Generative or Discriminative? Revisiting Text Classification in the Era of Transformers
This work addresses the problem of selecting optimal text classification approaches for practitioners, offering insights into performance trade-offs, but it is incremental as it extends classical analysis to new architectures without introducing a novel method.
The study revisits the comparison between generative and discriminative classifiers for text classification using modern transformer architectures, finding that the classical trade-offs in sample complexity and asymptotic error manifest distinctly across different models and training paradigms, with practical guidance provided for real-world constraints like latency and data limitations.
The comparison between discriminative and generative classifiers has intrigued researchers since Efron's seminal analysis of logistic regression versus discriminant analysis. While early theoretical work established that generative classifiers exhibit lower sample complexity but higher asymptotic error in simple linear settings, these trade-offs remain unexplored in the transformer era. We present the first comprehensive evaluation of modern generative and discriminative architectures - Auto-regressive modeling, Masked Language Modeling, Discrete Diffusion, and Encoders for text classification. Our study reveals that the classical 'two regimes' phenomenon manifests distinctly across different architectures and training paradigms. Beyond accuracy, we analyze sample efficiency, calibration, noise robustness, and ordinality across diverse scenarios. Our findings offer practical guidance for selecting the most suitable modeling approach based on real-world constraints such as latency and data limitations.