Masked language modeling (MLM) has become one of the most successful self-supervised pre-training task. Inspired by its success, Point-BERT, as a pioneer work in point cloud, proposed masked point modeling (MPM) to pre-train point transformer on large scale unanotated dataset. Despite its great performance, we find the inherent difference between language and point cloud tends to cause ambiguous tokenization for point cloud. For point cloud, there doesn't exist a gold standard for point cloud tokenization. Point-BERT use a discrete Variational AutoEncoder (dVAE) as tokenizer, but it might generate different token ids for semantically-similar patches and generate the same token ids for semantically-dissimilar patches. To tackle above problem, we propose our McP-BERT, a pre-training framework with multi-choice tokens. Specifically, we ease the previous single-choice constraint on patch token ids in Point-BERT, and provide multi-choice token ids for each patch as supervision. Moreover, we utilitze the high-level semantics learned by transformer to further refine our supervision signals. Extensive experiments on point cloud classification, few-shot classification and part segmentation tasks demonstrate the superiority of our method, e.g., the pre-trained transformer achieves 94.1% accuracy on ModelNet40, 84.28% accuracy on the hardest setting of ScanObjectNN and new state-of-the-art performance on few-shot learning. We also demonstrate that our method not only improves the performance of Point-BERT on all downstream tasks, but also incurs almost no extra computational overhead. The code will be released in https://github.com/fukexue/McP-BERT.
Sentence embeddings are commonly used in text clustering and semantic retrieval tasks. State-of-the-art sentence representation methods are based on artificial neural networks fine-tuned on large collections of manually labeled sentence pairs. Sufficient amount of annotated data is available for high-resource languages such as English or Chinese. In less popular languages, multilingual models have to be used, which offer lower performance. In this publication, we address this problem by proposing a method for training effective language-specific sentence encoders without manually labeled data. Our approach is to automatically construct a dataset of paraphrase pairs from sentence-aligned bilingual text corpora. We then use the collected data to fine-tune a Transformer language model with an additional recurrent pooling layer. Our sentence encoder can be trained in less than a day on a single graphics card, achieving high performance on a diverse set of sentence-level tasks. We evaluate our method on eight linguistic tasks in Polish, comparing it with the best available multilingual sentence encoders.
Visual Entailment with natural language explanations aims to infer the relationship between a text-image pair and generate a sentence to explain the decision-making process. Previous methods rely mainly on a pre-trained vision-language model to perform the relation inference and a language model to generate the corresponding explanation. However, the pre-trained vision-language models mainly build token-level alignment between text and image yet ignore the high-level semantic alignment between the phrases (chunks) and visual contents, which is critical for vision-language reasoning. Moreover, the explanation generator based only on the encoded joint representation does not explicitly consider the critical decision-making points of relation inference. Thus the generated explanations are less faithful to visual-language reasoning. To mitigate these problems, we propose a unified Chunk-aware Alignment and Lexical Constraint based method, dubbed as CALeC. It contains a Chunk-aware Semantic Interactor (arr. CSI), a relation inferrer, and a Lexical Constraint-aware Generator (arr. LeCG). Specifically, CSI exploits the sentence structure inherent in language and various image regions to build chunk-aware semantic alignment. Relation inferrer uses an attention-based reasoning network to incorporate the token-level and chunk-level vision-language representations. LeCG utilizes lexical constraints to expressly incorporate the words or chunks focused by the relation inferrer into explanation generation, improving the faithfulness and informativeness of the explanations. We conduct extensive experiments on three datasets, and experimental results indicate that CALeC significantly outperforms other competitor models on inference accuracy and quality of generated explanations.
Recently Convolution-augmented Transformer (Conformer) has shown promising results in Automatic Speech Recognition (ASR), outperforming the previous best published Transformer Transducer. In this work, we believe that the output information of each block in the encoder and decoder is not completely inclusive, in other words, their output information may be complementary. We study how to take advantage of the complementary information of each block in a parameter-efficient way, and it is expected that this may lead to more robust performance. Therefore we propose the Block-augmented Transformer for speech recognition, named Blockformer. We have implemented two block ensemble methods: the base Weighted Sum of the Blocks Output (Base-WSBO), and the Squeeze-and-Excitation module to Weighted Sum of the Blocks Output (SE-WSBO). Experiments have proved that the Blockformer significantly outperforms the state-of-the-art Conformer-based models on AISHELL-1, our model achieves a CER of 4.29\% without using a language model and 4.05\% with an external language model on the testset.
