Weakly-supervised text classification aims to induce text classifiers from only a few user-provided seed words. The vast majority of previous work assumes high-quality seed words are given. However, the expert-annotated seed words are sometimes non-trivial to come up with. Furthermore, in the weakly-supervised learning setting, we do not have any labeled document to measure the seed words' efficacy, making the seed word selection process "a walk in the dark". In this work, we remove the need for expert-curated seed words by first mining (noisy) candidate seed words associated with the category names. We then train interim models with individual candidate seed words. Lastly, we estimate the interim models' error rate in an unsupervised manner. The seed words that yield the lowest estimated error rates are added to the final seed word set. A comprehensive evaluation of six binary classification tasks on four popular datasets demonstrates that the proposed method outperforms a baseline using only category name seed words and obtained comparable performance as a counterpart using expert-annotated seed words.
We propose a framework using contrastive learning as a pre-training task to perform image classification in the presence of noisy labels. Recent strategies such as pseudo-labeling, sample selection with Gaussian Mixture models, weighted supervised contrastive learning have been combined into a fine-tuning phase following the pre-training. This paper provides an extensive empirical study showing that a preliminary contrastive learning step brings a significant gain in performance when using different loss functions: non-robust, robust, and early-learning regularized. Our experiments performed on standard benchmarks and real-world datasets demonstrate that: i) the contrastive pre-training increases the robustness of any loss function to noisy labels and ii) the additional fine-tuning phase can further improve accuracy but at the cost of additional complexity.
In this article, we propose the approach to procedural optimization of a neural network, based on the combination of information theory and braid theory. The network studied in the article implemented with the intersections between the braid strands, as well as simplified networks (a network with strands without intersections and a simple convolutional deep neural network), are used to solve various problems of multiclass image classification that allow us to analyze the comparative effectiveness of the proposed architecture. The simulation results showed BraidNet's comparative advantage in learning speed and classification accuracy.
The long-tailed distribution datasets poses great challenges for deep learning based classification models on how to handle the class imbalance problem. Existing solutions usually involve class-balacing strategies or transfer learing from head- to tail-classes or use two-stages learning strategy to re-train the classifier. However, the existing methods are difficult to solve the low quality problem when images are obtained by SAR. To address this problem, we establish a novel three-stages training strategy, which has excellent results for processing SAR image datasets with long-tailed distribution. Specifically, we divide training procedure into three stages. The first stage is to use all kinds of images for rough-training, so as to get the rough-training model with rich content. The second stage is to make the rough model learn the feature expression by using the residual dataset with the class 0 removed. The third stage is to fine tune the model using class-balanced datasets with all 10 classes (including the overall model fine tuning and classifier re-optimization). Through this new training strategy, we only use the information of SAR image dataset and the network model with very small parameters to achieve the top 1 accuracy of 22.34 in development phase.
Electromyography (EMG) data has been extensively adopted as an intuitive interface for instructing human-robot collaboration. A major challenge of the real-time detection of human grasp intent is the identification of dynamic EMG from hand movements. Previous studies mainly implemented steady-state EMG classification with a small number of grasp patterns on dynamic situations, which are insufficient to generate differentiated control regarding the muscular activity variation in practice. In order to better detect dynamic movements, more EMG variability could be integrated into the model. However, only limited research were concentrated on such detection of dynamic grasp motions, and most existing assessments on non-static EMG classification either require supervised ground-truth timestamps of the movement status, or only contain limited kinematic variations. In this study, we propose a framework for classifying dynamic EMG signals into gestures, and examine the impact of different movement phases, using an unsupervised method to segment and label the action transitions. We collected and utilized data from large gesture vocabularies with multiple dynamic actions to encode the transitions from one grasp intent to another based on common sequences of the grasp movements. The classifier for identifying the gesture label was constructed afterwards based on the dynamic EMG signal, with no supervised annotation of kinematic movements required. Finally, we evaluated the performances of several training strategies using EMG data from different movement phases, and explored the information revealed from each phase. All experiments were evaluated in a real-time style with the performance transitions over time presented.
