Augmented Inputs Research Articles

Advanced confidence methods in deep learning

The typical aim of classification tasks is to maximize the accuracy of the predicted label for a given input. This accuracy increases with the confidence, which is the maximal value of the output units, and when the accuracy equals confidence, calibration is achieved. Herein, several methods are proposed to enhance the accuracy of inputs with similar confidence, extending significantly beyond calibration. Using the first gap between the maximal and second maximal output values, the accuracy of the inputs with similar confidence is enhanced. The extension of the confidence or confidence gap to their minimal value among a set of augmented inputs further enhances the accuracy of inputs with similar confidence. Enhanced accuracies are demonstrated on EfficientNet-B0 trained on ImageNet and CIFAR-100, and VGG-16 trained on CIFAR-100. The results suggest improved applications for high-accuracy classification tasks that require manual operation for a given fraction of low-accuracy inputs.

Few-Shot Learning from Augmented Label-Uncertain Queries in Bongard-HOI

Detecting human-object interactions (HOI) in a few-shot setting remains a challenge. Existing meta-learning methods struggle to extract representative features for classification due to the limited data, while existing few-shot HOI models rely on HOI text labels for classification. Moreover, some query images may display visual similarity to those outside their class, such as similar backgrounds between different HOI classes. This makes learning more challenging, especially with limited samples. Bongard-HOI epitomizes this HOI few-shot problem, making it the benchmark we focus on in this paper. In our proposed method, we introduce novel label-uncertain query augmentation techniques to enhance the diversity of the query inputs, aiming to distinguish the positive HOI class from the negative ones. As these augmented inputs may or may not have the same class label as the original inputs, their class label is unknown. Those belonging to a different class become hard samples due to their visual similarity to the original ones. Additionally, we introduce a novel pseudo-label generation technique that enables a mean teacher model to learn from the augmented label-uncertain inputs. We propose to augment the negative support set for the student model to enrich the semantic information, fostering diversity that challenges and enhances the student’s learning. Experimental results demonstrate that our method sets a new state-of-the-art (SOTA) performance by achieving 68.74% accuracy on the Bongard-HOI benchmark, a significant improvement over the existing SOTA of 66.59%. In our evaluation on HICO-FS, a more general few-shot recognition dataset, our method achieves 73.27% accuracy, outperforming the previous SOTA of 71.20% in the 5- way 5-shot task.

Augmented inputs for surveillance re-identification

Augmented inputs for surveillance re-identification

Adversarial Word Dilution as Text Data Augmentation in Low-Resource Regime

Data augmentation is widely used in text classification, especially in the low-resource regime where a few examples for each class are available during training. Despite the success, generating data augmentations as hard positive examples that may increase their effectiveness is under-explored. This paper proposes an Adversarial Word Dilution (AWD) method that can generate hard positive examples as text data augmentations to train the low-resource text classification model efficiently. Our idea of augmenting the text data is to dilute the embedding of strong positive words by weighted mixing with unknown-word embedding, making the augmented inputs hard to be recognized as positive by the classification model. We adversarially learn the dilution weights through a constrained min-max optimization process with the guidance of the labels. Empirical studies on three benchmark datasets show that AWD can generate more effective data augmentations and outperform the state-of-the-art text data augmentation methods. The additional analysis demonstrates that the data augmentations generated by AWD are interpretable and can flexibly extend to new examples without further training.

JCBIE: a joint continual learning neural network for biomedical information extraction

Extracting knowledge from heterogeneous data sources is fundamental for the construction of structured biomedical knowledge graphs (BKGs), where entities and relations are represented as nodes and edges in the graphs, respectively. Previous biomedical knowledge extraction methods simply considered limited entity types and relations by using a task-specific training set, which is insufficient for large-scale BKGs development and downstream task applications in different scenarios. To alleviate this issue, we propose a joint continual learning biomedical information extraction (JCBIE) network to extract entities and relations from different biomedical information datasets. By empirically studying different joint learning and continual learning strategies, the proposed JCBIE can learn and expand different types of entities and relations from different datasets. JCBIE uses two separated encoders in joint-feature extraction, hence can effectively avoid the feature confusion problem comparing with using one hard-parameter sharing encoder. Specifically, it allows us to adopt entity augmented inputs to establish the interaction between named entity recognition and relation extraction. Finally, a novel evaluation mechanism is proposed for measuring cross-corpus generalization errors, which was ignored by traditional evaluation methods. Our empirical studies show that JCBIE achieves promising performance when continual learning strategy is adopted with multiple corpora.

Compensation and monitoring of transmitter and receiver impairments in 10,000-km single-mode fiber transmission by adaptive multi-layer filters with augmented inputs.