We introduce a zero-shot video captioning method that employs two frozen networks: the GPT-2 language model and the CLIP image-text matching model. The matching score is used to steer the language model toward generating a sentence that has a high average matching score to a subset of the video frames. Unlike zero-shot image captioning methods, our work considers the entire sentence at once. This is achieved by optimizing, during the generation process, part of the prompt from scratch, by modifying the representation of all other tokens in the prompt, and by repeating the process iteratively, gradually improving the specificity and comprehensiveness of the generated sentence. Our experiments show that the generated captions are coherent and display a broad range of real-world knowledge. Our code is available at: https://github.com/YoadTew/zero-shot-video-to-text
Recent breakthroughs in Natural Language Processing (NLP) have been driven by language models trained on a massive amount of plain text. While powerful, deriving supervision from textual resources is still an open question. For example, language model pretraining often neglects the rich, freely-available structures in textual data. In this thesis, we describe three lines of work that seek to improve the training and evaluation of neural models using naturally-occurring supervision. We first investigate self-supervised training losses to help enhance the performance of pretrained language models for various NLP tasks. Specifically, we alter the sentence prediction loss to make it better suited to other pretraining losses and more challenging to solve. We design an intermediate finetuning step that uses self-supervised training to promote models' ability in cross-task generalization. Then we describe methods to leverage the structures in Wikipedia and paraphrases. In particular, we propose training losses to exploit hyperlinks, article structures, and article category graphs for entity-, discourse-, entailment-related knowledge. We propose a framework that uses paraphrase pairs to disentangle semantics and syntax in sentence representations. We extend the framework for a novel generation task that controls the syntax of output text with a sentential exemplar. Lastly, we discuss our work on tailoring textual resources for establishing challenging evaluation tasks. We introduce three datasets by defining novel tasks using various fan-contributed websites, including a long-form data-to-text generation dataset, a screenplay summarization dataset, and a long-form story generation dataset. These datasets have unique characteristics offering challenges to future work in their respective task settings.
Pre-training has shown success in different areas of machine learning, such as Computer Vision, Natural Language Processing (NLP), and medical imaging. However, it has not been fully explored for clinical data analysis. An immense amount of clinical records are recorded, but still, data and labels can be scarce for data collected in small hospitals or dealing with rare diseases. In such scenarios, pre-training on a larger set of unlabelled clinical data could improve performance. In this paper, we propose novel unsupervised pre-training techniques designed for heterogeneous, multi-modal clinical data for patient outcome prediction inspired by masked language modeling (MLM), by leveraging graph deep learning over population graphs. To this end, we further propose a graph-transformer-based network, designed to handle heterogeneous clinical data. By combining masking-based pre-training with a transformer-based network, we translate the success of masking-based pre-training in other domains to heterogeneous clinical data. We show the benefit of our pre-training method in a self-supervised and a transfer learning setting, utilizing three medical datasets TADPOLE, MIMIC-III, and a Sepsis Prediction Dataset. We find that our proposed pre-training methods help in modeling the data at a patient and population level and improve performance in different fine-tuning tasks on all datasets.
Prompted models have demonstrated impressive few-shot learning abilities. Repeated interactions at test-time with a single model, or the composition of multiple models together, further expands capabilities. These compositions are probabilistic models, and may be expressed in the language of graphical models with random variables whose values are complex data types such as strings. Cases with control flow and dynamic structure require techniques from probabilistic programming, which allow implementing disparate model structures and inference strategies in a unified language. We formalize several existing techniques from this perspective, including scratchpads / chain of thought, verifiers, STaR, selection-inference, and tool use. We refer to the resulting programs as language model cascades.
