// MODEL OPTIMIZATION AND PROMPT SYNTAX TERM
MAE
Stands for Mean Absolute Error, a metric that measures the average magnitude of the errors in a set of predictions, without considering their direction.
TECHNICAL DEFINITION
Mean Absolute Error (MAE) is a regression evaluation metric that calculates the average of the absolute differences between predicted values and actual values, providing a robust measure of prediction error that is less sensitive to outliers than MSE.
BACKGROUND
Human–AI interaction is a field of research and a sub-field of human–computer interaction, focusing on user experience and psychological factors. With the proliferation of artificial intelligence (AI), there has developed a subsection of HCI research dedicated to artificial intelligence and how people interact with and are impacted by it.
READ MORE ON WIKIPEDIASYNONYMS & ALIASES
- Mean Absolute Deviation
- L1 Loss
USAGE NOTE
Useful when outliers should not heavily influence the error metric, or when the error magnitude is more important than its squared value.
DEVELOPERS
Organizations developing technology related to MAE.
Originally proposed and developed Masked Autoencoders (MAE) in their influential 2021 paper, 'Masked Autoencoders Are Scalable Vision Learners'. The lab continues to be a leader in self-supervised learning research for vision and other domains.
Conducts extensive research on self-supervised learning and transformer-based models. Their work on Masked Language Models (like in BERT) was a conceptual precursor to MAE, and they continue to develop related architectures for vision and multimodal AI.
Develops related self-supervised learning techniques for vision transformers. Their BEiT (Bidirectional Encoder representation from Image Transformers) model uses a masked image modeling task that is conceptually similar to MAE.
A leading academic lab that conducts foundational research in computer vision and self-supervised learning. Researchers at BAIR analyze, critique, and build upon state-of-the-art models like MAE.
An AI company that provides widely used open-source tools and platforms. They develop and maintain implementations of MAE and similar models in their `transformers` library, making the technology accessible for AI engineers to build with.
A prominent university research lab that contributes significantly to machine learning, particularly in computer vision. Their research frequently involves developing and applying novel self-supervised learning methods related to masked autoencoding.
Focuses on developing AI models and the hardware/software systems to train them efficiently. They research and implement scalable model architectures and training techniques, including those based on or inspired by MAE, for tasks in computer graphics and vision.