Embedding Models
Embedding models transform data into a condensed, fingerprint-like vector (list of numbers) that represents the meaning.
What are they?
A vector embedding model takes in data and outputs a semantic vector.
Embed(data) = [1.3433, 0.332, -8.423, ...]The length of the vector depends on the amount of dimensions in the embedding model. For example the face detection in iPhones uses 128 dimensional embedding vectors. Higher dimensions don't always mean higher accuracy.
Vector embedding models are something you would usually not train youself, rather you would use a pretrained model to do inference on your data. Training a text model would require for example a dataset of millions of webpages and the training model to "read" all of them to find and memorize patterns in the word useage.
Which usecases are there?
A few powerful examples
- Recommendations on a Webshop
- Text overlap and plagiarism checks
- Product comparison
- Facial recognition and matching
- Dating algorithms
- Symptom or patient data clustering
- File search and visualization
- Many more!
How can I use embeddings in aicuflow?
In aicuflow, embeddings are used in 2 ways:
- You can cluster and search your data and flows, which uses embeddings in the background.
- More actively, you can use our multimodal embedding models in your data flows.
Get started!