Detecting and mitigating harmful biases in modern language models are widely recognized as crucial, open problems. In this paper, we take a step back and investigate how language models come to be biased in the first place. We use a relatively small language model, using the LSTM architecture trained on an English Wikipedia corpus. With full access to the data and to the model parameters as they change during every step while training, we can map in detail how the representation of gender develops, what patterns in the dataset drive this, and how the model's internal state relates to the bias in a downstream task (semantic textual similarity). We find that the representation of gender is dynamic and identify different phases during training. Furthermore, we show that gender information is represented increasingly locally in the input embeddings of the model and that, as a consequence, debiasing these can be effective in reducing the downstream bias. Monitoring the training dynamics, allows us to detect an asymmetry in how the female and male gender are represented in the input embeddings. This is important, as it may cause naive mitigation strategies to introduce new undesirable biases. We discuss the relevance of the findings for mitigation strategies more generally and the prospects of generalizing our methods to larger language models, the Transformer architecture, other languages and other undesirable biases.
Transformer-based language models have recently achieved remarkable results in many natural language tasks. However, performance on leaderboards is generally achieved by leveraging massive amounts of training data, and rarely by encoding explicit linguistic knowledge into neural models. This has led many to question the relevance of linguistics for modern natural language processing. In this dissertation, I present several case studies to illustrate how theoretical linguistics and neural language models are still relevant to each other. First, language models are useful to linguists by providing an objective tool to measure semantic distance, which is difficult to do using traditional methods. On the other hand, linguistic theory contributes to language modelling research by providing frameworks and sources of data to probe our language models for specific aspects of language understanding. This thesis contributes three studies that explore different aspects of the syntax-semantics interface in language models. In the first part of my thesis, I apply language models to the problem of word class flexibility. Using mBERT as a source of semantic distance measurements, I present evidence in favour of analyzing word class flexibility as a directional process. In the second part of my thesis, I propose a method to measure surprisal at intermediate layers of language models. My experiments show that sentences containing morphosyntactic anomalies trigger surprisals earlier in language models than semantic and commonsense anomalies. Finally, in the third part of my thesis, I adapt several psycholinguistic studies to show that language models contain knowledge of argument structure constructions. In summary, my thesis develops new connections between natural language processing, linguistic theory, and psycholinguistics to provide fresh perspectives for the interpretation of language models.
Recently, for few-shot or even zero-shot learning, the new paradigm "pre-train, prompt, and predict" has achieved remarkable achievements compared with the "pre-train, fine-tune" paradigm. After the success of prompt-based GPT-3, a series of masked language model (MLM)-based (e.g., BERT, RoBERTa) prompt learning methods became popular and widely used. However, another efficient pre-trained discriminative model, ELECTRA, has probably been neglected. In this paper, we attempt to accomplish several NLP tasks in the zero-shot scenario using a novel our proposed replaced token detection (RTD)-based prompt learning method. Experimental results show that ELECTRA model based on RTD-prompt learning achieves surprisingly state-of-the-art zero-shot performance. Numerically, compared to MLM-RoBERTa-large and MLM-BERT-large, our RTD-ELECTRA-large has an average of about 8.4% and 13.7% improvement on all 15 tasks. Especially on the SST-2 task, our RTD-ELECTRA-large achieves an astonishing 90.1% accuracy without any training data. Overall, compared to the pre-trained masked language models, the pre-trained replaced token detection model performs better in zero-shot learning. The source code is available at: https://github.com/nishiwen1214/RTD-ELECTRA.
Building a universal Video-Language model for solving various video understanding tasks (\emph{e.g.}, text-video retrieval, video question answering) is an open challenge to the machine learning field. Towards this goal, most recent works build the model by stacking uni-modal and cross-modal feature encoders and train it with pair-wise contrastive pre-text tasks. Though offering attractive generality, the resulted models have to compromise between efficiency and performance. They mostly adopt different architectures to deal with different downstream tasks. We find this is because the pair-wise training cannot well \emph{align} and \emph{fuse} features from different modalities. We then introduce \textbf{Clover}\textemdash a Correlated Video-Language pre-training method\textemdash towards a universal Video-Language model for solving multiple video understanding tasks with neither performance nor efficiency compromise. It improves cross-modal feature alignment and fusion via a novel tri-modal alignment pre-training task. Additionally, we propose to enhance the tri-modal alignment via incorporating learning from semantic masked samples and a new pair-wise ranking loss. Clover establishes new state-of-the-arts on multiple downstream tasks, including three retrieval tasks for both zero-shot and fine-tuning settings, and eight video question answering tasks. Codes and pre-trained models will be released at \url{https://github.com/LeeYN-43/Clover}.
