We present Felix --- a flexible text-editing approach for generation, designed to derive the maximum benefit from the ideas of decoding with bi-directional contexts and self-supervised pre-training. In contrast to conventional sequence-to-sequence (seq2seq) models, Felix is efficient in low-resource settings and fast at inference time, while being capable of modeling flexible input-output transformations. We achieve this by decomposing the text-editing task into two sub-tasks: tagging to decide on the subset of input tokens and their order in the output text and insertion to in-fill the missing tokens in the output not present in the input. The tagging model employs a novel Pointer mechanism, while the insertion model is based on a Masked Language Model. Both of these models are chosen to be non-autoregressive to guarantee faster inference. Felix performs favourably when compared to recent text-editing methods and strong seq2seq baselines when evaluated on four NLG tasks: Sentence Fusion, Machine Translation Automatic Post-Editing, Summarization, and Text Simplification.
Masked language modeling (MLM) pre-training methods such as BERT corrupt the input by replacing some tokens with [MASK] and then train a model to reconstruct the original tokens. While they produce good results when transferred to downstream NLP tasks, they generally require large amounts of compute to be effective. As an alternative, we propose a more sample-efficient pre-training task called replaced token detection. Instead of masking the input, our approach corrupts it by replacing some tokens with plausible alternatives sampled from a small generator network. Then, instead of training a model that predicts the original identities of the corrupted tokens, we train a discriminative model that predicts whether each token in the corrupted input was replaced by a generator sample or not. Thorough experiments demonstrate this new pre-training task is more efficient than MLM because the task is defined over all input tokens rather than just the small subset that was masked out. As a result, the contextual representations learned by our approach substantially outperform the ones learned by BERT given the same model size, data, and compute. The gains are particularly strong for small models; for example, we train a model on one GPU for 4 days that outperforms GPT (trained using 30x more compute) on the GLUE natural language understanding benchmark. Our approach also works well at scale, where it performs comparably to RoBERTa and XLNet while using less than 1/4 of their compute and outperforms them when using the same amount of compute.
With growing amounts of available textual data, development of algorithms capable of automatic analysis, categorization and summarization of these data has become a necessity. In this research we present a novel algorithm for keyword identification, i.e., an extraction of one or multi-word phrases representing key aspects of a given document, called Transformer-based Neural Tagger for Keyword IDentification (TNT-KID). By adapting the transformer architecture for a specific task at hand and leveraging language model pretraining on a domain specific corpus, the model is capable of overcoming deficiencies of both supervised and unsupervised state-of-the-art approaches to keyword extraction by offering competitive and robust performance on a variety of different datasets while requiring only a fraction of manually labeled data required by the best performing systems. This study also offers thorough error analysis with valuable insights into the inner workings of the model and an ablation study measuring the influence of specific components of the keyword identification workflow on the overall performance.
Named entity recognition is a natural language processing task to recognize and extract spans of text associated with named entities and classify them in semantic Categories. Google BERT is a deep bidirectional language model, pre-trained on large corpora that can be fine-tuned to solve many NLP tasks such as question answering, named entity recognition, part of speech tagging and etc. In this paper, we use the pre-trained deep bidirectional network, BERT, to make a model for named entity recognition in Persian. We also compare the results of our model with the previous state of the art results achieved on Persian NER. Our evaluation metric is CONLL 2003 score in two levels of word and phrase. This model achieved second place in NSURL-2019 task 7 competition which associated with NER for the Persian language. our results in this competition are 83.5 and 88.4 f1 CONLL score respectively in phrase and word level evaluation.
Logs are extensively used during the development and maintenance of software systems. They collect runtime events and allow tracking of code execution, which enables a variety of critical tasks such as troubleshooting and fault detection. However, large-scale software systems generate massive volumes of semi-structured log records, posing a major challenge for automated analysis. Parsing semi-structured records with free-form text log messages into structured templates is the first and crucial step that enables further analysis. Existing approaches rely on log-specific heuristics or manual rule extraction. These are often specialized in parsing certain log types, and thus, limit performance scores and generalization. We propose a novel parsing technique called NuLog that utilizes a self-supervised learning model and formulates the parsing task as masked language modeling (MLM). In the process of parsing, the model extracts summarizations from the logs in the form of a vector embedding. This allows the coupling of the MLM as pre-training with a downstream anomaly detection task. We evaluate the parsing performance of NuLog on 10 real-world log datasets and compare the results with 12 parsing techniques. The results show that NuLog outperforms existing methods in parsing accuracy with an average of 99% and achieves the lowest edit distance to the ground truth templates. Additionally, two case studies are conducted to demonstrate the ability of the approach for log-based anomaly detection in both supervised and unsupervised scenario. The results show that NuLog can be successfully used to support troubleshooting tasks. The implementation is available at https://github.com/nulog/nulog.
