While Neural Machine Translation(NMT) has achieved great progress in recent years, it still suffers from inaccurate translation of entities (e.g., person/organization name, location), due to the lack of entity training instances. When we humans encounter an unknown entity during translation, we usually first look up in a dictionary and then organize the entity translation together with the translations of other parts to form a smooth target sentence. Inspired by this translation process, we propose an Extract-and-Attend approach to enhance entity translation in NMT, where the translation candidates of source entities are first extracted from a dictionary and then attended to by the NMT model to generate the target sentence. Specifically, the translation candidates are extracted by first detecting the entities in a source sentence and then translating the entities through looking up in a dictionary. Then, the extracted candidates are added as a prefix of the decoder input to be attended to by the decoder when generating the target sentence through self-attention. Experiments conducted on En-Zh and En-Ru demonstrate that the proposed method is effective on improving both the translation accuracy of entities and the overall translation quality, with up to 35% reduction on entity error rate and 0.85 gain on BLEU and 13.8 gain on COMET.
Summarizing lengthy documents is a common and essential task in our daily lives. Although recent advancements in neural summarization models can assist in crafting general-purpose summaries, human writers often have specific requirements that call for a more customized approach. To address this need, we introduce REVISE (Refinement and Editing via Iterative Summarization Enhancement), an innovative framework designed to facilitate iterative editing and refinement of draft summaries by human writers. Within our framework, writers can effortlessly modify unsatisfactory segments at any location or length and provide optional starting phrases -- our system will generate coherent alternatives that seamlessly integrate with the existing summary. At its core, REVISE incorporates a modified fill-in-the-middle model with the encoder-decoder architecture while developing novel evaluation metrics tailored for the summarization task. In essence, our framework empowers users to create high-quality, personalized summaries by effectively harnessing both human expertise and AI capabilities, ultimately transforming the summarization process into a truly collaborative and adaptive experience.
This paper introduces an end-to-end neural speech restoration model, HD-DEMUCS, demonstrating efficacy across multiple distortion environments. Unlike conventional approaches that employ cascading frameworks to remove undesirable noise first and then restore missing signal components, our model performs these tasks in parallel using two heterogeneous decoder networks. Based on the U-Net style encoder-decoder framework, we attach an additional decoder so that each decoder network performs noise suppression or restoration separately. We carefully design each decoder architecture to operate appropriately depending on its objectives. Additionally, we improve performance by leveraging a learnable weighting factor, aggregating the two decoder output waveforms. Experimental results with objective metrics across various environments clearly demonstrate the effectiveness of our approach over a single decoder or multi-stage systems for general speech restoration task.
We introduce a data-driven learning framework that assimilates two powerful ideas: ideal large eddy simulation (LES) from turbulence closure modeling and neural stochastic differential equations (SDE) for stochastic modeling. The ideal LES models the LES flow by treating each full-order trajectory as a random realization of the underlying dynamics, as such, the effect of small-scales is marginalized to obtain the deterministic evolution of the LES state. However, ideal LES is analytically intractable. In our work, we use a latent neural SDE to model the evolution of the stochastic process and an encoder-decoder pair for transforming between the latent space and the desired ideal flow field. This stands in sharp contrast to other types of neural parameterization of closure models where each trajectory is treated as a deterministic realization of the dynamics. We show the effectiveness of our approach (niLES - neural ideal LES) on a challenging chaotic dynamical system: Kolmogorov flow at a Reynolds number of 20,000. Compared to competing methods, our method can handle non-uniform geometries using unstructured meshes seamlessly. In particular, niLES leads to trajectories with more accurate statistics and enhances stability, particularly for long-horizon rollouts.
A robust summarization system should be able to capture the gist of the document, regardless of the specific word choices or noise in the input. In this work, we first explore the summarization models' robustness against perturbations including word-level synonym substitution and noise. To create semantic-consistent substitutes, we propose a SummAttacker, which is an efficient approach to generating adversarial samples based on language models. Experimental results show that state-of-the-art summarization models have a significant decrease in performance on adversarial and noisy test sets. Next, we analyze the vulnerability of the summarization systems and explore improving the robustness by data augmentation. Specifically, the first brittleness factor we found is the poor understanding of infrequent words in the input. Correspondingly, we feed the encoder with more diverse cases created by SummAttacker in the input space. The other factor is in the latent space, where the attacked inputs bring more variations to the hidden states. Hence, we construct adversarial decoder input and devise manifold softmixing operation in hidden space to introduce more diversity. Experimental results on Gigaword and CNN/DM datasets demonstrate that our approach achieves significant improvements over strong baselines and exhibits higher robustness on noisy, attacked, and clean datasets.
