Prior research notes that BERT's computational cost grows quadratically with sequence length thus leading to longer training times, higher GPU memory constraints and carbon emissions. While recent work seeks to address these scalability issues at pre-training, these issues are also prominent in fine-tuning especially for long sequence tasks like document classification. Our work thus focuses on optimizing the computational cost of fine-tuning for document classification. We achieve this by complementary learning of both topic and language models in a unified framework, named TopicBERT. This significantly reduces the number of self-attention operations - a main performance bottleneck. Consequently, our model achieves a 1.4x ($\sim40\%$) speedup with $\sim40\%$ reduction in $CO_2$ emission while retaining $99.9\%$ performance over 5 datasets.
Non-Intrusive Load Monitoring (NILM) aims to predict the status or consumption of domestic appliances in a household only by knowing the aggregated power load. NILM can be formulated as regression problem or most often as a classification problem. Most datasets gathered by smart meters allow to define naturally a regression problem, but the corresponding classification problem is a derived one, since it requires a conversion from the power signal to the status of each device by a thresholding method. We treat three different thresholding methods to perform this task, discussing their differences on various devices from the UK-DALE dataset. We analyze the performance of deep learning state-of-the-art architectures on both the regression and classification problems, introducing criteria to select the most convenient thresholding method.
Graph Neural Networks (GNNs) are effective in many applications. Still, there is a limited understanding of the effect of common graph structures on the learning process of GNNs. In this work, we systematically study the impact of community structure on the performance of GNNs in semi-supervised node classification on graphs. Following an ablation study on six datasets, we measure the performance of GNNs on the original graphs, and the change in performance in the presence and the absence of community structure. Our results suggest that communities typically have a major impact on the learning process and classification performance. For example, in cases where the majority of nodes from one community share a single classification label, breaking up community structure results in a significant performance drop. On the other hand, for cases where labels show low correlation with communities, we find that the graph structure is rather irrelevant to the learning process, and a feature-only baseline becomes hard to beat. With our work, we provide deeper insights in the abilities and limitations of GNNs, including a set of general guidelines for model selection based on the graph structure.
Contrastive learning (CL) has been successful as a powerful representation learning method. In this work we propose CLIM: Contrastive Learning with mutual Information Maximization, to explore the potential of CL on cross-domain sentiment classification. To the best of our knowledge, CLIM is the first to adopt contrastive learning for natural language processing (NLP) tasks across domains. Due to scarcity of labels on the target domain, we introduce mutual information maximization (MIM) apart from CL to exploit the features that best support the final prediction. Furthermore, MIM is able to maintain a relatively balanced distribution of the model's prediction, and enlarges the margin between classes on the target domain. The larger margin increases our model's robustness and enables the same classifier to be optimal across domains. Consequently, we achieve new state-of-the-art results on the Amazon-review dataset as well as the airlines dataset, showing the efficacy of our proposed method CLIM.
In this paper, we provide a theory of using graph neural networks (GNNs) for multi-node representation learning (where we are interested in learning a representation for a set of more than one node, such as link). We know that GNN is designed to learn single-node representations. When we want to learn a node set representation involving multiple nodes, a common practice in previous works is to directly aggregate the single-node representations obtained by a GNN into a joint node set representation. In this paper, we show a fundamental constraint of such an approach, namely the inability to capture the dependence between nodes in the node set, and argue that directly aggregating individual node representations does not lead to an effective joint representation for multiple nodes. Then, we notice that a few previous successful works for multi-node representation learning, including SEAL, Distance Encoding, and ID-GNN, all used node labeling. These methods first label nodes in the graph according to their relationships with the target node set before applying a GNN. Then, the node representations obtained in the labeled graph are aggregated into a node set representation. By investigating their inner mechanisms, we unify these node labeling techniques into a single and most general form -- labeling trick. We prove that with labeling trick a sufficiently expressive GNN learns the most expressive node set representations, thus in principle solves any joint learning tasks over node sets. Experiments on one important two-node representation learning task, link prediction, verified our theory. Our work explains the superior performance of previous node-labeling-based methods, and establishes a theoretical foundation of using GNNs for multi-node representation learning.