We propose an adaptive multi-layer (ML) filter architecture to compensate for linear impairments that occur in transmitter (Tx) and receiver (Rx) components in ultra-long-haul optical fiber transmission systems, in which large chromatic dispersion (CD) accumulates in the received signal. The architecture consists of strictly linear (SL) and widely linear (WL) filter layers, and the coefficients of the ML filters are adaptively controlled by gradient calculation with back propagation and stochastic gradient descent. Static CD compensation is performed on the received signal and its complex conjugate before the adaptive ML filters. These augmented signals are then the inputs of the first 2×1 SL filter layer of the ML filters, for compensation of in-phase (I) and quadrature (Q) impairments on the Rx side. Tx IQ impairments and polarization effects as well as Rx IQ impairments are adaptively compensated in the ML filters. By sweeping CD compensation filters before the ML filters, this architecture mitigates the computational complexity for back propagation of the ML filters especially for ultra-long-haul transmission, while mutual non-commutativity between the WL filter for IQ impairment compensation and the CD compensation filter is appropriately solved. We evaluated the proposed adaptive ML filter architecture with augmented inputs through both simulation and wavelength-division multiplexed transmission experiments of 32-Gbaud polarization-division-multiplexed 64-quadrature amplitude modulation-based probabilistic constellation shaped signals over 10,000 km of single-mode fiber (SMF). The results demonstrated that the proposed adaptive ML filter architecture effectively compensates for Tx and Rx IQ skews in ultra-long-haul SMF transmission, and that impairments can be monitored individually from the converged filter coefficients of the corresponding layers.

Identifying New Classes of High Temperature Superconductors With Convolutional Neural Networks

Applying machine learning to aid the search for high temperature superconductors has recently been a topic of significant interest due to the broad applications of these materials but is challenging due to the lack of a quantitative microscopic model. Here we analyze over 33,000 entries from the Superconducting Materials Database, maintained by the National Institute for Materials Science of Japan, assigning crystal structures to each entry by correlation with Materials project and other structural databases. These augmented inputs are combined with material-specific properties, including critical temperature, to train convolutional neural networks (CNNs) to identify superconductors. Classification models achieve accuracy >95% and regression models trained to predict critical temperature achieve R2 >0.92 and mean absolute error ≈ 5.6 K. A crystal-graph representation whereby an undirected graph encodes atom sites (graph vertices) and their bonding relationships (graph edges), is used to represent materials’ periodic crystal structure to the CNNs. Trained networks are used to search though 130,000 crystal structures in the Materials Project for high temperature superconductor candidates and predict their critical temperature; several materials with model-predicted TC >30 K are proposed, including rediscovery of the recently explored infinite layer nickelates.

Fingerprint Presentation Attack Detection in Open-Set Scenario Using Transient Liveness Factor

Background: In recent history, fingerprint presentation attack detection (FPAD) proposal came out in a variety of ways. A close-set approach uses pattern classification technique that best suits to a specific context and goal. Openset approach works fine in wider context, which is relatively robust with new fabrication material and independent of sensor type. In both case results were promising but not too generalizable because of unseen condition not fitting into method used. It is clear, the two key challenges in FPAD system, sensor interoperability and robustness with new fabrication materials not addressed to date. Objective: To address above challenge a liveness detection model is proposed using live sample using transient liveness factor and one-class CNN. Methods: In our architecture, liveness is predicted by using the fusion rule, score level fusion of two decisions. Here, ‘n’ high quality live samples are initially trained for quality. We have observed that fingerprint liveness information is ‘transitory’ in nature, a variation in the different live sample is natural. Thus, each live sample has a ‘transient liveness’ (TL) information. We use no-reference (NR) image quality measure (IQM) as a transient value corresponding to each live sample. A consensus agreement is collectively reached in transient value to predict adversarial input. Further, live sample at server are trained with augmented inputs on the one-class classifier to predict the outlier. So, by using the fusion rule, score level fusion of consensus agreement and appropriately characterized negative cases (or outliers) predicts liveness. Results: Our approach uses high quality 30-live sample only, out of 90 images available in dataset to reduce learning time. We used Time Series images from LivDet competition 2015. It has 90-live images and 45-spoof images made from Bodydouble, Ecoflex and Playdoh of each person. Fusion rule results in 100% accuracy in recognising live as live. Conclusion: We have presented an architecture for liveness-server for extraction/updating transient liveness factor. Our work explained here a significant step forward towards generalized and reproducible process with a consideration towards the provision for the universal scheme as a need of today. The proposed TLF approach has a solid presumption; it will address dataset heterogeneity as it incorporates wider scope-context. Similar results with other dataset are under validation. Implementation seems difficult now but have several advantages when carried out during the transformative process.

