Structure Retention in Embedding Spaces as a Predictor of Benchmark Performance
For researchers and practitioners using embedding models, this provides a predictor of benchmark performance without running full evaluations, potentially guiding model selection and training.
The authors show that high-performing embedding models organize their embedding spaces consistently, with nearest-neighbor overlap and ICA magnitude differences correlating up to 0.97 with task performance across 25 models on MTEB tasks. This reveals varying linearity and local information retention across tasks.
In this paper, we show that high-performing embedding models organize their embedding spaces in a consistent way. We evaluate 25 contemporary embedding models on five MTEB tasks spanning four diverse task categories (retrieval, bitext mining, pair classification, and summarization) in both English and multilingual settings, and reveal that nearest-neighbor overlap and magnitude differences in independent component analysis (ICA) between paired text instances strongly correlate (even up to 0.97) with performance on the given task. Ultimately, we show that embedding tasks display varying degrees of linearity and reliance on retention of local information. Our results further the understanding of embeddings, their relation to model performance, and shed light on possible future training objectives and optimizing conditional embeddings.