For future learning systems, incremental learning is desirable because it allows for: efficient resource usage by eliminating the need to retrain from scratch at the arrival of new data; reduced memory usage by preventing or limiting the amount of data required to be stored -- also important when privacy limitations are imposed; and learning that more closely resembles human learning. The main challenge for incremental learning is catastrophic forgetting, which refers to the precipitous drop in performance on previously learned tasks after learning a new one. Incremental learning of deep neural networks has seen explosive growth in recent years. Initial work focused on task-incremental learning, where a task-ID is provided at inference time. Recently, we have seen a shift towards class-incremental learning where the learner must discriminate at inference time between all classes seen in previous tasks without recourse to a task-ID. In this paper, we provide a complete survey of existing class-incremental learning methods for image classification, and in particular, we perform an extensive experimental evaluation on thirteen class-incremental methods. We consider several new experimental scenarios, including a comparison of class-incremental methods on multiple large-scale image classification datasets, an investigation into small and large domain shifts, and a comparison of various network architectures.
Emotions recognition is commonly employed for health assessment. However, the typical metric for evaluation in therapy is based on patient-doctor appraisal. This process can fall into the issue of subjectivity, while also requiring healthcare professionals to deal with copious amounts of information. Thus, machine learning algorithms can be a useful tool for the classification of emotions. While several models have been developed in this domain, there is a lack of userfriendly representations of the emotion classification systems for therapy. We propose a tool which enables users to take speech samples and identify a range of emotions (happy, sad, angry, surprised, neutral, clam, disgust, and fear) from audio elements through a machine learning model. The dashboard is designed based on local therapists' needs for intuitive representations of speech data in order to gain insights and informative analyses of their sessions with their patients.
Utterance classification performance in low-resource dialogue systems is constrained by an inevitably high degree of data imbalance in class labels. We present a new end-to-end pairwise learning framework that is designed specifically to tackle this phenomenon by inducing a few-shot classification capability in the utterance representations and augmenting data through an interpolation of utterance representations. Our approach is a general purpose training methodology, agnostic to the neural architecture used for encoding utterances. We show significant improvements in macro-F1 score over standard cross-entropy training for three different neural architectures, demonstrating improvements on a Virtual Patient dialogue dataset as well as a low-resourced emulation of the Switchboard dialogue act classification dataset.
The road is vital for many aspects of life, and road maintenance is crucial for human safety. One of the critical tasks to allow timely repair of road damages is to quickly and efficiently detect and classify them. This work details the strategies and experiments evaluated for these tasks. Specifically, we evaluate Detectron2's implementation of Faster R-CNN using different base models and configurations. We also experiment with these approaches using the Global Road Damage Detection Challenge 2020, A Track in the IEEE Big Data 2020 Big Data Cup Challenge dataset. The results show that the X101-FPN base model for Faster R-CNN with Detectron2's default configurations are efficient and general enough to be transferable to different countries in this challenge. This approach results in F1 scores of 51.0% and 51.4% for the test1 and test2 sets of the challenge, respectively. Though the visualizations show good prediction results, the F1 scores are low. Therefore, we also evaluate the prediction results against the existing annotations and discover some discrepancies. Thus, we also suggest strategies to improve the labeling process for this dataset.
We present MedMNIST, a collection of 10 pre-processed medical open datasets. MedMNIST is standardized to perform classification tasks on lightweight 28x28 images, which requires no background knowledge. Covering the primary data modalities in medical image analysis, it is diverse on data scale (from 100 to 100,000) and tasks (binary/multi-class, ordinal regression and multi-label). MedMNIST could be used for educational purpose, rapid prototyping, multi-modal machine learning or AutoML in medical image analysis. Moreover, MedMNIST Classification Decathlon is designed to benchmark AutoML algorithms on all 10 datasets; We have compared several baseline methods, including open-source or commercial AutoML tools. The datasets, evaluation code and baseline methods for MedMNIST are publicly available at https://medmnist.github.io/.
Semi-supervised learning via learning from limited quantities of labeled data has been investigated as an alternative to supervised counterparts. Maximizing knowledge gains from copious unlabeled data benefit semi-supervised learning settings. Moreover, learning multiple tasks within the same model further improves model generalizability. We propose a novel multitask learning model, namely MultiMix, which jointly learns disease classification and anatomical segmentation in a sparingly supervised manner, while preserving explainability through bridge saliency between the two tasks. Our extensive experimentation with varied quantities of labeled data in the training sets justify the effectiveness of our multitasking model for the classification of pneumonia and segmentation of lungs from chest X-ray images. Moreover, both in-domain and cross-domain evaluations across the tasks further showcase the potential of our model to adapt to challenging generalization scenarios.
Previous work on bridging anaphora recognition (Hou et al., 2013a) casts the problem as a subtask of learning fine-grained information status (IS). However, these systems heavily depend on many hand-crafted linguistic features. In this paper, we propose a simple discourse context-aware BERT model for fine-grained IS classification. On the ISNotes corpus (Markert et al., 2012), our model achieves new state-of-the-art performance on fine-grained IS classification, obtaining a 4.8 absolute overall accuracy improvement compared to Hou et al. (2013a). More importantly, we also show an improvement of 10.5 F1 points for bridging anaphora recognition without using any complex hand-crafted semantic features designed for capturing the bridging phenomenon. We further analyze the trained model and find that the most attended signals for each IS category correspond well to linguistic notions of information status.
A plethora of research in the literature shows how human eye fixation pattern varies depending on different factors, including genetics, age, social functioning, cognitive functioning, and so on. Analysis of these variations in visual attention has already elicited two potential research avenues: 1) determining the physiological or psychological state of the subject and 2) predicting the tasks associated with the act of viewing from the recorded eye-fixation data. To this end, this paper proposes a visual saliency based novel feature extraction method for automatic and quantitative classification of eye-tracking data, which is applicable to both of the research directions. Instead of directly extracting features from the fixation data, this method employs several well-known computational models of visual attention to predict eye fixation locations as saliency maps. Comparing the saliency amplitudes, similarity and dissimilarity of saliency maps with the corresponding eye fixations maps gives an extra dimension of information which is effectively utilized to generate discriminative features to classify the eye-tracking data. Extensive experimentation using Saliency4ASD, Age Prediction, and Visual Perceptual Task dataset show that our saliency-based feature can achieve superior performance, outperforming the previous state-of-the-art methods by a considerable margin. Moreover, unlike the existing application-specific solutions, our method demonstrates performance improvement across three distinct problems from the real-life domain: Autism Spectrum Disorder screening, toddler age prediction, and human visual perceptual task classification, providing a general paradigm that utilizes the extra-information inherent in saliency maps for a more accurate classification.
The data imbalance problem is a frequent bottleneck in the classification performance of neural networks. In this paper, we propose a novel supervised discriminative feature generation (DFG) method for a minority class dataset. DFG is based on the modified structure of a generative adversarial network consisting of four independent networks: generator, discriminator, feature extractor, and classifier. To augment the selected discriminative features of the minority class data by adopting an attention mechanism, the generator for the class-imbalanced target task is trained, and the feature extractor and classifier are regularized using the pre-trained features from a large source data. The experimental results show that the DFG generator enhances the augmentation of the label-preserved and diverse features, and the classification results are significantly improved on the target task. The feature generation model can contribute greatly to the development of data augmentation methods through discriminative feature generation and supervised attention methods.
In this paper, we explore text classification with extremely weak supervision, i.e., only relying on the surface text of class names. This is a more challenging setting than the seed-driven weak supervision, which allows a few seed words per class. We opt to attack this problem from a representation learning perspective -- ideal document representations should lead to nearly the same results between clustering and the desired classification. In particular, one can classify the same corpus differently (e.g., based on topics and locations), so document representations should be adaptive to the given class names. We propose a novel framework X-Class to realize the adaptive representations. Specifically, we first estimate class representations by incrementally adding the most similar word to each class until inconsistency arises. Following a tailored mixture of class attention mechanisms, we obtain the document representation via a weighted average of contextualized word representations. With the prior of each document assigned to its nearest class, we then cluster and align the documents to classes. Finally, we pick the most confident documents from each cluster to train a text classifier. Extensive experiments demonstrate that X-Class can rival and even outperform seed-driven weakly supervised methods on 7 benchmark datasets. Our dataset and code are released at https://github.com/ZihanWangKi/XClass/ .
A recent approach for few-shot text classification is to convert textual inputs to cloze questions that contain some form of task description, process them with a pretrained language model and map the predicted words to labels. Manually defining this mapping between words and labels requires both domain expertise and an understanding of the language model's abilities. To mitigate this issue, we devise an approach that automatically finds such a mapping given small amounts of training data. For a number of tasks, the mapping found by our approach performs almost as well as hand-crafted label-to-word mappings.
Deep metric learning plays a key role in various machine learning tasks. Most of the previous works have been confined to sampling from a mini-batch, which cannot precisely characterize the global geometry of the embedding space. Although researchers have developed proxy- and classification-based methods to tackle the sampling issue, those methods inevitably incur a redundant computational cost. In this paper, we propose a novel Proxy-based deep Graph Metric Learning (ProxyGML) approach from the perspective of graph classification, which uses fewer proxies yet achieves better comprehensive performance. Specifically, multiple global proxies are leveraged to collectively approximate the original data points for each class. To efficiently capture local neighbor relationships, a small number of such proxies are adaptively selected to construct similarity subgraphs between these proxies and each data point. Further, we design a novel reverse label propagation algorithm, by which the neighbor relationships are adjusted according to ground-truth labels, so that a discriminative metric space can be learned during the process of subgraph classification. Extensive experiments carried out on widely-used CUB-200-2011, Cars196, and Stanford Online Products datasets demonstrate the superiority of the proposed ProxyGML over the state-of-the-art methods in terms of both effectiveness and efficiency. The source code is publicly available at https://github.com/YuehuaZhu/ProxyGML.
To deploy machine learning-based algorithms for real-time applications with strict latency constraints, we consider an edge-computing setting where a subset of inputs are offloaded to the edge for processing by an accurate but resource-intensive model, and the rest are processed only by a less-accurate model on the device itself. Both models have computational costs that match available compute resources, and process inputs with low-latency. But offloading incurs network delays, and to manage these delays to meet application deadlines, we use a token bucket to constrain the average rate and burst length of transmissions from the device. We introduce a Markov Decision Process-based framework to make offload decisions under these constraints, based on the local model's confidence and the token bucket state, with the goal of minimizing a specified error measure for the application. Beyond isolated decisions for individual devices, we also propose approaches to allow multiple devices connected to the same access switch to share their bursting allocation. We evaluate and analyze the policies derived using our framework on the standard ImageNet image classification benchmark.
Videos are a commonly-used type of content in learning during Web search. Many e-learning platforms provide quality content, but sometimes educational videos are long and cover many topics. Humans are good in extracting important sections from videos, but it remains a significant challenge for computers. In this paper, we address the problem of assigning importance scores to video segments, that is how much information they contain with respect to the overall topic of an educational video. We present an annotation tool and a new dataset of annotated educational videos collected from popular online learning platforms. Moreover, we propose a multimodal neural architecture that utilizes state-of-the-art audio, visual and textual features. Our experiments investigate the impact of visual and temporal information, as well as the combination of multimodal features on importance prediction.
The inclusion of spatial information into spectral classifiers for fine-resolution hyperspectral imagery has led to significant improvements in terms of classification performance. The task of spectral-spatial hyperspectral image classification has remained challenging because of high intraclass spectrum variability and low interclass spectral variability. This fact has made the extraction of spatial information highly active. In this work, a novel hyperspectral image classification framework using the fusion of dual spatial information is proposed, in which the dual spatial information is built by both exploiting pre-processing feature extraction and post-processing spatial optimization. In the feature extraction stage, an adaptive texture smoothing method is proposed to construct the structural profile (SP), which makes it possible to precisely extract discriminative features from hyperspectral images. The SP extraction method is used here for the first time in the remote sensing community. Then, the extracted SP is fed into a spectral classifier. In the spatial optimization stage, a pixel-level classifier is used to obtain the class probability followed by an extended random walker-based spatial optimization technique. Finally, a decision fusion rule is utilized to fuse the class probabilities obtained by the two different stages. Experiments performed on three data sets from different scenes illustrate that the proposed method can outperform other state-of-the-art classification techniques. In addition, the proposed feature extraction method, i.e., SP, can effectively improve the discrimination between different land covers.