Multiple Choice Learning: Learning to Produce Multiple Structured Outputs

By Abner Guzmán-Rivera, Dhruv Batra, Pushmeet Kohli. 2012

224 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Statistical Learning Theory

Citations in the corpus

Segment Anything (2023, methodology): “This is a common technique used for models with multiple outputs [15, 45, 64].”
Evaluating the State-of-the-Art of End-to-End Natural Language Generation: The E2E NLG Challenge (2019, background): “HARV takes yet another approach in order to both stabilise training and increase diversity, called diverse ensembling (Guzman-Rivera, Batra, and Kohli 2012).”
Mixture Models for Diverse Machine Translation: Tricks of the Trade (2019, background): “Among these variants, the hMup objective is also known as multiple choice learning (MCL) for an ensemble of learners where the oracle loss is minimized (Guzman-Rivera et al., 2012), and He et al. (2018) has considered the sMup objective to train a sequence-to-sequence mixture model.”
End-to-End Content and Plan Selection for Data-to-Text Generation (2018, methodology): “:;f Kbe the Kmodels in the mixture. These models can either be completely disjoint or share a subset of their parameters (e.g. the word embeddings, the encoder, or both encoder and decoder). FollowingGuzman-Rivera et al. (2012), we introduce an unobserved random variable w˘Cat(1=K) that assigns a we”
Towards Oracle Knowledge Distillation with Neural Architecture Search (2019, methodology): “Model ensemble (Breiman 1996; Freund, Schapire, and Abe 1999; Guzman-Rivera, Batra, and Kohli 2012; Lee et al. 2017; 2016; 2015; Mun et al. 2018) is a useful technique to boost performance using multiple models trained ∗Both authors contributed equally Copyright c© 2020, Association for the…”
Geometric Image Synthesis (2018, methodology): “e innitely plausible outputs due to dierent possible lighting conditions, object colors etc. Thus, we task our network to produce Kdiverse outputs from the last layer using multiple-choice learning [40,2]. More specically, we compute the loss for each of these outputs, but only back propagate the gr”
MultiSeg: Semantically Meaningful, Scale-Diverse Segmentations From Minimal User Input (2019, methodology): “Diverse Predictions: While there is a large body of work on making diverse predictions [3, 9, 13, 19, 20], we briefly discuss works that produce a diverse set of solutions in interactive image segmentation.”
A Probabilistic U-Net for Segmentation of Ambiguous Images (2018, methodology): “In order to overcome this, several approaches train models jointly using the oracle set loss [10], i.”
Text-to-image Synthesis via Symmetrical Distillation Networks (2018, unknown): “s function for diverse synthesis because directly using above loss functions on this modified architecture will make the diverse outputs exactly the same. Therefore, we adopt multiple choice learning [12] to modify the distillation loss in two stages as follows: LossDiv I = Õ l ||ϕD l (Ir)−ϕ G +1 (φ”
Residual Mixture of Experts (2022, background): “In the meanwhile, instead of designing experts in the architecture, there are similar research ideas to generate a set of outputs and pick the correct one, such as multi-choice learning [20, 33, 34].”
Mixture Content Selection for Diverse Sequence Generation (2019, background): “Among these, the closest to ours is the mixture decoder (Shen et al., 2019) that also adopts hard-EM for training, where a minimum-loss predictor is assigned to each data point, which is also known as multiple choice learning (Guzman-Rivera et al., 2012; Lee et al., 2016).”
CKD: Cross-Task Knowledge Distillation for Text-to-Image Synthesis (2020, methodology): “Therefore, we employ multiple choice learning [47] to change the loss function in three stages as follows: Loss DivS 3 = min where i ps ( ϕ t ; θ ) denotes p -th synthetic image output by the ﬁnal layer of T2IS model.”
Embracing Uncertainty: Decoupling and De-bias for Robust Temporal Grounding (2021, methodology): “To tackle the dilemma between multiple predictions and single-style annotations, we introduce a min-loss from MCL [10] to learn diverse predictions without extra annotations.”
Diverse Image Synthesis From Semantic Layouts via Conditional IMLE (2018, methodology): “For example, [13] proposed a model that outputs a fixed number of different predictions simultaneously, which was an approach adopted by Chen and Koltun [3] to generate different images for the same semantic layout.”
Confident Multiple Choice Learning (2017, methodology): “In this section, we describe the basic concept of multiple choice learning (MCL) (Guzman-Rivera et al., 2014; 2012).”
Stochastic Multiple Choice Learning for Training Diverse Deep Ensembles (2016, background): “To summarize, we propose Stochastic Multiple Choice Learning (sMCL), an SGD-based technique for training diverse deep ensembles that follows a ‘winner-take-gradient’ training strategy.”
Gesture Recognition Portfolios for Personalization (2014, methodology): “More formally, where the subject-speciﬁc likelihood L s (Θ) is deﬁned as Such a formulation is related to the Multiple Choice Learning framework [9], which used a coordinate descent procedure to minimize the set loss.”
Identifying Ambiguous Similarity Conditions via Semantic Matching (2022, methodology): “Following the multi-choice learning [2], it learns disentangled embeddings ΨK with hard condition assignments.”