Global Autoregressive Models for Data-Efficient Sequence Learning

By Tetiana Parshakova, Jean-Marc Andreoli, Marc Dymetman. 2019

23 citations recorded by the source corpus.

Standard autoregressive seq2seq models are easily trained by max-likelihood, but tend to show poor results under small-data conditions. We introduce a class of seq2seq models, GAMs (Global Autoregressive Models), which combine an autoregressive component with a log-linear component, allowing the use of global a priori features to compensate for lack of data. We train these models in two steps. In the first step, we obtain an unnormalized GAM that maximizes the likelihood of the data, but is improper for fast inference or evaluation. In the second step, we use this GAM to train (by distillation) a second autoregressive model that approximates the normalized distribution associated with the GAM, and can be used for fast inference and evaluation. Our experiments focus on language modelling under synthetic conditions and show a strong perplexity reduction of using the second autoregressive model over the standard one.

Subfields: Artificial Intelligence, Model Architectures

How researchers see this paper

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

In Artificial Intelligence

In Model Architectures

Citations in the corpus

Residual Energy-Based Models for Text Generation (2020, methodology): “We want to note that the same formulation has been proposed in (Wang & Ou, 2018b; Parshakova et al., 2019).”
On Reinforcement Learning and Distribution Matching for Fine-Tuning Language Models with no Catastrophic Forgetting (2022, methodology): “While the connections between these two seemingly distinct paradigms have been noted (Parshakova et al., 2019b; Korbak et al., 2022b), they have not been explored in detail.”
Distributional Reinforcement Learning for Energy-Based Sequential Models (2019, methodology): “This is the approach to Training-2 taken in [15], using rejection sampling at training time for obtaining the samples, and then training θ on these samples to obtain a ﬁnal AM π θ which can be used for eﬃcient sampling at test time and for evaluation.”
Residual Energy-Based Models for Text (2021, methodology): “In this work, we leverage generations produced by pretrained language models as negative samples (Wang and Ou, 2018b; Parshakova et al., 2019).”
Energy-Based Models for Code Generation under Compilability Constraints (2021, methodology): “Parshakova et al. (2019b) and Deng et al. (2020) augment a autoregressive language models with an additional global factor to obtain a lower perplexity on the training data.”
COLD Decoding: Energy-based Constrained Text Generation with Langevin Dynamics (2022, unknown): “Recent work has thus made attempts to train text-based EBMs each for specific tasks (Deng et al., 2020; Khalifa et al., 2020; Parshakova et al., 2019; Hu et al., 2018).”
An approximate sampler for energy-based models with divergence diagnostics (2022, methodology): “The vector λ is found using self-normalized importance sampling (Owen, 2013; Parshakova et al., 2019a) and stochastic optimization.”
Energy-Based Reranking: Improving Neural Machine Translation Using Energy-Based Models (2020, background): “Rosenfeld et al.; Parshakova et al. (2001; 2019) deﬁne q ( y ) as autoregressive model and ω θ ( y ) us-ing a log-linear model: ω θ ( y ) = exp( θ T φ ( y )) , where φ ( y ) is the vector of sufﬁcient statistics (features) evaluated at y .”
RankGen: Improving Text Generation with Large Ranking Models (2022, unknown): “R ANK G EN is also closely related to efforts in energy-based methods (LeCun et al., 2006) for generative modeling (Grover et al., 2019; Parshakova et al., 2019), speech recognition (Wang and Ou, 2018), open-ended text generation (Bakhtin et al., 2019; Deng et al., 2020), machine translation (Shen…”