Recent advances in large language models have shown that autoregressive modeling can generate complex and novel sequences that have many real-world applications. However, these models must generate outputs autoregressively, which becomes time-consuming when dealing with long sequences. Hierarchical autoregressive approaches that compress data have been proposed as a solution, but these methods still generate outputs at the original data frequency, resulting in slow and memory-intensive models. In this paper, we propose a model based on the Hierarchical Recurrent Encoder Decoder (HRED) architecture. This model independently encodes input sub-sequences without global context, processes these sequences using a lower-frequency model, and decodes outputs at the original data frequency. By interpreting the encoder as an implicitly defined embedding matrix and using sampled softmax estimation, we develop a training algorithm that can train the entire model without a high-frequency decoder, which is the most memory and compute-intensive part of hierarchical approaches. In a final, brief phase, we train the decoder to generate data at the original granularity. Our algorithm significantly reduces memory requirements for training autoregressive models and it also improves the total training wall-clock time.
Multi-view 3D object detection is becoming popular in autonomous driving due to its high effectiveness and low cost. Most of the current state-of-the-art detectors follow the query-based bird's-eye-view (BEV) paradigm, which benefits from both BEV's strong perception power and end-to-end pipeline. Despite achieving substantial progress, existing works model objects via globally leveraging temporal and spatial information of BEV features, resulting in problems when handling the challenging complex and dynamic autonomous driving scenarios. In this paper, we proposed an Object-Centric query-BEV detector OCBEV, which can carve the temporal and spatial cues of moving targets more effectively. OCBEV comprises three designs: Object Aligned Temporal Fusion aligns the BEV feature based on ego-motion and estimated current locations of moving objects, leading to a precise instance-level feature fusion. Object Focused Multi-View Sampling samples more 3D features from an adaptive local height ranges of objects for each scene to enrich foreground information. Object Informed Query Enhancement replaces part of pre-defined decoder queries in common DETR-style decoders with positional features of objects on high-confidence locations, introducing more direct object positional priors. Extensive experimental evaluations are conducted on the challenging nuScenes dataset. Our approach achieves a state-of-the-art result, surpassing the traditional BEVFormer by 1.5 NDS points. Moreover, we have a faster convergence speed and only need half of the training iterations to get comparable performance, which further demonstrates its effectiveness.
We propose DocFormerv2, a multi-modal transformer for Visual Document Understanding (VDU). The VDU domain entails understanding documents (beyond mere OCR predictions) e.g., extracting information from a form, VQA for documents and other tasks. VDU is challenging as it needs a model to make sense of multiple modalities (visual, language and spatial) to make a prediction. Our approach, termed DocFormerv2 is an encoder-decoder transformer which takes as input - vision, language and spatial features. DocFormerv2 is pre-trained with unsupervised tasks employed asymmetrically i.e., two novel document tasks on encoder and one on the auto-regressive decoder. The unsupervised tasks have been carefully designed to ensure that the pre-training encourages local-feature alignment between multiple modalities. DocFormerv2 when evaluated on nine datasets shows state-of-the-art performance over strong baselines e.g. TabFact (4.3%), InfoVQA (1.4%), FUNSD (1%). Furthermore, to show generalization capabilities, on three VQA tasks involving scene-text, Doc- Formerv2 outperforms previous comparably-sized models and even does better than much larger models (such as GIT2, PaLi and Flamingo) on some tasks. Extensive ablations show that due to its pre-training, DocFormerv2 understands multiple modalities better than prior-art in VDU.
State-of-the-art non-autoregressive text-to-speech (TTS) models based on FastSpeech 2 can efficiently synthesise high-fidelity and natural speech. For expressive speech datasets however, we observe characteristic audio distortions. We demonstrate that such artefacts are introduced to the vocoder reconstruction by over-smooth mel-spectrogram predictions, which are induced by the choice of mean-squared-error (MSE) loss for training the mel-spectrogram decoder. With MSE loss FastSpeech 2 is limited to learn conditional averages of the training distribution, which might not lie close to a natural sample if the distribution still appears multimodal after all conditioning signals. To alleviate this problem, we introduce TVC-GMM, a mixture model of Trivariate-Chain Gaussian distributions, to model the residual multimodality. TVC-GMM reduces spectrogram smoothness and improves perceptual audio quality in particular for expressive datasets as shown by both objective and subjective evaluation.