We present Geo2DR (Geometric to Distributed Representations), a GPU ready Python library for unsupervised learning on graph-structured data using discrete substructure patterns and neural language models. It contains efficient implementations of popular graph decomposition algorithms and neural language models in PyTorch which can be combined to learn representations of graphs using the distributive hypothesis. Furthermore, Geo2DR comes with general data processing and loading methods to bring substantial speed-up in the training of the neural language models. Through this we provide a modular set of tools and methods to quickly construct systems capable of learning distributed representations of graphs. This is useful for replication of existing methods, modification, or development of completely new methods. This paper serves to present the Geo2DR library and perform a comprehensive comparative analysis of existing methods re-implemented using Geo2DR across widely used graph classification benchmarks. Geo2DR displays a high reproducibility of results in published methods and interoperability with other libraries useful for distributive language modelling.
Deep networks often suffer from vanishing or exploding gradients due to inefficient signal propagation, leading to long training times or convergence difficulties. Various architecture designs, sophisticated residual-style networks, and initialization schemes have been shown to improve deep signal propagation. Recently, Pennington et al. used free probability theory to show that dynamical isometry plays an integral role in efficient deep learning. We show that the simplest architecture change of gating each residual connection using a single zero-initialized parameter satisfies initial dynamical isometry and outperforms more complex approaches. Although much simpler than its predecessors, this gate enables training thousands of fully connected layers with fast convergence and better test performance for ResNets trained on CIFAR-10. We apply this technique to language modeling and find that we can easily train 120-layer Transformers. When applied to 12 layer Transformers, it converges 56% faster on enwiki8.
Learning what to share between tasks has been a topic of great importance recently, as strategic sharing of knowledge has been shown to improve downstream task performance. This is particularly important for multilingual applications, as most languages in the world are under-resourced. Here, we consider the setting of training models on multiple different languages at the same time, when little or no data is available for languages other than English. We show that this challenging setup can be approached using meta-learning, where, in addition to training a source language model, another model learns to select which training instances are the most beneficial to the first. We experiment using standard supervised, zero-shot cross-lingual, as well as few-shot cross-lingual settings for different natural language understanding tasks (natural language inference, question answering). Our extensive experimental setup demonstrates the consistent effectiveness of meta-learning for a total of 15 languages. We improve upon the state-of-the-art for zero-shot and few-shot NLI (on MultiNLI and XNLI) and QA (on the MLQA dataset). A comprehensive error analysis indicates that the correlation of typological features between languages can partly explain when parameter sharing learned via meta-learning is beneficial.
Interests in the automatic generation of cooking recipes have been growing steadily over the past few years thanks to a large amount of online cooking recipes. We present RecipeGPT, a novel online recipe generation and evaluation system. The system provides two modes of text generations: (1) instruction generation from given recipe title and ingredients; and (2) ingredient generation from recipe title and cooking instructions. Its back-end text generation module comprises a generative pre-trained language model GPT-2 fine-tuned on a large cooking recipe dataset. Moreover, the recipe evaluation module allows the users to conveniently inspect the quality of the generated recipe contents and store the results for future reference. RecipeGPT can be accessed online at https://recipegpt.org/.
We introduce "talking-heads attention" - a variation on multi-head attention which includes linearprojections across the attention-heads dimension, immediately before and after the softmax operation.While inserting only a small number of additional parameters and a moderate amount of additionalcomputation, talking-heads attention leads to better perplexities on masked language modeling tasks, aswell as better quality when transfer-learning to language comprehension and question answering tasks.
Language model based pre-trained models such as BERT have provided significant gains across different NLP tasks. In this paper, we study different types of transformer based pre-trained models such as auto-regressive models (GPT-2), auto-encoder models (BERT), and seq2seq models (BART) for conditional data augmentation. We show that prepending the class labels to text sequences provides a simple yet effective way to condition the pre-trained models for data augmentation. Additionally, on three classification benchmarks, pre-trained Seq2Seq model outperforms other data augmentation methods in a low-resource setting. Further, we explore how different pre-trained model based data augmentation differs in-terms of data diversity, and how well such methods preserve the class-label information.
