Recent advances with self-supervised learning have allowed speech recognition systems to achieve state-of-the-art (SOTA) word error rates (WER) while requiring only a fraction of the labeled training data needed by its predecessors. Notwithstanding, while such models achieve SOTA performance in matched train/test conditions, their performance degrades substantially when tested in unseen conditions. To overcome this problem, strategies such as data augmentation and/or domain shift training have been explored. Available models, however, are still too large to be considered for edge speech applications on resource-constrained devices, thus model compression tools are needed. In this paper, we explore the effects that train/test mismatch conditions have on speech recognition accuracy based on compressed self-supervised speech models. In particular, we report on the effects that parameter quantization and model pruning have on speech recognition accuracy based on the so-called robust wav2vec 2.0 model under noisy, reverberant, and noise-plus-reverberation conditions.
Counterfactual examples have proven to be valuable in the field of natural language processing (NLP) for both evaluating and improving the robustness of language models to spurious correlations in datasets. Despite their demonstrated utility for NLP, multimodal counterfactual examples have been relatively unexplored due to the difficulty of creating paired image-text data with minimal counterfactual changes. To address this challenge, we introduce a scalable framework for automatic generation of counterfactual examples using text-to-image diffusion models. We use our framework to create COCO-Counterfactuals, a multimodal counterfactual dataset of paired image and text captions based on the MS-COCO dataset. We validate the quality of COCO-Counterfactuals through human evaluations and show that existing multimodal models are challenged by our counterfactual image-text pairs. Additionally, we demonstrate the usefulness of COCO-Counterfactuals for improving out-of-domain generalization of multimodal vision-language models via training data augmentation.
We present a locality-aware method for interpreting the latent space of wavelet-based Generative Adversarial Networks (GANs), that can well capture the large spatial and spectral variability that is characteristic to satellite imagery. By focusing on preserving locality, the proposed method is able to decompose the weight-space of pre-trained GANs and recover interpretable directions that correspond to high-level semantic concepts (such as urbanization, structure density, flora presence) - that can subsequently be used for guided synthesis of satellite imagery. In contrast to typically used approaches that focus on capturing the variability of the weight-space in a reduced dimensionality space (i.e., based on Principal Component Analysis, PCA), we show that preserving locality leads to vectors with different angles, that are more robust to artifacts and can better preserve class information. Via a set of quantitative and qualitative examples, we further show that the proposed approach can outperform both baseline geometric augmentations, as well as global, PCA-based approaches for data synthesis in the context of data augmentation for satellite scene classification.
Active stereo technique using single pattern projection, a.k.a. one-shot 3D scan, have drawn a wide attention from industry, medical purposes, etc. One severe drawback of one-shot 3D scan is sparse reconstruction. In addition, since spatial pattern becomes complicated for the purpose of efficient embedding, it is easily affected by noise, which results in unstable decoding. To solve the problems, we propose a pixel-wise interpolation technique for one-shot scan, which is applicable to any types of static pattern if the pattern is regular and periodic. This is achieved by U-net which is pre-trained by CG with efficient data augmentation algorithm. In the paper, to further overcome the decoding instability, we propose a robust correspondence finding algorithm based on Markov random field (MRF) optimization. We also propose a shape refinement algorithm based on b-spline and Gaussian kernel interpolation using explicitly detected laser curves. Experiments are conducted to show the effectiveness of the proposed method using real data with strong noises and textures.
In recent years, dynamic parameterization of acoustic environments has raised increasing attention in the field of audio processing. One of the key parameters that characterize the local room acoustics in isolation from orientation and directivity of sources and receivers is the geometric room volume. Convolutional neural networks (CNNs) have been widely selected as the main models for conducting blind room acoustic parameter estimation, which aims to learn a direct mapping from audio spectrograms to corresponding labels. With the recent trend of self-attention mechanisms, this paper introduces a purely attention-based model to blindly estimate room volumes based on single-channel noisy speech signals. We demonstrate the feasibility of eliminating the reliance on CNN for this task and the proposed Transformer architecture takes Gammatone magnitude spectral coefficients and phase spectrograms as inputs. To enhance the model performance given the task-specific dataset, cross-modality transfer learning is also applied. Experimental results demonstrate that the proposed model outperforms traditional CNN models across a wide range of real-world acoustics spaces, especially with the help of the dedicated pretraining and data augmentation schemes.