Building on the success of the ADReSS Challenge at Interspeech 2020, which attracted the participation of 34 teams from across the world, the ADReSSo Challenge targets three difficult automatic prediction problems of societal and medical relevance, namely: detection of Alzheimer's Dementia, inference of cognitive testing scores, and prediction of cognitive decline. This paper presents these prediction tasks in detail, describes the datasets used, and reports the results of the baseline classification and regression models we developed for each task. A combination of acoustic and linguistic features extracted directly from audio recordings, without human intervention, yielded a baseline accuracy of 78.87% for the AD classification task, an MMSE prediction root mean squared (RMSE) error of 5.28, and 68.75% accuracy for the cognitive decline prediction task.
The tabla is a unique percussion instrument due to the combined harmonic and percussive nature of its timbre, and the contrasting harmonic frequency ranges of its two drums. This allows a tabla player to uniquely emphasize parts of the rhythmic cycle (theka) in order to mark the salient positions. An analysis of the loudness dynamics and timing deviations at various cycle positions is an important part of musicological studies on the expressivity in tabla accompaniment. To achieve this at a corpus-level, and not restrict it to the few recordings that manual annotation can afford, it is helpful to have access to an automatic tabla transcription system. Although a few systems have been built by training models on labeled tabla strokes, the achieved accuracy does not necessarily carry over to unseen instruments. In this article, we report our work towards building an instrument-independent stroke classification system for accompaniment tabla based on the more easily available tabla solo audio tracks. We present acoustic features that capture the distinctive characteristics of tabla strokes and build an automatic system to predict the label as one of a reduced, but musicologically motivated, target set of four stroke categories. To address the lack of sufficient labeled training data, we turn to common data augmentation methods and find the use of pitch-shifting based augmentation to be most promising. We then analyse the important features and highlight the problem of their instrument-dependence while motivating the use of more task-specific data augmentation strategies to improve the diversity of training data.
Adversarial examples are helpful for analyzing and improving the robustness of text classifiers. Generating high-quality adversarial examples is a challenging task as it requires generating fluent adversarial sentences that are semantically similar to the original sentences and preserve the original labels, while causing the classifier to misclassify them. Existing methods prioritize misclassification by maximizing each perturbation's effectiveness at misleading a text classifier; thus, the generated adversarial examples fall short in terms of fluency and similarity. In this paper, we propose a rewrite and rollback (R&R) framework for adversarial attack. It improves the quality of adversarial examples by optimizing a critique score which combines the fluency, similarity, and misclassification metrics. R&R generates high-quality adversarial examples by allowing exploration of perturbations that do not have immediate impact on the misclassification metric but can improve fluency and similarity metrics. We evaluate our method on 5 representative datasets and 3 classifier architectures. Our method outperforms current state-of-the-art in attack success rate by +16.2%, +12.8%, and +14.0% on the classifiers respectively. Code is available at https://github.com/DAI-Lab/fibber
Deep learning techniques have achieved great success in many fields, while at the same time deep learning models are getting more complex and expensive to compute. It severely hinders the wide applications of these models. In order to alleviate this problem, model distillation emerges as an effective means to compress a large model into a smaller one without a significant drop in accuracy. In this paper, we study a related but orthogonal issue, data distillation, which aims to distill the knowledge from a large training dataset down to a smaller and synthetic one. It has the potential to address the large and growing neural network training problem based on the small dataset. We develop a novel data distillation method for text classification. We evaluate our method on eight benchmark datasets. The results that the distilled data with the size of 0.1% of the original text data achieves approximately 90% performance of the original is rather impressive.
We present a Three-level Hierarchical Transformer Network (3-level-HTN) for modeling long-term dependencies across clinical notes for the purpose of patient-level prediction. The network is equipped with three levels of Transformer-based encoders to learn progressively from words to sentences, sentences to notes, and finally notes to patients. The first level from word to sentence directly applies a pre-trained BERT model as a fully trainable component. While the second and third levels both implement a stack of transformer-based encoders, before the final patient representation is fed into a classification layer for clinical predictions. Compared to conventional BERT models, our model increases the maximum input length from 512 tokens to much longer sequences that are appropriate for modeling large numbers of clinical notes. We empirically examine different hyper-parameters to identify an optimal trade-off given computational resource limits. Our experiment results on the MIMIC-III dataset for different prediction tasks demonstrate that the proposed Hierarchical Transformer Network outperforms previous state-of-the-art models, including but not limited to BigBird.