Manipulation tasks in daily life, such as pouring water, unfold intentionally under specialized manipulation contexts. Being able to process contextual knowledge in these Activities of Daily Living (ADLs) over time can help us understand manipulation intentions, which are essential for an intelligent robot to transition smoothly between various manipulation actions. In this paper, to model the intended concepts of manipulation, we present a vision dataset under a strictly constrained knowledge domain for both robot and human manipulations, where manipulation concepts and relations are stored by an ontology system in a taxonomic manner. Furthermore, we propose a scheme to generate a combination of visual attentions and an evolving knowledge graph filled with commonsense knowledge. Our scheme works with real-world camera streams and fuses an attention-based Vision-Language model with the ontology system. The experimental results demonstrate that the proposed scheme can successfully represent the evolution of an intended object manipulation procedure for both robots and humans. The proposed scheme allows the robot to mimic human-like intentional behaviors by watching real-time videos. We aim to develop this scheme further for real-world robot intelligence in Human-Robot Interaction.
In this paper, we introduce the Chinese corpus from CLUE organization, CLUECorpus2020, a large-scale corpus that can be used directly for self-supervised learning such as pre-training of a language model, or language generation. It has 100G raw corpus with 35 billion Chinese characters, which is retrieved from Common Crawl. To better understand this corpus, we conduct language understanding experiments on both small and large scale, and results show that the models trained on this corpus can achieve excellent performance on Chinese. We release a new Chinese vocabulary with a size of 8K, which is only one-third of the vocabulary size used in Chinese Bert released by Google. It saves computational cost and memory while works as good as original vocabulary. We also release both large and tiny versions of the pre-trained model on this corpus. The former achieves the state-of-the-art result, and the latter retains most precision while accelerating training and prediction speed for eight times compared to Bert-base. To facilitate future work on self-supervised learning on Chinese, we release our dataset, new vocabulary, codes, and pre-trained models on Github.
Tensor networks are a powerful modeling framework developed for computational many-body physics, which have only recently been applied within machine learning. In this work we utilize a uniform matrix product state (u-MPS) model for probabilistic modeling of sequence data. We first show that u-MPS enable sequence-level parallelism, with length-n sequences able to be evaluated in depth O(log n). We then introduce a novel generative algorithm giving trained u-MPS the ability to efficiently sample from a wide variety of conditional distributions, each one defined by a regular expression. Special cases of this algorithm correspond to autoregressive and fill-in-the-blank sampling, but more complex regular expressions permit the generation of richly structured data in a manner that has no direct analogue in neural generative models. Experiments on sequence modeling with synthetic and real text data show u-MPS outperforming a variety of baselines and effectively generalizing their predictions in the presence of limited data.
Molecule generation is a challenging open problem in cheminformatics. Currently, deep generative approaches addressing the challenge belong to two broad categories, differing in how molecules are represented. One approach encodes molecular graphs as strings of text, and learns their corresponding character-based language model. Another, more expressive, approach operates directly on the molecular graph. In this work, we address two limitations of the former: generation of invalid and duplicate molecules. To improve validity rates, we develop a language model for small molecular substructures called fragments, loosely inspired by the well-known paradigm of Fragment-Based Drug Design. In other words, we generate molecules fragment by fragment, instead of atom by atom. To improve uniqueness rates, we present a frequency-based masking strategy that helps generate molecules with infrequent fragments. We show experimentally that our model largely outperforms other language model-based competitors, reaching state-of-the-art performances typical of graph-based approaches. Moreover, generated molecules display molecular properties similar to those in the training sample, even in absence of explicit task-specific supervision.
We propose to pre-train a unified language model for both autoencoding and partially autoregressive language modeling tasks using a novel training procedure, referred to as a pseudo-masked language model (PMLM). Given an input text with masked tokens, we rely on conventional masks to learn inter-relations between corrupted tokens and context via autoencoding, and pseudo masks to learn intra-relations between masked spans via partially autoregressive modeling. With well-designed position embeddings and self-attention masks, the context encodings are reused to avoid redundant computation. Moreover, conventional masks used for autoencoding provide global masking information, so that all the position embeddings are accessible in partially autoregressive language modeling. In addition, the two tasks pre-train a unified language model as a bidirectional encoder and a sequence-to-sequence decoder, respectively. Our experiments show that the unified language models pre-trained using PMLM achieve new state-of-the-art results on a wide range of natural language understanding and generation tasks across several widely used benchmarks.
