An official website of the United States government
Abstract We develop a method of classified mixed model prediction based on generalized linear mixed models that incorporate pseudo‐prior information to improve prediction accuracy. We establish consistency of the proposed method both in terms of prediction of the true mixed effect of interest and in terms of correctly identifying the potential class corresponding to the new observations if such a class matching one of the training data classes exists. Empirical results, including simulation studies and real‐data validation, fully support the theoretical findings.
more »
« less
Pseudo-Bayesian Classified Mixed Model Prediction
We propose a new classified mixed model prediction (CMMP) procedure, called pseudo-Bayesian CMMP,that uses network information in matching the group index between the training data and new data, whosecharacteristics of interest one wishes to predict. The current CMMP procedures do not incorporate suchinformation; as a result, the methods are not consistent in terms of matching the group index. Although, asthe number of training data groups increases, the current CMMP method can predict the mixed effects ofinterest consistently, its accuracy is not guaranteed when the number of groups is moderate, as is the case inmany potential applications. The proposed pseudo-Bayesian CMMP procedure assumes a flexible workingprobability model for the group index of the new observation to match the index of a training data group,which may be viewed as a pseudo prior. We show that, given any working model satisfying mild conditions,the pseudo-Bayesian CMMP procedure is consistent and asymptotically optimal both in terms of matchingthe group index and in terms of predicting the mixed effect of interest associated with the new observations.The theoretical results are fully supported by results of empirical studies, including Monte-Carlo simulationsand real-data validation.
more »
« less
- Award ID(s):
- 1914465
- PAR ID:
- 10478957
- Publisher / Repository:
- Taylor & Francis
- Date Published:
- Journal Name:
- Journal of the American Statistical Association
- Volume:
- 118
- Issue:
- 543
- ISSN:
- 0162-1459
- Page Range / eLocation ID:
- 1747 to 1759
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Machine learning models often perform poorly under subpopulation shifts in the data distribution. Developing methods that allow machine learning models to better generalize to such shifts is crucial for safe deployment in real-world settings. In this paper, we develop a family of group-aware prior (GAP) distributions over neural network parameters that explicitly favor models that generalize well under subpopulation shifts. We design a simple group-aware prior that only requires access to a small set of data with group information and demonstrate that training with this prior yields state-of-the-art performance -- even when only retraining the final layer of a previously trained non-robust model. Group aware-priors are conceptually simple, complementary to existing approaches, such as attribute pseudo labeling and data reweighting, and open up promising new avenues for harnessing Bayesian inference to enable robustness to subpopulation shifts.more » « less
-
Machine learning models often perform poorly under subpopulation shifts in the data distribution. Developing methods that allow machine learning models to better generalize to such shifts is crucial for safe deployment in real-world settings. In this paper, we develop a family of group-aware prior (GAP) distributions over neural network parameters that explicitly favor models that generalize well under subpopulation shifts. We design a simple group-aware prior that only requires access to a small set of data with group information and demonstrate that training with this prior yields state-of-the-art performance -- even when only retraining the final layer of a previously trained non-robust model. Group aware-priors are conceptually simple, complementary to existing approaches, such as attribute pseudo labeling and data reweighting, and open up promising new avenues for harnessing Bayesian inference to enable robustness to subpopulation shifts.more » « less
-
null (Ed.)We consider a novel application of inverse reinforcement learning with behavioral economics constraints to model, learn and predict the commenting behavior of YouTube viewers. Each group of users is modeled as a rationally inattentive Bayesian agent which solves a contextual bandit problem. Our methodology integrates three key components. First, to identify distinct commenting patterns, we use deep embedded clustering to estimate framing information (essential extrinsic features) that clusters users into distinct groups. Second, we present an inverse reinforcement learning algorithm that uses Bayesian revealed preferences to test for rationality: does there exist a utility function that rationalizes the given data, and if yes, can it be used to predict commenting behavior? Finally, we impose behavioral economics constraints stemming from rational inattention to characterize the attention span of groups of users. The test imposes a Rényi mutual information cost constraint which impacts how the agent can select attention strategies to maximize their expected utility. After a careful analysis of a massive YouTube dataset, our surprising result is that in most YouTube user groups, the commenting behavior is consistent with optimizing a Bayesian utility with rationally inattentive constraints. The paper also highlights how the rational inattention model can accurately predict commenting behavior. The massive YouTube dataset and analysis used in this paper are available on GitHub and completely reproduciblemore » « less
-
Despite continued interest in mixed-species groups, we still lack a unified understanding of how ecological and social processes work across scales to influence group formation. Recent work has revealed ecological correlates of mixed-species group formation, but the mechanisms by which concomitant social dynamics produce these patterns, if at all, is unknown. Here, we use camera trap data for six mammalian grazer species in Serengeti National Park. Building on previous work, we found that ecological variables, and especially forage quality, influenced the chances of species overlap over small spatio-temporal scales (i.e. on the scales of several metres and hours). Migratory species (gazelle, wildebeest and zebra) were more likely to have heterospecific partners available in sites with higher forage quality, but the opposite was true for resident species (buffalo, hartebeest and topi). These findings illuminate the circumstances under which mixed-species group formation is even possible. Next, we found that greater heterospecific availability was associated with an increased probability of mixed-species group formation in gazelle, hartebeest, wildebeest and zebra, but ecological variables did not further shape these patterns. Overall, our results are consistent with a model whereby ecological and social drivers of group formation are species-specific and operate on different spatio-temporal scales. This article is part of the theme issue ‘Mixed-species groups and aggregations: shaping ecological and behavioural patterns and processes’.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1080/01621459.2021.2008944
