Non-deterministic Behavior of Ranking-based Metrics when Evaluating Embeddings
This addresses a reliability and security issue in machine learning evaluations, particularly for embedding-based systems, but is incremental as it builds on known quantization effects.
The paper tackles the problem of ambiguous performance metrics in embedding evaluations due to quantized distances, showing that this ambiguity can have measurable effects in state-of-the-art systems and be exploited for unfair gains. It provides bounds on the effect and suggests a simple solution to make ranking-based metrics deterministic and secure.
Embedding data into vector spaces is a very popular strategy of pattern recognition methods. When distances between embeddings are quantized, performance metrics become ambiguous. In this paper, we present an analysis of the ambiguity quantized distances introduce and provide bounds on the effect. We demonstrate that it can have a measurable effect in empirical data in state-of-the-art systems. We also approach the phenomenon from a computer security perspective and demonstrate how someone being evaluated by a third party can exploit this ambiguity and greatly outperform a random predictor without even access to the input data. We also suggest a simple solution making the performance metrics, which rely on ranking, totally deterministic and impervious to such exploits.