Enhancing images in low-light scenes is a challenging but widely concerned task in the computer vision. The mainstream learning-based methods mainly acquire the enhanced model by learning the data distribution from the specific scenes, causing poor adaptability (even failure) when meeting real-world scenarios that have never been encountered before. The main obstacle lies in the modeling conundrum from distribution discrepancy across different scenes. To remedy this, we first explore relationships between diverse low-light scenes based on statistical analysis, i.e., the network parameters of the encoder trained in different data distributions are close. We introduce the bilevel paradigm to model the above latent correspondence from the perspective of hyperparameter optimization. A bilevel learning framework is constructed to endow the scene-irrelevant generality of the encoder towards diverse scenes (i.e., freezing the encoder in the adaptation and testing phases). Further, we define a reinforced bilevel learning framework to provide a meta-initialization for scene-specific decoder to further ameliorate visual quality. Moreover, to improve the practicability, we establish a Retinex-induced architecture with adaptive denoising and apply our built learning framework to acquire its parameters by using two training losses including supervised and unsupervised forms. Extensive experimental evaluations on multiple datasets verify our adaptability and competitive performance against existing state-of-the-art works. The code and datasets will be available at https://github.com/vis-opt-group/BL.
Manual medical image segmentation is subjective and suffers from annotator-related bias, which can be mimicked or amplified by deep learning methods. Recently, researchers have suggested that such bias is the combination of the annotator preference and stochastic error, which are modeled by convolution blocks located after decoder and pixel-wise independent Gaussian distribution, respectively. It is unlikely that convolution blocks can effectively model the varying degrees of preference at the full resolution level. Additionally, the independent pixel-wise Gaussian distribution disregards pixel correlations, leading to a discontinuous boundary. This paper proposes a Transformer-based Annotation Bias-aware (TAB) medical image segmentation model, which tackles the annotator-related bias via modeling annotator preference and stochastic errors. TAB employs the Transformer with learnable queries to extract the different preference-focused features. This enables TAB to produce segmentation with various preferences simultaneously using a single segmentation head. Moreover, TAB takes the multivariant normal distribution assumption that models pixel correlations, and learns the annotation distribution to disentangle the stochastic error. We evaluated our TAB on an OD/OC segmentation benchmark annotated by six annotators. Our results suggest that TAB outperforms existing medical image segmentation models which take into account the annotator-related bias.
Information extraction, e.g., attribute value extraction, has been extensively studied and formulated based only on text. However, many attributes can benefit from image-based extraction, like color, shape, pattern, among others. The visual modality has long been underutilized, mainly due to multimodal annotation difficulty. In this paper, we aim to patch the visual modality to the textual-established attribute information extractor. The cross-modality integration faces several unique challenges: (C1) images and textual descriptions are loosely paired intra-sample and inter-samples; (C2) images usually contain rich backgrounds that can mislead the prediction; (C3) weakly supervised labels from textual-established extractors are biased for multimodal training. We present PV2TEA, an encoder-decoder architecture equipped with three bias reduction schemes: (S1) Augmented label-smoothed contrast to improve the cross-modality alignment for loosely-paired image and text; (S2) Attention-pruning that adaptively distinguishes the visual foreground; (S3) Two-level neighborhood regularization that mitigates the label textual bias via reliability estimation. Empirical results on real-world e-Commerce datasets demonstrate up to 11.74% absolute (20.97% relatively) F1 increase over unimodal baselines.
We conduct a preliminary inquiry into the ability of generative transformer models to deductively reason from premises provided. We observe notable differences in the performance of models coming from different training setups and find that the deductive reasoning ability increases with scale. Further, we discover that the performance generally does not decrease with the length of the deductive chain needed to reach the conclusion, with the exception of OpenAI GPT-3 and GPT-3.5 models. Our study considers a wide variety of transformer-decoder models, ranging from 117 million to 175 billion parameters in size.
Summarizing a video requires a diverse understanding of the video, ranging from recognizing scenes to evaluating how much each frame is essential enough to be selected as a summary. Self-supervised learning (SSL) is acknowledged for its robustness and flexibility to multiple downstream tasks, but the video SSL has not shown its value for dense understanding tasks like video summarization. We claim an unsupervised autoencoder with sufficient self-supervised learning does not need any extra downstream architecture design or fine-tuning weights to be utilized as a video summarization model. The proposed method to evaluate the importance score of each frame takes advantage of the reconstruction score of the autoencoder's decoder. We evaluate the method in major unsupervised video summarization benchmarks to show its effectiveness under various experimental settings.
Clinical notes in healthcare facilities are tagged with the International Classification of Diseases (ICD) code; a list of classification codes for medical diagnoses and procedures. ICD coding is a challenging multilabel text classification problem due to noisy clinical document inputs and long-tailed label distribution. Recent automated ICD coding efforts improve performance by encoding medical notes and codes with additional data and knowledge bases. However, most of them do not reflect how human coders generate the code: first, the coders select general code categories and then look for specific subcategories that are relevant to a patient's condition. Inspired by this, we propose a two-stage decoding mechanism to predict ICD codes. Our model uses the hierarchical properties of the codes to split the prediction into two steps: At first, we predict the parent code and then predict the child code based on the previous prediction. Experiments on the public MIMIC-III data set show that our model performs well in single-model settings without external data or knowledge.