Tamil is a Dravidian language that is commonly used and spoken in the southern part of Asia. In the era of social media, memes have been a fun moment in the day-to-day life of people. Here, we try to analyze the true meaning of Tamil memes by categorizing them as troll and non-troll. We propose an ingenious model comprising of a transformer-transformer architecture that tries to attain state-of-the-art by using attention as its main component. The dataset consists of troll and non-troll images with their captions as text. The task is a binary classification task. The objective of the model is to pay more attention to the extracted features and to ignore the noise in both images and text.
The evolution of the Internet has increased the amount of information that is expressed by people on different platforms. This information can be product reviews, discussions on forums, or social media platforms. Accessibility of these opinions and peoples feelings open the door to opinion mining and sentiment analysis. As language and speech technologies become more advanced, many languages have been used and the best models have been obtained. However, due to linguistic diversity and lack of datasets, African languages have been left behind. In this study, by using the current state-of-the-art model, multilingual BERT, we perform sentiment classification on Swahili datasets. The data was created by extracting and annotating 8.2k reviews and comments on different social media platforms and the ISEAR emotion dataset. The data were classified as either positive or negative. The model was fine-tuned and achieve the best accuracy of 87.59%.
Although research on emotion classification has significantly progressed in high-resource languages, it is still infancy for resource-constrained languages like Bengali. However, unavailability of necessary language processing tools and deficiency of benchmark corpora makes the emotion classification task in Bengali more challenging and complicated. This work proposes a transformer-based technique to classify the Bengali text into one of the six basic emotions: anger, fear, disgust, sadness, joy, and surprise. A Bengali emotion corpus consists of 6243 texts is developed for the classification task. Experimentation carried out using various machine learning (LR, RF, MNB, SVM), deep neural networks (CNN, BiLSTM, CNN+BiLSTM) and transformer (Bangla-BERT, m-BERT, XLM-R) based approaches. Experimental outcomes indicate that XLM-R outdoes all other techniques by achieving the highest weighted $f_1$-score of $69.73\%$ on the test data. The dataset is publicly available at https://github.com/omar-sharif03/NAACL-SRW-2021.
We explore Few-Shot Learning (FSL) for Relation Classification (RC). Focusing on the realistic scenario of FSL, in which a test instance might not belong to any of the target categories (none-of-the-above, aka NOTA), we first revisit the recent popular dataset structure for FSL, pointing out its unrealistic data distribution. To remedy this, we propose a novel methodology for deriving more realistic few-shot test data from available datasets for supervised RC, and apply it to the TACRED dataset. This yields a new challenging benchmark for FSL RC, on which state of the art models show poor performance. Next, we analyze classification schemes within the popular embedding-based nearest-neighbor approach for FSL, with respect to constraints they impose on the embedding space. Triggered by this analysis we propose a novel classification scheme, in which the NOTA category is represented as learned vectors, shown empirically to be an appealing option for FSL.
Multi-domain sentiment classification deals with the scenario where labeled data exists for multiple domains but insufficient for training effective sentiment classifiers that work across domains. Thus, fully exploiting sentiment knowledge shared across domains is crucial for real world applications. While many existing works try to extract domain-invariant features in high-dimensional space, such models fail to explicitly distinguish between shared and private features at text-level, which to some extent lacks interpretablity. Based on the assumption that removing domain-related tokens from texts would help improve their domain-invariance, we instead first transform original sentences to be domain-agnostic. To this end, we propose the BertMasker network which explicitly masks domain-related words from texts, learns domain-invariant sentiment features from these domain-agnostic texts, and uses those masked words to form domain-aware sentence representations. Empirical experiments on a well-adopted multiple domain sentiment classification dataset demonstrate the effectiveness of our proposed model on both multi-domain sentiment classification and cross-domain settings, by increasing the accuracy by 0.94% and 1.8% respectively. Further analysis on masking proves that removing those domain-related and sentiment irrelevant tokens decreases texts' domain distinction, resulting in the performance degradation of a BERT-based domain classifier by over 12%.
We propose the neural string edit distance model for string-pair matching and string transduction based on learnable string edit distance. We modify the original expectation-maximization learned edit distance algorithm into a differentiable loss function, allowing us to integrate it into a neural network providing a contextual representation of the input. We evaluate on cognate detection, transliteration, and grapheme-to-phoneme conversion, and show that we can trade off between performance and interpretability in a single framework. Using contextual representations, which are difficult to interpret, we match the performance of state-of-the-art string-pair matching models. Using static embeddings and a slightly different loss function, we force interpretability, at the expense of an accuracy drop.
