Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userConsistent Accelerated Inference via Confident Adaptive Transformers
137
0
0.0
(
0
)
Added by
Tal Schuster
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
We develop a novel approach for confidently accelerating inference in the large and expensive multilayer Transformers that are now ubiquitous in natural language processing (NLP). Amortized or approximate computational methods increase efficiency, but can come with unpredictable performance costs. In this work, we present CATs -- Confident Adaptive Transformers -- in which we simultaneously increase computational efficiency, while guaranteeing a specifiable degree of consistency with the original model with high confidence. Our method trains additional prediction heads on top of intermediate layers, and dynamically decides when to stop allocating computational effort to each input using a meta consistency classifier. To calibrate our early prediction stopping rule, we formulate a unique extension of conformal prediction. We demonstrate the effectiveness of this approach on four classification and regression tasks.
rate research
Read More
State-of-the-art models in natural language processing rely on separate rigid subword tokenization algorithms, which limit their generalization ability and adaptation to new settings. In this paper, we propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. To this end, we introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters in a data-driven fashion. Concretely, GBST enumerates candidate subword blocks and learns to score them in a position-wise fashion using a block scoring network. We additionally introduce Charformer, a deep Transformer model that integrates GBST and operates on the byte level. Via extensive experiments on English GLUE, multilingual, and noisy text datasets, we show that Charformer outperforms a series of competitive byte-level baselines while generally performing on par and sometimes outperforming subword-based models. Additionally, Charformer is fast, improving the speed of both vanilla byte-level and subword-level Transformers by 28%-100% while maintaining competitive quality. We believe this work paves the way for highly performant token-free models that are trained completely end-to-end.
Depth-adaptive neural networks can dynamically adjust depths according to the hardness of input words, and thus improve efficiency. The main challenge is how to measure such hardness and decide the required depths (i.e., layers) to conduct. Previous works generally build a halting unit to decide whether the computation should continue or stop at each layer. As there is no specific supervision of depth selection, the halting unit may be under-optimized and inaccurate, which results in suboptimal and unstable performance when modeling sentences. In this paper, we get rid of the halting unit and estimate the required depths in advance, which yields a faster depth-adaptive model. Specifically, two approaches are proposed to explicitly measure the hardness of input words and estimate corresponding adaptive depth, namely 1) mutual information (MI) based estimation and 2) reconstruction loss based estimation. We conduct experiments on the text classification task with 24 datasets in various sizes and domains. Results confirm that our approaches can speed up the vanilla Transformer (up to 7x) while preserving high accuracy. Moreover, efficiency and robustness are significantly improved when compared with other depth-adaptive approaches.
Low-resource language translation is a challenging but socially valuable NLP task. Building on recent work adapting the Transformers normalization to this setting, we propose QKNorm, a normalization technique that modifies the attention mechanism to make the softmax function less prone to arbitrary saturation without sacrificing expressivity. Specifically, we apply $ell_2$ normalization along the head dimension of each query and key matrix prior to multiplying them and then scale up by a learnable parameter instead of dividing by the square root of the embedding dimension. We show improvements averaging 0.928 BLEU over state-of-the-art bilingual benchmarks for 5 low-resource translation pairs from the TED Talks corpus and IWSLT15.
Translation-based embedding models have gained significant attention in link prediction tasks for knowledge graphs. TransE is the primary model among translation-based embeddings and is well-known for its low complexity and high efficiency. Therefore, most of the earlier works have modified the score function of the TransE approach in order to improve the performance of link prediction tasks. Nevertheless, proven theoretically and experimentally, the performance of TransE strongly depends on the loss function. Margin Ranking Loss (MRL) has been one of the earlier loss functions which is widely used for training TransE. However, the scores of positive triples are not necessarily enforced to be sufficiently small to fulfill the translation from head to tail by using relation vector (original assumption of TransE). To tackle this problem, several loss functions have been proposed recently by adding upper bounds and lower bounds to the scores of positive and negative samples. Although highly effective, previously developed models suffer from an expansion in search space for a selection of the hyperparameters (in particular the upper and lower bounds of scores) on which the performance of the translation-based models is highly dependent. In this paper, we propose a new loss function dubbed Adaptive Margin Loss (AML) for training translation-based embedding models. The formulation of the proposed loss function enables an adaptive and automated adjustment of the margin during the learning process. Therefore, instead of obtaining two values (upper bound and lower bound), only the center of a margin needs to be determined. During learning, the margin is expanded automatically until it converges. In our experiments on a set of standard benchmark datasets including Freebase and WordNet, the effectiveness of AML is confirmed for training TransE on link prediction tasks.
The ability to perform arithmetic tasks is a remarkable trait of human intelligence and might form a critical component of more complex reasoning tasks. In this work, we investigate if the surface form of a number has any influence on how sequence-to-sequence language models learn simple arithmetic tasks such as addition and subtraction across a wide range of values. We find that how a number is represented in its surface form has a strong influence on the models accuracy. In particular, the model fails to learn addition of five-digit numbers when using subwords (e.g., 32), and it struggles to learn with character-level representations (e.g., 3 2). By introducing position tokens (e.g., 3 10e1 2), the model learns to accurately add and subtract numbers up to 60 digits. We conclude that modern pretrained language models can easily learn arithmetic from very few examples, as long as we use the proper surface representation. This result bolsters evidence that subword tokenizers and positional encodings are components in current transformer designs that might need improvement. Moreover, we show that regardless of the number of parameters and training examples, models cannot learn addition rules that are independent of the length of the numbers seen during training. Code to reproduce our experiments is available at https://github.com/castorini/transformers-arithmetic
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
