End-to-end ASR models trained on large amount of data tend to be implicitly biased towards language semantics of the training data. Internal language model estimation (ILME) has been proposed to mitigate this bias for autoregressive models such as attention-based encoder-decoder and RNN-T. Typically, ILME is performed by modularizing the acoustic and language components of the model architecture, and eliminating the acoustic input to perform log-linear interpolation with the text-only posterior. However, for CTC-based ASR, it is not as straightforward to decouple the model into such acoustic and language components, as CTC log-posteriors are computed in a non-autoregressive manner. In this work, we propose a novel ILME technique for CTC-based ASR models. Our method iteratively masks the audio timesteps to estimate a pseudo log-likelihood of the internal LM by accumulating log-posteriors for only the masked timesteps. Extensive evaluation across multiple out-of-domain datasets reveals that the proposed approach improves WER by up to 9.8% and OOV F1-score by up to 24.6% relative to Shallow Fusion, when only text data from target domain is available. In the case of zero-shot domain adaptation, with no access to any target domain data, we demonstrate that removing the source domain bias with ILME can still outperform Shallow Fusion to improve WER by up to 9.3% relative.
相關內容
Scene text removal (STR) aims at replacing text strokes in natural scenes with visually coherent backgrounds. Recent STR approaches rely on iterative refinements or explicit text masks, resulting in higher complexity and sensitivity to the accuracy of text localization. Moreover, most existing STR methods utilize convolutional neural networks (CNNs) for feature representation while the potential of vision Transformers (ViTs) remains largely unexplored. In this paper, we propose a simple-yet-effective ViT-based text eraser, dubbed ViTEraser. Following a concise encoder-decoder framework, different types of ViTs can be easily integrated into ViTEraser to enhance the long-range dependencies and global reasoning. Specifically, the encoder hierarchically maps the input image into the hidden space through ViT blocks and patch embedding layers, while the decoder gradually upsamples the hidden features to the text-erased image with ViT blocks and patch splitting layers. As ViTEraser implicitly integrates text localization and inpainting, we propose a novel end-to-end pretraining method, termed SegMIM, which focuses the encoder and decoder on the text box segmentation and masked image modeling tasks, respectively. To verify the effectiveness of the proposed methods, we comprehensively explore the architecture, pretraining, and scalability of the ViT-based encoder-decoder for STR, which provides deep insights into the application of ViT to STR. Experimental results demonstrate that ViTEraser with SegMIM achieves state-of-the-art performance on STR by a substantial margin. Furthermore, the extended experiment on tampered scene text detection demonstrates the generality of ViTEraser to other tasks. We believe this paper can inspire more research on ViT-based STR approaches. Code will be available at //github.com/shannanyinxiang/ViTEraser.
The Plackett--Luce model is a popular approach for ranking data analysis, where a utility vector is employed to determine the probability of each outcome based on Luce's choice axiom. In this paper, we investigate the asymptotic theory of utility vector estimation by maximizing different types of likelihood, such as the full-, marginal-, and quasi-likelihood. We provide a rank-matching interpretation for the estimating equations of these estimators and analyze their asymptotic behavior as the number of items being compared tends to infinity. In particular, we establish the uniform consistency of these estimators under conditions characterized by the topology of the underlying comparison graph sequence and demonstrate that the proposed conditions are sharp for common sampling scenarios such as the nonuniform random hypergraph model and the hypergraph stochastic block model; we also obtain the asymptotic normality of these estimators and discuss the trade-off between statistical efficiency and computational complexity for practical uncertainty quantification. Both results allow for nonuniform and inhomogeneous comparison graphs with varying edge sizes and different asymptotic orders of edge probabilities. We verify our theoretical findings by conducting detailed numerical experiments.