The earth system is exceedingly complex and often chaotic in nature, making prediction incredibly challenging: we cannot expect to make perfect predictions all of the time. Instead, we look for specific states of the system that lead to more predictable behavior than others, often termed "forecasts of opportunity." When these opportunities are not present, scientists need prediction systems that are capable of saying "I don't know." We introduce a novel loss function, termed the "NotWrong loss", that allows neural networks to identify forecasts of opportunity for classification problems. The NotWrong loss introduces an abstention class that allows the network to identify the more confident samples and abstain (say "I don't know") on the less confident samples. The abstention loss is designed to abstain on a user-defined fraction of the samples via a PID controller. Unlike many machine learning methods used to reject samples post-training, the NotWrong loss is applied during training to preferentially learn from the more confident samples. We show that the NotWrong loss outperforms other existing loss functions for multiple climate use cases. The implementation of the proposed loss function is straightforward in most network architectures designed for classification as it only requires the addition of an abstention class to the output layer and modification of the loss function.
We analyze the properties of adversarial training for learning adversarially robust halfspaces in the presence of agnostic label noise. Denoting $\mathsf{OPT}_{p,r}$ as the best robust classification error achieved by a halfspace that is robust to perturbations of $\ell_{p}$ balls of radius $r$, we show that adversarial training on the standard binary cross-entropy loss yields adversarially robust halfspaces up to (robust) classification error $\tilde O(\sqrt{\mathsf{OPT}_{2,r}})$ for $p=2$, and $\tilde O(d^{1/4} \sqrt{\mathsf{OPT}_{\infty, r}} + d^{1/2} \mathsf{OPT}_{\infty,r})$ when $p=\infty$. Our results hold for distributions satisfying anti-concentration properties enjoyed by log-concave isotropic distributions among others. We additionally show that if one instead uses a nonconvex sigmoidal loss, adversarial training yields halfspaces with an improved robust classification error of $O(\mathsf{OPT}_{2,r})$ for $p=2$, and $O(d^{1/4}\mathsf{OPT}_{\infty, r})$ when $p=\infty$. To the best of our knowledge, this is the first work to show that adversarial training provably yields robust classifiers in the presence of noise.
Graph neural networks have achieved state-of-the-art accuracy for graph node classification. However, GNNs are difficult to scale to large graphs, for example frequently encountering out-of-memory errors on even moderate size graphs. Recent works have sought to address this problem using a two-stage approach, which first aggregates data along graph edges, then trains a classifier without using additional graph information. These methods can run on much larger graphs and are orders of magnitude faster than GNNs, but achieve lower classification accuracy. We propose a novel two-stage algorithm based on a simple but effective observation: we should first train a classifier then aggregate, rather than the other way around. We show our algorithm is faster and can handle larger graphs than existing two-stage algorithms, while achieving comparable or higher accuracy than popular GNNs. We also present a theoretical basis to explain our algorithm's improved accuracy, by giving a synthetic nonlinear dataset in which performing aggregation before classification actually decreases accuracy compared to doing classification alone, while our classify then aggregate approach substantially improves accuracy compared to classification alone.
Objective: Traumatic brain injury can be caused by head impacts, but many brain injury risk estimation models are not equally accurate across the variety of impacts that patients may undergo and the characteristics of different types of impacts are not well studied. We investigated the spectral characteristics of different head impact types with kinematics classification. Methods: Data was analyzed from 3,262 head impacts from lab reconstruction, American football, mixed martial arts, and publicly available car crash data. A random forest classifier with spectral densities of linear acceleration and angular velocity was built to classify head impact types (e.g., football, car crash, mixed martial arts). To test the classifier robustness, another 271 lab-reconstructed impacts were obtained from 5 other instrumented mouthguards. Finally, with the classifier, type-specific, nearest-neighbor regression models were built for brain strain. Results: The classifier reached a median accuracy of 96% over 1,000 random partitions of training and test sets. The most important features in the classification included both low-frequency and high-frequency features, both linear acceleration features and angular velocity features. Different head impact types had different distributions of spectral densities in low-frequency and high-frequency ranges (e.g., the spectral densities of MMA impacts were higher in high-frequency range than in the low-frequency range). The type-specific regression showed a generally higher R^2-value than baseline models without classification. Conclusion: The machine-learning-based classifier enables a better understanding of the impact kinematics spectral density in different sports, and it can be applied to evaluate the quality of impact-simulation systems and on-field data augmentation.