Despite the basic premise that next-generation wireless networks (e.g., 6G) will be artificial intelligence (AI)-native, to date, most existing efforts remain either qualitative or incremental extensions to existing "AI for wireless" paradigms. Indeed, creating AI-native wireless networks faces significant technical challenges due to the limitations of data-driven, training-intensive AI. These limitations include the black-box nature of the AI models, their curve-fitting nature, which can limit their ability to reason and adapt, their reliance on large amounts of training data, and the energy inefficiency of large neural networks. In response to these limitations, this article presents a comprehensive, forward-looking vision that addresses these shortcomings by introducing a novel framework for building AI-native wireless networks; grounded in the emerging field of causal reasoning. Causal reasoning, founded on causal discovery, causal representation learning, and causal inference, can help build explainable, reasoning-aware, and sustainable wireless networks. Towards fulfilling this vision, we first highlight several wireless networking challenges that can be addressed by causal discovery and representation, including ultra-reliable beamforming for terahertz (THz) systems, near-accurate physical twin modeling for digital twins, training data augmentation, and semantic communication. We showcase how incorporating causal discovery can assist in achieving dynamic adaptability, resilience, and cognition in addressing these challenges. Furthermore, we outline potential frameworks that leverage causal inference to achieve the overarching objectives of future-generation networks, including intent management, dynamic adaptability, human-level cognition, reasoning, and the critical element of time sensitivity.
This work aims to improve unsupervised audio-visual pre-training. Inspired by the efficacy of data augmentation in visual contrastive learning, we propose a novel speed co-augmentation method that randomly changes the playback speeds of both audio and video data. Despite its simplicity, the speed co-augmentation method possesses two compelling attributes: (1) it increases the diversity of audio-visual pairs and doubles the size of negative pairs, resulting in a significant enhancement in the learned representations, and (2) it changes the strict correlation between audio-visual pairs but introduces a partial relationship between the augmented pairs, which is modeled by our proposed SoftInfoNCE loss to further boost the performance. Experimental results show that the proposed method significantly improves the learned representations when compared to vanilla audio-visual contrastive learning.
Synthesizing realistic human movements, dynamically responsive to the environment, is a long-standing objective in character animation, with applications in computer vision, sports, and healthcare, for motion prediction and data augmentation. Recent kinematics-based generative motion models offer impressive scalability in modeling extensive motion data, albeit without an interface to reason about and interact with physics. While simulator-in-the-loop learning approaches enable highly physically realistic behaviors, the challenges in training often affect scalability and adoption. We introduce DROP, a novel framework for modeling Dynamics Responses of humans using generative mOtion prior and Projective dynamics. DROP can be viewed as a highly stable, minimalist physics-based human simulator that interfaces with a kinematics-based generative motion prior. Utilizing projective dynamics, DROP allows flexible and simple integration of the learned motion prior as one of the projective energies, seamlessly incorporating control provided by the motion prior with Newtonian dynamics. Serving as a model-agnostic plug-in, DROP enables us to fully leverage recent advances in generative motion models for physics-based motion synthesis. We conduct extensive evaluations of our model across different motion tasks and various physical perturbations, demonstrating the scalability and diversity of responses.
Deep learning models fail on cross-domain challenges if the model is oversensitive to domain-specific attributes, e.g., lightning, background, camera angle, etc. To alleviate this problem, data augmentation coupled with consistency regularization are commonly adopted to make the model less sensitive to domain-specific attributes. Consistency regularization enforces the model to output the same representation or prediction for two views of one image. These constraints, however, are either too strict or not order-preserving for the classification probabilities. In this work, we propose the Order-preserving Consistency Regularization (OCR) for cross-domain tasks. The order-preserving property for the prediction makes the model robust to task-irrelevant transformations. As a result, the model becomes less sensitive to the domain-specific attributes. The comprehensive experiments show that our method achieves clear advantages on five different cross-domain tasks.
The EMelodyGen system focuses on emotional melody generation in ABC notation controlled by the musical feature template. Owing to the scarcity of well-structured and emotionally labeled sheet music, we designed a template for controlling emotional melody generation by statistical correlations between musical features and emotion labels derived from small-scale emotional symbolic music datasets and music psychology conclusions. We then automatically annotated a large, well-structured sheet music collection with rough emotional labels by the template, converted them into ABC notation, and reduced label imbalance by data augmentation, resulting in a dataset named Rough4Q. Our system backbone pre-trained on Rough4Q can achieve up to 99% music21 parsing rate and melodies generated by our template can lead to a 91% alignment on emotional expressions in blind listening tests. Ablation studies further validated the effectiveness of the feature controls in the template. Available code and demos are at https://github.com/monetjoe/EMelodyGen.
Imitation learning is an effective approach for training game-playing agents and, consequently, for efficient game production. However, generalization - the ability to perform well in related but unseen scenarios - is an essential requirement that remains an unsolved challenge for game AI. Generalization is difficult for imitation learning agents because it requires the algorithm to take meaningful actions outside of the training distribution. In this paper we propose a solution to this challenge. Inspired by the success of data augmentation in supervised learning, we augment the training data so the distribution of states and actions in the dataset better represents the real state-action distribution. This study evaluates methods for combining and applying data augmentations to observations, to improve generalization of imitation learning agents. It also provides a performance benchmark of these augmentations across several 3D environments. These results demonstrate that data augmentation is a promising framework for improving generalization in imitation learning agents.
