a dataset of images is split into 3 sets. The train set contains data used to train a model. With a validation set, engineers evaluate model learning progress at the end of each epoch. The test set is a set of images used to assess the performance of a fully trained network.
• Backbone — in terms of neural networks, the main part of a convolutional neural network (CNN) that extracts features from an image, in our case, features are transmitted to the top of the network for classification.
• Freeze a backbone — the process of freezing the weights of a backbone. Freezing of the model’s weights is used to lock some layers of a model from changes while training other layers.
• Transfer learning — a popular deep learning approach of training an already pre-trained model on new types of data or training it for new purposes. Actually, it is reusing and retraining of a pre-trained model to save resources, as it requires less data than other approaches.
• Fine-tuning — a method of using weights from a trained neural model to retrain a new model to identify new classes for solving new tasks. It is usually a second step of the model training process, as it speeds up training and allows using smaller training datasets.
• Cyclic learning rate — a neural network training technique that helps to avoid overfitting and increase model’s accuracy (or other target metrics). As soon as a learning rate is a hyperparameter used to adjust the weights of a neural network, this technique allows setting cyclical variations of a learning rate between reasonable boundary values.

How can one apply modern technology to business needs?

Our articles dedicated to AI, machine learning, neural networks and image recognition, facial recognition, face mask detection, emotion recognition, and people counting solutions are here for your attention.

1. Face Mask Detection and Correct Mask Wearing Recognition Software. How to Save Your Business from Quarantine and Closure
2. Research of Descriptor-Based Image Normalization and Comparative Analysis of SURF, SIFT, BRISK, ORB, KAZE, AKAZE Descriptors
3. What Is Facial Recognition And How Is It Used?
4. The Beauty of Emotions Embodied in Software Analyzing Facial Expressions
5. Helping One Look Good: the Less Known Human Appearance-Related Uses of Image Recognition
6. Developing a Face Recognition System: How We’ve Done This and What We Have Learned Along the Way
7. 9 Things to Know While Looking to Purchase a Face Recognition App
8. Image Recognition: Can an Image Recognition App Become the Quality Boost Your Business Needs?
9. Face Recognition: What Can a Face Recognition App Be Capable of and How to Make It Happen?
10. Facial Recognition: the Advent of a New Era in Non-Digital Marketing?
11. Data Science and Business Analysis: a Marriage to Shape Most Businesses’ Future
12. Industry 4.0: What Can Help Fuel the Change?
13. Artificial Intelligence & Data Science In Telecom: How Can Telcos Gain?
14. 5 Popular Uses Of Face Recognition

Enjoy useful reading!