LEDet: A Single-Shot Real-Time Object Detector Based on Low-Light Image Enhancement

Abstract Recently, significant breakthroughs have been achieved in the field of object detection. However, existing methods mostly focus on the generic object detection task. Performance degradation can be unavoidable when applying the existing methods to some specific situations directly, e.g. a low-light environment. To address this issue, we propose a single-shot real-time object Detector based on Low-light image Enhancement, namely LEDet. LEDet adapts itself to the low-light detection task in three aspects. First, a low-light enhancement module is introduced as the image preprocessor, producing the augmented inputs from the low-light images. Second, two modules, i.e. low-light and enhanced features fusion module and the scale-aware channel attention dilated convolution module are designed. These two modules aim at learning robust and discriminative features from objects of various sizes hidden in the darkness. In experiments, we validate the effectiveness of each part of our LEDet model via several ablation studies. We also compare LEDet with various methods on the Exclusively Dark dataset, showing that our model achieves the state-of-the-art performance on the balance between speed and accuracy.

Robustness and performance of Deep Reinforcement Learning

Deep Reinforcement Learning (DRL) has recently obtained considerable attentions. It empowers Reinforcement Learning (RL) with Deep Learning (DL) techniques to address various difficult tasks. In this paper, a novel approach called the Genetic Algorithm of Neuron Coverage (GANC) is proposed. It is motivated for improving the robustness and performance of a DRL network. The GANC uses Genetic Algorithm (GA) to maximise the Neuron Coverage (NC) of a DRL network by producing augmented inputs. We apply this method in the self-driving car applications, where it is crucial to accurately provide a correct decision for different road tracking views. We evaluate our method on the SYNTHIA-SEQS-05 databases in four different driving environments. Our outcomes are very promising – the best driving accuracy reached 97.75% – and are superior to the state-of-the-art results.

Heat and Mass Transfer of Natural Convective Flow with Slanted Magnetic Field via Fractional Operators

This article explores the MHD natural convective viscous and incompressible fluid flow along with radiation and chemical reaction. The flow is confined to a moving tilted plate under slanted magnetic field with variable temperature in a porous medium. Non-dimensional parameter along Laplace transformation and inversion algorithm are used to investigate the solution of system of dimensionless governing equations. Fractional differential operators namely, Caputo (C), Caputo-Fabrizio (CF) and Atangana-Baleanu in Caputo sense (ABC) are used to compare graphical behavior of for velocity, temperature and concentration for emerging parameters. On comparison, it is observed that fractional order model is better in explaining the memory effect as compared to classical model. Velocity showing increasing behavior for fractional parameter a whereas there is a decline in temperature, and concentration profiles for a. Fluid velocity goes through a decay due to rise in the values of M, Sc and j. However, velocity shows a reverse profile for augmented inputs of Kp , Gr and S. Tabular comparison is made for velocity and Nusselt number and Sherwood number for fractional models.

Rich Visual Knowledge-Based Augmentation Network for Visual Question Answering

Visual question answering (VQA) that involves understanding an image and paired questions develops very quickly with the boost of deep learning in relevant research fields, such as natural language processing and computer vision. Existing works highly rely on the knowledge of the data set. However, some questions require more professional cues other than the data set knowledge to answer questions correctly. To address such an issue, we propose a novel framework named a knowledge-based augmentation network (KAN) for VQA. We introduce object-related open-domain knowledge to assist the question answering. Concretely, we extract more visual information from images and introduce a knowledge graph to provide the necessary common sense or experience for the reasoning process. For these two augmented inputs, we design an attention module that can adjust itself according to the specific questions, such that the importance of external knowledge against detected objects can be balanced adaptively. Extensive experiments show that our KAN achieves state-of-the-art performance on three challenging VQA data sets, i.e., VQA v2, VQA-CP v2, and FVQA. In addition, our open-domain knowledge is also beneficial to VQA baselines. Code is available at https://github.com/yyyanglz/KAN.

Iterative Machine Learning for Output Tracking

This article develops iterative machine learning (IML) for output tracking. The input-output data generated during iterations to develop the model used in the iterative update. The main contribution of this article to propose the use of kernel-based machine learning to iteratively update both the model and the model-inversion-based input simultaneously. Additionally, augmented inputs with persistency of excitation are proposed to promote learning of the model during the iteration process. The proposed approach is illustrated with a simulation example.

Improved Rice Variety Adoption and its Welfare Impact on Rural Farming Households in Akwa Ibom State of Nigeria

ABSTRACT In this study cross-sectional data from 200 rice farmers were used to examine the various factors that influence the adoption of improved rice varieties distributed by the State Agricultural Development Programme (AKADEP) and its welfare impact on the farmers. The Foster, Greer and Thorbecke (FGT) class of measures was used to determine the incidence, the depth and severity of poverty among rice farming households who are adopters and non-adopters of improved rice varieties. The incidence, depth, and severity of poverty were higher among households who were non-adopters of improved rice varieties. The Tobit regression model was used to determine the factors that affect adoption and poverty. Educational attainment, access to extension agents, access to credit, access to augmented inputs, farm size, and crop yield were significant determinants of adoption of improved rice varieties. The results of the determinants of household poverty revealed that age, educational attainment, extent of commerciali...