To train transcriptor models that produce robust results, a large and diverse labeled dataset is required. Finding such data with the necessary characteristics is a challenging task, especially for languages less popular than English. Moreover, producing such data requires significant effort and often money. Therefore, a strategy to mitigate this problem is the use of data augmentation techniques. In this work, we propose a framework that approaches data augmentation based on deepfake audio. To validate the produced framework, experiments were conducted using existing deepfake and transcription models. A voice cloner and a dataset produced by Indians (in English) were selected, ensuring the presence of a single accent in the dataset. Subsequently, the augmented data was used to train speech to text models in various scenarios.
Mixup is an effective data augmentation method that generates new augmented samples by aggregating linear combinations of different original samples. However, if there are noises or aberrant features in the original samples, Mixup may propagate them to the augmented samples, leading to over-sensitivity of the model to these outliers . To solve this problem, this paper proposes a new Mixup method called AMPLIFY. This method uses the Attention mechanism of Transformer itself to reduce the influence of noises and aberrant values in the original samples on the prediction results, without increasing additional trainable parameters, and the computational cost is very low, thereby avoiding the problem of high resource consumption in common Mixup methods such as Sentence Mixup . The experimental results show that, under a smaller computational resource cost, AMPLIFY outperforms other Mixup methods in text classification tasks on 7 benchmark datasets, providing new ideas and new ways to further improve the performance of pre-trained models based on the Attention mechanism, such as BERT, ALBERT, RoBERTa, and GPT. Our code can be obtained at https://github.com/kiwi-lilo/AMPLIFY.
Sequential recommendation (SRS) has become the technical foundation in many applications recently, which aims to recommend the next item based on the user's historical interactions. However, sequential recommendation often faces the problem of data sparsity, which widely exists in recommender systems. Besides, most users only interact with a few items, but existing SRS models often underperform these users. Such a problem, named the long-tail user problem, is still to be resolved. Data augmentation is a distinct way to alleviate these two problems, but they often need fabricated training strategies or are hindered by poor-quality generated interactions. To address these problems, we propose a Diffusion Augmentation for Sequential Recommendation (DiffuASR) for a higher quality generation. The augmented dataset by DiffuASR can be used to train the sequential recommendation models directly, free from complex training procedures. To make the best of the generation ability of the diffusion model, we first propose a diffusion-based pseudo sequence generation framework to fill the gap between image and sequence generation. Then, a sequential U-Net is designed to adapt the diffusion noise prediction model U-Net to the discrete sequence generation task. At last, we develop two guide strategies to assimilate the preference between generated and origin sequences. To validate the proposed DiffuASR, we conduct extensive experiments on three real-world datasets with three sequential recommendation models. The experimental results illustrate the effectiveness of DiffuASR. As far as we know, DiffuASR is one pioneer that introduce the diffusion model to the recommendation.
We present MosaicFusion, a simple yet effective diffusion-based data augmentation approach for large vocabulary instance segmentation. Our method is training-free and does not rely on any label supervision. Two key designs enable us to employ an off-the-shelf text-to-image diffusion model as a useful dataset generator for object instances and mask annotations. First, we divide an image canvas into several regions and perform a single round of diffusion process to generate multiple instances simultaneously, conditioning on different text prompts. Second, we obtain corresponding instance masks by aggregating cross-attention maps associated with object prompts across layers and diffusion time steps, followed by simple thresholding and edge-aware refinement processing. Without bells and whistles, our MosaicFusion can produce a significant amount of synthetic labeled data for both rare and novel categories. Experimental results on the challenging LVIS long-tailed and open-vocabulary benchmarks demonstrate that MosaicFusion can significantly improve the performance of existing instance segmentation models, especially for rare and novel categories. Code: https://github.com/Jiahao000/MosaicFusion.
In digital advertising, the selection of the optimal item (recommendation) and its best creative presentation (creative optimization) have traditionally been considered separate disciplines. However, both contribute significantly to user satisfaction, underpinning our assumption that it relies on both an item's relevance and its presentation, particularly in the case of visual creatives. In response, we introduce the task of {\itshape Generative Creative Optimization (GCO)}, which proposes the use of generative models for creative generation that incorporate user interests, and {\itshape AdBooster}, a model for personalized ad creatives based on the Stable Diffusion outpainting architecture. This model uniquely incorporates user interests both during fine-tuning and at generation time. To further improve AdBooster's performance, we also introduce an automated data augmentation pipeline. Through our experiments on simulated data, we validate AdBooster's effectiveness in generating more relevant creatives than default product images, showing its potential of enhancing user engagement.
