Disentangled and Distilled Encoder for Out-of-Distribution Reasoning with Rademacher Guarantees
This work addresses the deployment challenge for multi-label OOD reasoning in resource-limited settings, representing an incremental improvement over existing disentangled VAE methods.
The paper tackles the problem of deploying out-of-distribution reasoning models on resource-constrained devices by proposing a disentangled distilled encoder framework that compresses models while preserving disentanglement, achieving a 50% reduction in model size with minimal performance loss.
Recently, the disentangled latent space of a variational autoencoder (VAE) has been used to reason about multi-label out-of-distribution (OOD) test samples that are derived from different distributions than training samples. Disentangled latent space means having one-to-many maps between latent dimensions and generative factors or important characteristics of an image. This paper proposes a disentangled distilled encoder (DDE) framework to decrease the OOD reasoner size for deployment on resource-constrained devices while preserving disentanglement. DDE formalizes student-teacher distillation for model compression as a constrained optimization problem while preserving disentanglement with disentanglement constraints. Theoretical guarantees for disentanglement during distillation based on Rademacher complexity are established. The approach is evaluated empirically by deploying the compressed model on an NVIDIA