Product Quantization
A vector quantization method that splits high-dimensional vectors into low-dimensional subvectors and independently trains a small codebook for each subspace. It is used for vector compression in large-scale image retrieval and can express exponentially many combinations within limited codebook storage.
Timeline (last 90 days)
ProTok uses product quantization to assign an independent codebook group to each modality; during training, after randomly dropping the input of a certain modality, the shared codebooks must still produce meaningful discrete tokens, so that a single checkpoint can support flexible inference over any subset of modalities without retraining.
Product quantization was first proposed by Jégou et al. in 2011 for vector compression in large-scale image retrieval.
All Facts (2)
ProTok uses product quantization to assign an independent codebook group to each modality; during training, after randomly dropping the input of a certain modality, the shared codebooks must still produce meaningful discrete tokens, so that a single checkpoint can support flexible inference over any subset of modalities without retraining.
50%UnverifiedProduct quantization was first proposed by Jégou et al. in 2011 for vector compression in large-scale image retrieval.
50%