<ul><li>Performative Drift is a special type of Concept Drift that occurs when a model's predictions influence the future instances the model will encounter.</li><li>The Generative Domain Adversarial Network (GDAN) is introduced to create drift-resistant classifiers by generating domain-invariant representations of incoming data and reversing the effects of performative drift.</li><li>Empirical evaluation of GDAN shows promising results, with limited performance degradation over several timesteps.</li><li>GDAN's generative network can also be used in combination with other models to mitigate performance degradation in the presence of performative drift.</li></ul>

Performative Drift Resistant Classification Using Generative Domain Adversarial Networks

Discover more