The framework of visually-guided sound source separation generally consists of three parts: visual feature extraction, multimodal feature fusion, and sound signal processing. An ongoing trend in this field has been to tailor involved visual feature extractor for informative visual guidance and separately devise module for feature fusion, while utilizing U-Net by default for sound analysis. However, such divide-and-conquer paradigm is parameter inefficient and, meanwhile, may obtain suboptimal performance as jointly optimizing and harmonizing various model components is challengeable. By contrast, this paper presents a novel approach, dubbed audio-visual predictive coding (AVPC), to tackle this task in a parameter efficient and more effective manner. The network of AVPC features a simple ResNet-based video analysis network for deriving semantic visual features, and a predictive coding-based sound separation network that can extract audio features, fuse multimodal information, and predict sound separation masks in the same architecture. By iteratively minimizing the prediction error between features, AVPC integrates audio and visual information recursively, leading to progressively improved performance. In addition, we develop a valid self-supervised learning strategy for AVPC via co-predicting two audio-visual representations of the same sound source. Extensive evaluations demonstrate that AVPC outperforms several baselines in separating musical instrument sounds, while reducing the model size significantly. Code is available at: //github.com/zjsong/Audio-Visual-Predictive-Coding.
Language model (LM) based audio generation frameworks, e.g., AudioLM, have recently achieved new state-of-the-art performance in zero-shot audio generation. In this paper, we explore the feasibility of LMs for zero-shot voice conversion. An intuitive approach is to follow AudioLM - Tokenizing speech into semantic and acoustic tokens respectively by HuBERT and SoundStream, and converting source semantic tokens to target acoustic tokens conditioned on acoustic tokens of the target speaker. However, such an approach encounters several issues: 1) the linguistic content contained in semantic tokens may get dispersed during multi-layer modeling while the lengthy speech input in the voice conversion task makes contextual learning even harder; 2) the semantic tokens still contain speaker-related information, which may be leaked to the target speech, lowering the target speaker similarity; 3) the generation diversity in the sampling of the LM can lead to unexpected outcomes during inference, leading to unnatural pronunciation and speech quality degradation. To mitigate these problems, we propose LM-VC, a two-stage language modeling approach that generates coarse acoustic tokens for recovering the source linguistic content and target speaker's timbre, and then reconstructs the fine for acoustic details as converted speech. Specifically, to enhance content preservation and facilitates better disentanglement, a masked prefix LM with a mask prediction strategy is used for coarse acoustic modeling. This model is encouraged to recover the masked content from the surrounding context and generate target speech based on the target speaker's utterance and corrupted semantic tokens. Besides, to further alleviate the sampling error in the generation, an external LM, which employs window attention to capture the local acoustic relations, is introduced to participate in the coarse acoustic modeling.
In recent years, online social networks have been the target of adversaries who seek to introduce discord into societies, to undermine democracies and to destabilize communities. Often the goal is not to favor a certain side of a conflict but to increase disagreement and polarization. To get a mathematical understanding of such attacks, researchers use opinion-formation models from sociology, such as the Friedkin--Johnsen model, and formally study how much discord the adversary can produce when altering the opinions for only a small set of users. In this line of work, it is commonly assumed that the adversary has full knowledge about the network topology and the opinions of all users. However, the latter assumption is often unrealistic in practice, where user opinions are not available or simply difficult to estimate accurately. To address this concern, we raise the following question: Can an attacker sow discord in a social network, even when only the network topology is known? We answer this question affirmatively. We present approximation algorithms for detecting a small set of users who are highly influential for the disagreement and polarization in the network. We show that when the adversary radicalizes these users and if the initial disagreement/polarization in the network is not very high, then our method gives a constant-factor approximation on the setting when the user opinions are known. To find the set of influential users, we provide a novel approximation algorithm for a variant of MaxCut in graphs with positive and negative edge weights. We experimentally evaluate our methods, which have access only to the network topology, and we find that they have similar performance as methods that have access to the network topology and all user opinions. We further present an NP-hardness proof, which was an open question by Chen and Racz [IEEE Trans. Netw. Sci. Eng., 2021].
An algorithm is said to be adaptive to a certain parameter (of the problem) if it does not need a priori knowledge of such a parameter but performs competitively to those that know it. This dissertation presents our work on adaptive algorithms in following scenarios: 1. In the stochastic optimization setting, we only receive stochastic gradients and the level of noise in evaluating them greatly affects the convergence rate. Tuning is typically required when without prior knowledge of the noise scale in order to achieve the optimal rate. Considering this, we designed and analyzed noise-adaptive algorithms that can automatically ensure (near)-optimal rates under different noise scales without knowing it. 2. In training deep neural networks, the scales of gradient magnitudes in each coordinate can scatter across a very wide range unless normalization techniques, like BatchNorm, are employed. In such situations, algorithms not addressing this problem of gradient scales can behave very poorly. To mitigate this, we formally established the advantage of scale-free algorithms that adapt to the gradient scales and presented its real benefits in empirical experiments. 3. Traditional analyses in non-convex optimization typically rely on the smoothness assumption. Yet, this condition does not capture the properties of some deep learning objective functions, including the ones involving Long Short-Term Memory networks and Transformers. Instead, they satisfy a much more relaxed condition, with potentially unbounded smoothness. Under this condition, we show that a generalized SignSGD algorithm can theoretically match the best-known convergence rates obtained by SGD with gradient clipping but does not need explicit clipping at all, and it can empirically match the performance of Adam and beat others. Moreover, it can also be made to automatically adapt to the unknown relaxed smoothness.