Objects are usually associated with multiple attributes, and these attributes often exhibit high correlations. Modeling complex relationships between attributes poses a great challenge for multi-attribute learning. This paper proposes a simple yet generic framework named Label2Label to exploit the complex attribute correlations. Label2Label is the first attempt for multi-attribute prediction from the perspective of language modeling. Specifically, it treats each attribute label as a "word" describing the sample. As each sample is annotated with multiple attribute labels, these "words" will naturally form an unordered but meaningful "sentence", which depicts the semantic information of the corresponding sample. Inspired by the remarkable success of pre-training language models in NLP, Label2Label introduces an image-conditioned masked language model, which randomly masks some of the "word" tokens from the label "sentence" and aims to recover them based on the masked "sentence" and the context conveyed by image features. Our intuition is that the instance-wise attribute relations are well grasped if the neural net can infer the missing attributes based on the context and the remaining attribute hints. Label2Label is conceptually simple and empirically powerful. Without incorporating task-specific prior knowledge and highly specialized network designs, our approach achieves state-of-the-art results on three different multi-attribute learning tasks, compared to highly customized domain-specific methods. Code is available at https://github.com/Li-Wanhua/Label2Label.
Transformer-based models show their effectiveness across multiple domains and tasks. The self-attention allows to combine information from all sequence elements into context-aware representations. However, global and local information has to be stored mostly in the same element-wise representations. Moreover, the length of an input sequence is limited by quadratic computational complexity of self-attention. In this work, we propose and study a memory-augmented segment-level recurrent Transformer (RMT). Memory allows to store and process local and global information as well as to pass information between segments of the long sequence with the help of recurrence. We implement a memory mechanism with no changes to Transformer model by adding special memory tokens to the input or output sequence. Then the model is trained to control both memory operations and sequence representations processing. Results of experiments show that RMT performs on par with the Transformer-XL on language modeling for smaller memory sizes and outperforms it for tasks that require longer sequence processing. We show that adding memory tokens to Tr-XL is able to improve its performance. This makes Recurrent Memory Transformer a promising architecture for applications that require learning of long-term dependencies and general purpose in memory processing, such as algorithmic tasks and reasoning.
Context-aware STR methods typically use internal autoregressive (AR) language models (LM). Inherent limitations of AR models motivated two-stage methods which employ an external LM. The conditional independence of the external LM on the input image may cause it to erroneously rectify correct predictions, leading to significant inefficiencies. Our method, PARSeq, learns an ensemble of internal AR LMs with shared weights using Permutation Language Modeling. It unifies context-free non-AR and context-aware AR inference, and iterative refinement using bidirectional context. Using synthetic training data, PARSeq achieves state-of-the-art (SOTA) results in STR benchmarks (91.9% accuracy) and more challenging datasets. It establishes new SOTA results (96.0% accuracy) when trained on real data. PARSeq is optimal on accuracy vs parameter count, FLOPS, and latency because of its simple, unified structure and parallel token processing. Due to its extensive use of attention, it is robust on arbitrarily-oriented text which is common in real-world images. Code, pretrained weights, and data are available at: https://github.com/baudm/parseq.
We leverage pre-trained language models to solve the task of complex NER for two low-resource languages: Chinese and Spanish. We use the technique of Whole Word Masking(WWM) to boost the performance of masked language modeling objective on large and unsupervised corpora. We experiment with multiple neural network architectures, incorporating CRF, BiLSTMs, and Linear Classifiers on top of a fine-tuned BERT layer. All our models outperform the baseline by a significant margin and our best performing model obtains a competitive position on the evaluation leaderboard for the blind test set.
