Timezone: »

 
Poster
Marginal Distribution Adaptation for Discrete Sets via Module-Oriented Divergence Minimization
Hanjun Dai · Mengjiao Yang · Yuan Xue · Dale Schuurmans · Bo Dai

Tue Jul 19 03:30 PM -- 05:30 PM (PDT) @ Hall E #333

Distributions over discrete sets capture the essential statistics including the high-order correlation among elements. They provide powerful insight for decision making in many application domains, e.g., product assortment based on the distributions of products in shopping carts. Although deep generative models can be trained to capture these distributions from pre-collected data, such pre-trained models are usually not able to align well with the target domain due to reasons such as temporal shift or the change in the population mix. We develop a general framework to adapt a generative model subject to a (possibly counterfactual) target data distribution with both sampling and computation efficiency. Concretely, instead of re-training a full model from scratch, we reuse the learned modules to preserve the correlations between set elements, while only adjusting corresponding components to align with target marginal constraints. We instantiate the approach for three commonly used forms of discrete set distribution---latent variable, autoregressive, and energy based models---and provide efficient solutions for marginal-constrained optimization in either primal or dual forms. Experiments on both synthetic and real-world e-commerce and EHR datasets show that the proposed framework is able to practically align a generative model to match marginal constraints under distribution shift.

Author Information

Hanjun Dai (Google Brain)
Mengjiao Yang (Google Brain)
Yuan Xue (Google)
Dale Schuurmans (Google / University of Alberta)
Bo Dai (Google Brain)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors