This paper describes an approach to the facial action units detections. The involved action units (AU) include AU1 (Inner Brow Raiser), AU2 (Outer Brow Raiser), AU4 (Brow Lowerer), AU6 (Cheek Raise), AU12 (Lip Corner Puller), AU15 (Lip Corner Depressor), AU20 (Lip Stretcher), and AU25 (Lip Part). Our work relies on the dataset released by the FG-2020 Competition: Affective Behavior Analysis In-the-Wild (ABAW). The proposed method consists of the data preprocessing, the feature extraction and the AU classification. The data preprocessing includes the detection of face texture and landmarks. The texture static and landmark dynamic features are extracted through neural networks and then fused as the feature latent representation. Finally, the fused feature is taken as the initial hidden state of a recurrent neural network with a trainable lookup AU table. The output of the RNN is the results of AU classification. The detected accuracy is evaluated with 0.5$\times$accuracy + 0.5$\times$F1. Our method achieve 0.56 with the validation data that is specified by the organization committee.
In this paper, we present a novel algorithm, FastWordBug, to efficiently generate small text perturbations in a black-box setting that forces a sentiment analysis or text classification mode to make an incorrect prediction. By combining the part of speech attributes of words, we propose a scoring method that can quickly identify important words that affect text classification. We evaluate FastWordBug on three real-world text datasets and two state-of-the-art machine learning models under black-box setting. The results show that our method can significantly reduce the accuracy of the model, and at the same time, we can call the model as little as possible, with the highest attack efficiency. We also attack two popular real-world cloud services of NLP, and the results show that our method works as well.
Text classification models, especially neural networks based models, have reached very high accuracy on many popular benchmark datasets. Yet, such models when deployed in real world applications, tend to perform badly. The primary reason is that these models are not tested against sufficient real world natural data. Based on the application users, the vocabulary and the style of the model's input may greatly vary. This emphasizes the need for a model agnostic test dataset, which consists of various corruptions that are natural to appear in the wild. Models trained and tested on such benchmark datasets, will be more robust against real world data. However, such data sets are not easily available. In this work, we address this problem, by extending the benchmark datasets along naturally occurring corruptions such as Spelling Errors, Text Noise and Synonyms and making them publicly available. Through extensive experiments, we compare random and targeted corruption strategies using Local Interpretable Model-Agnostic Explanations(LIME). We report the vulnerabilities in two popular text classification models along these corruptions and also find that targeted corruptions can expose vulnerabilities of a model better than random choices in most cases.
Spike-based neuromorphic hardware promises to reduce the energy consumption of image classification and other deep learning applications, particularly on mobile phones or other edge devices. However, direct training of deep spiking neural networks is difficult, and previous methods for converting trained artificial neural networks to spiking neurons were inefficient because the neurons had to emit too many spikes. We show that a substantially more efficient conversion arises when one optimizes the spiking neuron model for that purpose, so that it not only matters for information transmission how many spikes a neuron emits, but also when it emits those spikes. This advances the accuracy that can be achieved for image classification with spiking neurons, and the resulting networks need on average just two spikes per neuron for classifying an image. In addition, our new conversion method improves latency and throughput of the resulting spiking networks.
In recent years, image manipulation is becoming increasingly more accessible, yielding more natural-looking images, owing to the modern tools in image processing and computer vision techniques. The task of the identification of forged images has become very challenging. Amongst different types of forgeries, the cases of Copy-Move forgery are increasing manifold, due to the difficulties involved to detect this tampering. To tackle such problems, publicly available datasets are insufficient. In this paper, we propose to create a synthetic forged dataset using deep semantic image inpainting and copy-move forgery algorithm. However, models trained on these datasets have a significant drop in performance when tested on more realistic data. To alleviate this problem, we use unsupervised domain adaptation networks to detect copy-move forgery in new domains by mapping the feature space from our synthetically generated dataset. Furthermore, we improvised the F1 score on CASIA and CoMoFoD dataset to 80.3% and 78.8%, respectively. Our approach can be helpful in those cases where the classification of data is unavailable.
Understanding on-road vehicle behaviour from a temporal sequence of sensor data is gaining in popularity. In this paper, we propose a pipeline for understanding vehicle behaviour from a monocular image sequence or video. A monocular sequence along with scene semantics, optical flow and object labels are used to get spatial information about the object (vehicle) of interest and other objects (semantically contiguous set of locations) in the scene. This spatial information is encoded by a Multi-Relational Graph Convolutional Network (MR-GCN), and a temporal sequence of such encodings is fed to a recurrent network to label vehicle behaviours. The proposed framework can classify a variety of vehicle behaviours to high fidelity on datasets that are diverse and include European, Chinese and Indian on-road scenes. The framework also provides for seamless transfer of models across datasets without entailing re-annotation, retraining and even fine-tuning. We show comparative performance gain over baseline Spatio-temporal classifiers and detail a variety of ablations to showcase the efficacy of the framework.
Chest X-ray scan is a most often used modality by radiologists to diagnose many chest related diseases in their initial stages. The proposed system aids the radiologists in making decision about the diseases found in the scans more efficiently. Our system combines the techniques of image processing for feature enhancement and deep learning for classification among diseases. We have used the ChestX-ray14 database in order to train our deep learning model on the 14 different labeled diseases found in it. The proposed research shows the significant improvement in the results by using wavelet transforms as pre-processing technique.
The recent success in implementing supervised learning to classify modulation types suggests that other problems akin to modulation classification would eventually benefit from that implementation. One of these problems is classifying the interference type added to a signal-of-interest, also known as interference classification. In this paper, we propose an interference classification method using a deep neural network. We generate five distinct types of interfering signals then use both the power-spectral density (PSD) and the cyclic spectrum of the received signal as input features to the network. The computer experiments reveal that using the received signal PSD outperforms using its cyclic spectrum in terms of accuracy. In addition, the same experiments show that the feed-forward networks yield better accuracy than classic methods. The proposed classifier aids the subsequent stage in the receiver chain with choosing the appropriate mitigation algorithm and also can coexist with modulation-classification methods to further improve the classifier accuracy.
Noninvasive brain-computer interface (BCI) decodes brain signals to understand user intention. Recent advances have been developed for the BCI-based drone control system as the demand for drone control increases. Especially, drone swarm control based on brain signals could provide various industries such as military service or industry disaster. This paper presents a prototype of a brain swarm interface system for a variety of scenarios using a visual imagery paradigm. We designed the experimental environment that could acquire brain signals under a drone swarm control simulator environment. Through the system, we collected the electroencephalogram (EEG) signals with respect to four different scenarios. Seven subjects participated in our experiment and evaluated classification performances using the basic machine learning algorithm. The grand average classification accuracy is higher than the chance level accuracy. Hence, we could confirm the feasibility of the drone swarm control system based on EEG signals for performing high-level tasks.
Real data collected from different applications that have additional topological structures and connection information are amenable to be represented as a weighted graph. Considering the node labeling problem, Graph Neural Networks (GNNs) is a powerful tool, which can mimic experts' decision on node labeling. GNNs combine node features, connection patterns, and graph structure by using a neural network to embed node information and pass it through edges in the graph. We want to identify the patterns in the input data used by the GNN model to make a decision and examine if the model works as we desire. However, due to the complex data representation and non-linear transformations, explaining decisions made by GNNs is challenging. In this work, we propose new graph features' explanation methods to identify the informative components and important node features. Besides, we propose a pipeline to identify the key factors used for node classification. We use four datasets (two synthetic and two real) to validate our methods. Our results demonstrate that our explanation approach can mimic data patterns used for node classification by human interpretation and disentangle different features in the graphs. Furthermore, our explanation methods can be used for understanding data, debugging GNN models, and examine model decisions.
Electromyography (EMG) signals have been successfully employed for driving prosthetic limbs of a single or double degree of freedom. This principle works by using the amplitude of the EMG signals to decide between one or two simpler movements. This method underperforms as compare to the contemporary advances done at the mechanical, electronics, and robotics end, and it lacks intuition. Recently, research on myoelectric control based on pattern recognition (PR) shows promising results with the aid of machine learning classifiers. Using the approach termed as, EMG-PR, EMG signals are divided into analysis windows, and features are extracted for each window. These features are then fed to the machine learning classifiers as input. By offering multiple class movements and intuitive control, this method has the potential to power an amputated subject to perform everyday life movements. In this paper, we investigate the effect of the analysis window and feature selection on classification accuracy of different hand and wrist movements using time-domain features. We show that effective data preprocessing and optimum feature selection helps to improve the classification accuracy of hand movements. We use publicly available hand and wrist gesture dataset of $40$ intact subjects for experimentation. Results computed using different classification algorithms show that the proposed preprocessing and features selection outperforms the baseline and achieve up to $98\%$ classification accuracy.
Many recent works on deep speaker embeddings train their feature extraction networks on large classification tasks, distinguishing between all speakers in a training set. Empirically, this has been shown to produce speaker-discriminative embeddings, even for unseen speakers. However, it is not clear that this is the optimal means of training embeddings that generalize well. This work proposes two approaches to learning embeddings, based on the notion of dropping classes during training. We demonstrate that both approaches can yield performance gains in speaker verification tasks. The first proposed method, DropClass, works via periodically dropping a random subset of classes from the training data and the output layer throughout training, resulting in a feature extractor trained on many different classification tasks. Combined with an additive angular margin loss, this method can yield a 7.9% relative improvement in equal error rate (EER) over a strong baseline on VoxCeleb. The second proposed method, DropAdapt, is a means of adapting a trained model to a set of enrolment speakers in an unsupervised manner. This is performed by fine-tuning a model on only those classes which produce high probability predictions when the enrolment speakers are used as input, again also dropping the relevant rows from the output layer. This method yields a large 13.2% relative improvement in EER on VoxCeleb. The code for this paper has been made publicly available.
A number of techniques for interpretability have been presented for deep learning in computer vision, typically with the goal of understanding what the networks have based their classification on. However, interpretability for deep video architectures is still in its infancy and we do not yet have a clear concept of how to decode spatiotemporal features. In this paper, we present a study comparing how 3D convolutional networks and convolutional LSTM networks learn features across temporally dependent frames. This is the first comparison of two video models that both convolve to learn spatial features but have principally different methods of modeling time. Additionally, we extend the concept of meaningful perturbation introduced by \cite{MeaningFulPert} to the temporal dimension, to identify the temporal part of a sequence most meaningful to the network for a classification decision. Our findings indicate that the 3D convolutional model concentrates on shorter events in the input sequence, and places its spatial focus on fewer, contiguous areas.
Domain-adapted sentiment classification refers to training on a labeled source domain to well infer document-level sentiment on an unlabeled target domain. Most existing relevant models involve a feature extractor and a sentiment classifier, where the feature extractor works towards learning domain-invariant features from both domains, and the sentiment classifier is trained only on the source domain to guide the feature extractor. As such, they lack a mechanism to use sentiment polarity lying in the target domain. To improve domain-adapted sentiment classification by learning sentiment from the target domain as well, we devise a novel deep adversarial mutual learning approach involving two groups of feature extractors, domain discriminators, sentiment classifiers, and label probers. The domain discriminators enable the feature extractors to obtain domain-invariant features. Meanwhile, the label prober in each group explores document sentiment polarity of the target domain through the sentiment prediction generated by the classifier in the peer group, and guides the learning of the feature extractor in its own group. The proposed approach achieves the mutual learning of the two groups in an end-to-end manner. Experiments on multiple public datasets indicate our method obtains the state-of-the-art performance, validating the effectiveness of mutual learning through label probers.
Over the past year, the emergence of transfer learning with large-scale language models (LM) has led to dramatic performance improvements across a broad range of natural language understanding tasks. However, the size and memory footprint of these large LMs makes them difficult to deploy in many scenarios (e.g. on mobile phones). Recent research points to knowledge distillation as a potential solution, showing that when training data for a given task is abundant, it is possible to distill a large (teacher) LM into a small task-specific (student) network with minimal loss of performance. However, when such data is scarce, there remains a significant performance gap between large pretrained LMs and smaller task-specific models, even when training via distillation. In this paper, we bridge this gap with a novel training approach, called generation-distillation, that leverages large finetuned LMs in two ways: (1) to generate new (unlabeled) training examples, and (2) to distill their knowledge into a small network using these examples. Across three low-resource text classification datsets, we achieve comparable performance to BERT while using 300x fewer parameters, and we outperform prior approaches to distillation for text classification while using 3x fewer parameters.
A brain-computer interface (BCI) provides a direct communication pathway between user and external devices. Electroencephalogram (EEG) motor imagery (MI) paradigm is widely used in non-invasive BCI to obtain encoded signals contained user intention of movement execution. However, EEG has intricate and non-stationary properties resulting in insufficient decoding performance. By imagining numerous movements of a single-arm, decoding performance can be improved without artificial command matching. In this study, we collected intuitive EEG data contained the nine different types of movements of a single-arm from 9 subjects. We propose an end-to-end role assigned convolutional neural network (ERA-CNN) which considers discriminative features of each upper limb region by adopting the principle of a hierarchical CNN architecture. The proposed model outperforms previous methods on 3-class, 5-class and two different types of 7-class classification tasks. Hence, we demonstrate the possibility of decoding user intention by using only EEG signals with robust performance using an ERA-CNN.
Graphs are versatile tools for representing structured data. As a result, a variety of machine learning methods have been studied for graph data analysis. Although many such learning methods depend on the measurement of differences between input graphs, defining an appropriate distance metric for graphs remains a controversial issue. Hence, we propose a supervised distance metric learning method for the graph classification problem. Our method, named interpretable graph metric learning (IGML), learns discriminative metrics in a subgraph-based feature space, which has a strong graph representation capability. By introducing a sparsity-inducing penalty on the weight of each subgraph, IGML can identify a small number of important subgraphs that can provide insight into the given classification task. Because our formulation has a large number of optimization variables, an efficient algorithm that uses pruning techniques based on safe screening and working set selection methods is also proposed. An important property of IGML is that solution optimality is guaranteed because the problem is formulated as a convex problem and our pruning strategies only discard unnecessary subgraphs. Furthermore, we show that IGML is also applicable to other structured data such as itemset and sequence data, and that it can incorporate vertex-label similarity by using a transportation-based subgraph feature. We empirically evaluate the computational efficiency and classification performance of IGML on several benchmark datasets and provide some illustrative examples of how IGML identifies important subgraphs from a given graph dataset.
Graph neural networks are currently leading the performance charts in learning-based molecule property prediction and classification. Computational chemistry has, therefore, become the a prominent testbed for generic graph neural networks, as well as for specialized message passing methods. In this work, we demonstrate that the replacement of the underlying networks with hypernetworks leads to a boost in performance, obtaining state of the art results in various benchmarks. A major difficulty in the application of hypernetworks is their lack of stability. We tackle this by combining the current message and the first message. A recent work has tackled the training instability of hypernetworks in the context of error correcting codes, by replacing the activation function of the message passing network with a low-order Taylor approximation of it. We demonstrate that our generic solution can replace this domain-specific solution.
This thesis combines audio-analysis with computer vision to approach Music Information Retrieval (MIR) tasks from a multi-modal perspective. This thesis focuses on the information provided by the visual layer of music videos and how it can be harnessed to augment and improve tasks of the MIR research domain. The main hypothesis of this work is based on the observation that certain expressive categories such as genre or theme can be recognized on the basis of the visual content alone, without the sound being heard. This leads to the hypothesis that there exists a visual language that is used to express mood or genre. In a further consequence it can be concluded that this visual information is music related and thus should be beneficial for the corresponding MIR tasks such as music genre classification or mood recognition. A series of comprehensive experiments and evaluations are conducted which are focused on the extraction of visual information and its application in different MIR tasks. A custom dataset is created, suitable to develop and test visual features which are able to represent music related information. Evaluations range from low-level visual features to high-level concepts retrieved by means of Deep Convolutional Neural Networks. Additionally, new visual features are introduced capturing rhythmic visual patterns. In all of these experiments the audio-based results serve as benchmark for the visual and audio-visual approaches. The experiments are conducted for three MIR tasks Artist Identification, Music Genre Classification and Cross-Genre Classification. Experiments show that an audio-visual approach harnessing high-level semantic information gained from visual concept detection, outperforms audio-only genre-classification accuracy by 16.43%.
The tiled convolutional neural network (tiled CNN) has been applied only to computer vision for learning invariances. We adjust its architecture to NLP to improve the extraction of the most salient features for sentiment analysis. Knowing that the major drawback of the tiled CNN in the NLP field is its inflexible filter structure, we propose a novel architecture called hybrid tiled CNN that applies a filter only on the words that appear in the similar contexts and on their neighbor words (a necessary step for preventing the loss of some n-grams). The experiments on the datasets of IMDB movie reviews and SemEval 2017 demonstrate the efficiency of the hybrid tiled CNN that performs better than both CNN and tiled CNN.