We present a novel method that integrates subspace modeling with an adaptive generative image prior for high-dimensional MR image reconstruction. The subspace model imposes an explicit low-dimensional representation of the high-dimensional images, while the generative image prior serves as a spatial constraint on the "contrast-weighted" images or the spatial coefficients of the subspace model. A formulation was introduced to synergize these two components with complimentary regularization such as joint sparsity. A special pretraining plus subject-specific network adaptation strategy was proposed to construct an accurate generative-model-based representation for images with varying contrasts, validated by experimental data. An iterative algorithm was introduced to jointly update the subspace coefficients and the multiresolution latent space of the generative image model that leveraged a recently developed intermediate layer optimization technique for network inversion. We evaluated the utility of the proposed method in two high-dimensional imaging applications: accelerated MR parameter mapping and high-resolution MRSI. Improved performance over state-of-the-art subspace-based methods was demonstrated in both cases. Our work demonstrated the potential of integrating data-driven and adaptive generative models with low-dimensional representation for high-dimensional imaging problems.
Automated audio captioning (AAC) is an important cross-modality translation task, aiming at generating descriptions for audio clips. However, captions generated by previous AAC models have faced ``false-repetition'' errors due to the training objective. In such scenarios, we propose a new task of AAC error correction and hope to reduce such errors by post-processing AAC outputs. To tackle this problem, we use observation-based rules to corrupt captions without errors, for pseudo grammatically-erroneous sentence generation. One pair of corrupted and clean sentences can thus be used for training. We train a neural network-based model on the synthetic error dataset and apply the model to correct real errors in AAC outputs. Results on two benchmark datasets indicate that our approach significantly improves fluency while maintaining semantic information.
In federated frequency estimation (FFE), multiple clients work together to estimate the frequencies of their collective data by communicating with a server that respects the privacy constraints of Secure Summation (SecSum), a cryptographic multi-party computation protocol that ensures that the server can only access the sum of client-held vectors. For single-round FFE, it is known that count sketching is nearly information-theoretically optimal for achieving the fundamental accuracy-communication trade-offs [Chen et al., 2022]. However, we show that under the more practical multi-round FEE setting, simple adaptations of count sketching are strictly sub-optimal, and we propose a novel hybrid sketching algorithm that is provably more accurate. We also address the following fundamental question: how should a practitioner set the sketch size in a way that adapts to the hardness of the underlying problem? We propose a two-phase approach that allows for the use of a smaller sketch size for simpler problems (e.g. near-sparse or light-tailed distributions). We conclude our work by showing how differential privacy can be added to our algorithm and verifying its superior performance through extensive experiments conducted on large-scale datasets.
Object detection typically assumes that training and test data are drawn from an identical distribution, which, however, does not always hold in practice. Such a distribution mismatch will lead to a significant performance drop. In this work, we aim to improve the cross-domain robustness of object detection. We tackle the domain shift on two levels: 1) the image-level shift, such as image style, illumination, etc, and 2) the instance-level shift, such as object appearance, size, etc. We build our approach based on the recent state-of-the-art Faster R-CNN model, and design two domain adaptation components, on image level and instance level, to reduce the domain discrepancy. The two domain adaptation components are based on H-divergence theory, and are implemented by learning a domain classifier in adversarial training manner. The domain classifiers on different levels are further reinforced with a consistency regularization to learn a domain-invariant region proposal network (RPN) in the Faster R-CNN model. We evaluate our newly proposed approach using multiple datasets including Cityscapes, KITTI, SIM10K, etc. The results demonstrate the effectiveness of our proposed approach for robust object detection in various domain shift scenarios.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191