Matryoshka Embeddings with Aditya Kusupati, Zach Nussbaum, and Zain Hasan - Weaviate Podcast #89!
Weaviate Podcast
Optimizing Representation Learning with Binary Embeddings
During training of the representation learning pipeline, converting the D dimensional vectors into binary values ensures that the embeddings align with the sign of the hash function. This alignment allows for an order one lookup during inference, enabling efficient search for approximate nearest neighbors. By nesting bits within a binary code, such as ensuring the first four bits are within the eight bits, which are then within the 16 bits, a more structured and efficient embedding space can be created.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