This work focuses on improving the Spoken Language Identification (LangId) system for a challenge that focuses on developing robust language identification systems that are reliable for non-standard, accented (Singaporean accent), spontaneous code-switched, and child-directed speech collected via Zoom. We propose a two-stage Encoder-Decoder-based E2E model. The encoder module consists of 1D depth-wise separable convolutions with Squeeze-and-Excitation (SE) layers with a global context. The decoder module uses an attentive temporal pooling mechanism to get fixed length time-independent feature representation. The total number of parameters in the model is around 22.1 M, which is relatively light compared to using some large-scale pre-trained speech models. We achieved an EER of 15.6% in the closed track and 11.1% in the open track (baseline system 22.1%). We also curated additional LangId data from YouTube videos (having Singaporean speakers), which will be released for public use.
This paper presents techniques and findings for improving the performance of low-resource speech to text translation (ST). We conducted experiments on both simulated and real-low resource setups, on language pairs English - Portuguese, and Tamasheq - French respectively. Using the encoder-decoder framework for ST, our results show that a multilingual automatic speech recognition system acts as a good initialization under low-resource scenarios. Furthermore, using the CTC as an additional objective for translation during training and decoding helps to reorder the internal representations and improves the final translation. Through our experiments, we try to identify various factors (initializations, objectives, and hyper-parameters) that contribute the most for improvements in low-resource setups. With only 300 hours of pre-training data, our model achieved 7.3 BLEU score on Tamasheq - French data, outperforming prior published works from IWSLT 2022 by 1.6 points.
We consider the problem of optimizing the discriminator in generative adversarial networks (GANs) subject to higher-order gradient regularization. We show analytically, via the least-squares (LSGAN) and Wasserstein (WGAN) GAN variants, that the discriminator optimization problem is one of interpolation in $n$-dimensions. The optimal discriminator, derived using variational Calculus, turns out to be the solution to a partial differential equation involving the iterated Laplacian or the polyharmonic operator. The solution is implementable in closed-form via polyharmonic radial basis function (RBF) interpolation. In view of the polyharmonic connection, we refer to the corresponding GANs as Poly-LSGAN and Poly-WGAN. Through experimental validation on multivariate Gaussians, we show that implementing the optimal RBF discriminator in closed-form, with penalty orders $m \approx\lceil \frac{n}{2} \rceil $, results in superior performance, compared to training GAN with arbitrarily chosen discriminator architectures. We employ the Poly-WGAN discriminator to model the latent space distribution of the data with encoder-decoder-based GAN flavors such as Wasserstein autoencoders.
Self-supervised learning (SSL) has recently allowed leveraging large datasets of unlabeled speech signals to reach impressive performance on speech tasks using only small amounts of annotated data. The high number of proposed approaches fostered the need and rise of extended benchmarks that evaluate their performance on a set of downstream tasks exploring various aspects of the speech signal. However, and while the number of considered tasks has been growing, most rely upon a single decoding architecture that maps the frozen SSL representations to the downstream labels. This work investigates the robustness of such benchmarking results to changes in the decoder architecture. Interestingly, it appears that varying the architecture of the downstream decoder leads to significant variations in the leaderboards of most tasks. Concerningly, our study reveals that benchmarking using limited decoders may cause a counterproductive increase in the sizes of the developed SSL models.
Deep learning-based medical image segmentation models often suffer from domain shift, where the models trained on a source domain do not generalize well to other unseen domains. As a prompt-driven foundation model with powerful generalization capabilities, the Segment Anything Model (SAM) shows potential for improving the cross-domain robustness of medical image segmentation. However, SAM performs significantly worse in automatic segmentation scenarios than when manually prompted, hindering its direct application to domain generalization. Upon further investigation, we discovered that the degradation in performance was related to the coupling effect of inevitable poor prompts and mask generation. To address the coupling effect, we propose the Decoupled SAM (DeSAM). DeSAM modifies SAM's mask decoder by introducing two new modules: a prompt-relevant IoU module (PRIM) and a prompt-decoupled mask module (PDMM). PRIM predicts the IoU score and generates mask embeddings, while PDMM extracts multi-scale features from the intermediate layers of the image encoder and fuses them with the mask embeddings from PRIM to generate the final segmentation mask. This decoupled design allows DeSAM to leverage the pre-trained weights while minimizing the performance degradation caused by poor prompts. We conducted experiments on publicly available cross-site prostate and cross-modality abdominal image segmentation datasets. The results show that our DeSAM leads to a substantial performance improvement over previous state-of-theart domain generalization methods. The code is publicly available at https://github.com/yifangao112/DeSAM.