Part 3/8:
Classification: Assessing activities based on sensor data.
Translation: Understanding information from less accessible sensors.
Addressing over 30 real-world datasets across 20 domains, Totem aims to scale machine learning methods for these tasks.
Modern Machine Learning Ingredients
Building the underlying framework for Totem requires understanding several critical components:
Discrete Data Representation: This means handling data within a fixed space for effective transformation.
Transformer Architecture: Leveraging this prevalent architecture ensures efficient processing of sequences.
Generalist Training: The method focuses on jointly training across various data domains.