Rapid and accurate identification of Venous thromboembolism (VTE), a severe cardiovascular condition including deep vein thrombosis (DVT) and pulmonary embolism (PE), is important for effective treatment. Leveraging Natural Language Processing (NLP) on radiology reports, automated methods have shown promising advancements in identifying VTE events from retrospective data cohorts or aiding clinical experts in identifying VTE events from radiology reports. However, effectively training Deep Learning (DL) and the NLP models is challenging due to limited labeled medical text data, the complexity and heterogeneity of radiology reports, and data imbalance. This study proposes novel method combinations of DL methods, along with data augmentation, adaptive pre-trained NLP model selection, and a clinical expert NLP rule-based classifier, to improve the accuracy of VTE identification in unstructured (free-text) radiology reports. Our experimental results demonstrate the model's efficacy, achieving an impressive 97\% accuracy and 97\% F1 score in predicting DVT, and an outstanding 98.3\% accuracy and 98.4\% F1 score in predicting PE. These findings emphasize the model's robustness and its potential to significantly contribute to VTE research.
Biomedical image datasets can be imbalanced due to the rarity of targeted diseases. Generative Adversarial Networks play a key role in addressing this imbalance by enabling the generation of synthetic images to augment datasets. It is important to generate synthetic images that incorporate a diverse range of features to accurately represent the distribution of features present in the training imagery. Furthermore, the absence of diverse features in synthetic images can degrade the performance of machine learning classifiers. The mode collapse problem impacts Generative Adversarial Networks' capacity to generate diversified images. Mode collapse comes in two varieties: intra-class and inter-class. In this paper, both varieties of the mode collapse problem are investigated, and their subsequent impact on the diversity of synthetic X-ray images is evaluated. This work contributes an empirical demonstration of the benefits of integrating the adaptive input-image normalization with the Deep Convolutional GAN and Auxiliary Classifier GAN to alleviate the mode collapse problems. Synthetically generated images are utilized for data augmentation and training a Vision Transformer model. The classification performance of the model is evaluated using accuracy, recall, and precision scores. Results demonstrate that the DCGAN and the ACGAN with adaptive input-image normalization outperform the DCGAN and ACGAN with un-normalized X-ray images as evidenced by the superior diversity scores and classification scores.
We expose a surprising failure of generalization in auto-regressive large language models (LLMs). If a model is trained on a sentence of the form "A is B", it will not automatically generalize to the reverse direction "B is A". This is the Reversal Curse. For instance, if a model is trained on "Valentina Tereshkova was the first woman to travel to space", it will not automatically be able to answer the question, "Who was the first woman to travel to space?". Moreover, the likelihood of the correct answer ("Valentina Tershkova") will not be higher than for a random name. Thus, models do not generalize a prevalent pattern in their training set: if "A is B" occurs, "B is A" is more likely to occur. It is worth noting, however, that if "A is B" appears in-context, models can deduce the reverse relationship. We provide evidence for the Reversal Curse by finetuning GPT-3 and Llama-1 on fictitious statements such as "Uriah Hawthorne is the composer of Abyssal Melodies" and showing that they fail to correctly answer "Who composed Abyssal Melodies?". The Reversal Curse is robust across model sizes and model families and is not alleviated by data augmentation. We also evaluate ChatGPT (GPT-3.5 and GPT-4) on questions about real-world celebrities, such as "Who is Tom Cruise's mother? [A: Mary Lee Pfeiffer]" and the reverse "Who is Mary Lee Pfeiffer's son?". GPT-4 correctly answers questions like the former 79% of the time, compared to 33% for the latter. Code available at: https://github.com/lukasberglund/reversal_curse.
Deep learning has achieved remarkable success in bearing fault diagnosis. However, its performance oftentimes deteriorates when dealing with highly imbalanced or long-tailed data, while such cases are prevalent in industrial settings because fault is a rare event that occurs with an extremely low probability. Conventional data augmentation methods face fundamental limitations due to the scarcity of samples pertaining to the minority class. In this paper, we propose a supervised contrastive learning approach with a class-aware loss function to enhance the feature extraction capability of neural networks for fault diagnosis. The developed class-weighted contrastive learning quadratic network (CCQNet) consists of a quadratic convolutional residual network backbone, a contrastive learning branch utilizing a class-weighted contrastive loss, and a classifier branch employing logit-adjusted cross-entropy loss. By utilizing class-weighted contrastive loss and logit-adjusted cross-entropy loss, our approach encourages equidistant representation of class features, thereby inducing equal attention on all the classes. We further analyze the superior feature extraction ability of quadratic network by establishing the connection between quadratic neurons and autocorrelation in signal processing. Experimental results on public and proprietary datasets are used to validate the effectiveness of CCQNet, and computational results reveal that CCQNet outperforms SOTA methods in handling extremely imbalanced data substantially.