Language models are defined over a finite set of inputs, which creates a vocabulary bottleneck when we attempt to scale the number of supported languages. Tackling this bottleneck results in a trade-off between what can be represented in the embedding matrix and computational issues in the output layer. This paper introduces PIXEL, the Pixel-based Encoder of Language, which suffers from neither of these issues. PIXEL is a pretrained language model that renders text as images, making it possible to transfer representations across languages based on orthographic similarity or the co-activation of pixels. PIXEL is trained to reconstruct the pixels of masked patches instead of predicting a distribution over tokens. We pretrain the 86M parameter PIXEL model on the same English data as BERT and evaluate on syntactic and semantic tasks in typologically diverse languages, including various non-Latin scripts. We find that PIXEL substantially outperforms BERT on syntactic and semantic processing tasks on scripts that are not found in the pretraining data, but PIXEL is slightly weaker than BERT when working with Latin scripts. Furthermore, we find that PIXEL is more robust than BERT to orthographic attacks and linguistic code-switching, further confirming the benefits of modelling language with pixels.
Transformer models have recently emerged as one of the foundational models in natural language processing, and as a byproduct, there is significant recent interest and investment in scaling these models. However, the training and inference costs of these large Transformer language models are prohibitive, thus necessitating more research in identifying more efficient variants. In this work, we propose a simple yet effective modification to the Transformer architecture inspired by the literature in statistical language modeling, by augmenting the model with n-grams that are constructed from a discrete latent representation of the text sequence. We evaluate our model, the N-Grammer on language modeling on the C4 data-set as well as text classification on the SuperGLUE data-set, and find that it outperforms several strong baselines such as the Transformer and the Primer. We open-source our model for reproducibility purposes in Jax.
Goal-conditioned policies for robotic navigation can be trained on large, unannotated datasets, providing for good generalization to real-world settings. However, particularly in vision-based settings where specifying goals requires an image, this makes for an unnatural interface. Language provides a more convenient modality for communication with robots, but contemporary methods typically require expensive supervision, in the form of trajectories annotated with language descriptions. We present a system, LM-Nav, for robotic navigation that enjoys the benefits of training on unannotated large datasets of trajectories, while still providing a high-level interface to the user. Instead of utilizing a labeled instruction following dataset, we show that such a system can be constructed entirely out of pre-trained models for navigation (ViNG), image-language association (CLIP), and language modeling (GPT-3), without requiring any fine-tuning or language-annotated robot data. We instantiate LM-Nav on a real-world mobile robot and demonstrate long-horizon navigation through complex, outdoor environments from natural language instructions. For videos of our experiments, code release, and an interactive Colab notebook that runs in your browser, please check out our project page https://sites.google.com/view/lmnav
Opinion formation and propagation are crucial phenomena in social networks and have been extensively studied across several disciplines. Traditionally, theoretical models of opinion dynamics have been proposed to describe the interactions between individuals (i.e., social interaction) and their impact on the evolution of collective opinions. Although these models can incorporate sociological and psychological knowledge on the mechanisms of social interaction, they demand extensive calibration with real data to make reliable predictions, requiring much time and effort. Recently, the widespread use of social media platforms provides new paradigms to learn deep learning models from a large volume of social media data. However, these methods ignore any scientific knowledge about the mechanism of social interaction. In this work, we present the first hybrid method called Sociologically-Informed Neural Network (SINN), which integrates theoretical models and social media data by transporting the concepts of physics-informed neural networks (PINNs) from natural science (i.e., physics) into social science (i.e., sociology and social psychology). In particular, we recast theoretical models as ordinary differential equations (ODEs). Then we train a neural network that simultaneously approximates the data and conforms to the ODEs that represent the social scientific knowledge. In addition, we extend PINNs by integrating matrix factorization and a language model to incorporate rich side information (e.g., user profiles) and structural knowledge (e.g., cluster structure of the social interaction network). Moreover, we develop an end-to-end training procedure for SINN, which involves Gumbel-Softmax approximation to include stochastic mechanisms of social interaction. Extensive experiments on real-world and synthetic datasets show SINN outperforms six baseline methods in predicting opinion dynamics.