Early rumor detection (ERD) on social media platform is very challenging when limited, incomplete and noisy information is available. Most of the existing methods have largely worked on event-level detection that requires the collection of posts relevant to a specific event and relied only on user-generated content. They are not appropriate to detect rumor sources in the very early stages, before an event unfolds and becomes widespread. In this paper, we address the task of ERD at the message level. We present a novel hybrid neural network architecture, which combines a task-specific character-based bidirectional language model and stacked Long Short-Term Memory (LSTM) networks to represent textual contents and social-temporal contexts of input source tweets, for modelling propagation patterns of rumors in the early stages of their development. We apply multi-layered attention models to jointly learn attentive context embeddings over multiple context inputs. Our experiments employ a stringent leave-one-out cross-validation (LOO-CV) evaluation setup on seven publicly available real-life rumor event data sets. Our models achieve state-of-the-art(SoA) performance for detecting unseen rumors on large augmented data which covers more than 12 events and 2,967 rumors. An ablation study is conducted to understand the relative contribution of each component of our proposed model.
This paper deals with the two fundamental problems concerning the handling of large n-gram language models: indexing, that is compressing the n-gram strings and associated satellite data without compromising their retrieval speed; and estimation, that is computing the probability distribution of the strings from a large textual source. Regarding the problem of indexing, we describe compressed, exact and lossless data structures that achieve, at the same time, high space reductions and no time degradation with respect to state-of-the-art solutions and related software packages. In particular, we present a compressed trie data structure in which each word following a context of fixed length k, i.e., its preceding k words, is encoded as an integer whose value is proportional to the number of words that follow such context. Since the number of words following a given context is typically very small in natural languages, we lower the space of representation to compression levels that were never achieved before. Despite the significant savings in space, our technique introduces a negligible penalty at query time. Regarding the problem of estimation, we present a novel algorithm for estimating modified Kneser-Ney language models, that have emerged as the de-facto choice for language modeling in both academia and industry, thanks to their relatively low perplexity performance. Estimating such models from large textual sources poses the challenge of devising algorithms that make a parsimonious use of the disk. The state-of-the-art algorithm uses three sorting steps in external memory: we show an improved construction that requires only one sorting step thanks to exploiting the properties of the extracted n-gram strings. With an extensive experimental analysis performed on billions of n-grams, we show an average improvement of 4.5X on the total running time of the state-of-the-art approach.
Optimization in machine learning, both theoretical and applied, is presently dominated by first-order gradient methods such as stochastic gradient descent. Second-order optimization methods, that involve second derivatives and/or second order statistics of the data, are far less prevalent despite strong theoretical properties, due to their prohibitive computation, memory and communication costs. In an attempt to bridge this gap between theoretical and practical optimization, we present a scalable implementation of a second-order preconditioned method (concretely, a variant of full-matrix Adagrad), that along with several critical algorithmic and numerical improvements, provides significant convergence and wall-clock time improvements compared to conventional first-order methods on state-of-the-art deep models. Our novel design effectively utilizes the prevalent heterogeneous hardware architecture for training deep models, consisting of a multicore CPU coupled with multiple accelerator units. We demonstrate superior performance compared to state-of-the-art on very large learning tasks such as machine translation with Transformers, language modeling with BERT, click-through rate prediction on Criteo, and image classification on ImageNet with ResNet-50.
Transformers have been successfully applied to sequential, auto-regressive tasks despite being feedforward networks. Unlike recurrent neural networks, Transformers use attention to capture temporal relations while processing input tokens in parallel. While this parallelization makes them computationally efficient, it restricts the model from fully exploiting the sequential nature of the input. The representation at a given layer can only access representations from lower layers, rather than the higher level representations already available. In this work, we propose the Feedback Transformer architecture that exposes all previous representations to all future representations, meaning the lowest representation of the current timestep is formed from the highest-level abstract representation of the past. We demonstrate on a variety of benchmarks in language modeling, machine translation, and reinforcement learning that the increased representation capacity can create small, shallow models with much stronger performance than comparable Transformers.