Non-learning Methods Research Articles

Learning to route and schedule links in reconfigurable networks

This paper considers networks with a reconfigurable topology with so called 60 GHz dynamic links that can be activated or disabled over time. A fundamental problem is to jointly determine which 60 GHz dynamic links are active and the route chosen by source nodes over time. To this end, this paper outlines a hierarchical deep reinforcement learning solution that can be used to compute the optimal policy that determines for each time slot (i) active dynamic links, and (ii) the route used by each source–destination pair. The results show that the proposed approach results in a maximum average queue length that is 80% shorter than non-learning methods.

Texture and motion aware perception in-loop filter for AV1

Lossy compression introduces artifacts, and many conventional in-loop filters have been adopted in the AV1 standard to reduce these artifacts. Researchers have explored deep learning-based filters to remove artifacts in the compression loop. However, the high computational complexity of CNN-based filters remains a challenge. In this paper, a Texture- and Motion-Aware Perception (TMAP) in-loop filter is proposed to addresses this issue by selectively applying CNNs to texture-rich and high-motion regions, while utilizing non-learning methods to detect these regions. The proposed method introduces a new CNN structure, the Dense-Dual-Field Network (DDFN), which leverages a larger receptive field to enhance the quality of reconstructed frames by incorporating more contextual information. Furthermore, to improve perceptual quality, a novel loss function integrating wavelet-based perceptual information is presented. Experimental results demonstrate the superiority of our proposed models over other lightweight CNN models, and the effectiveness of the perceptual loss function is validated using the VMAF metric.

Autofocus methods based on laser illumination.

Autofocusing system plays an important role in microscopic measurement. However, natural-image-based autofocus methods encounter difficulties in improving focusing accuracy and robustness due to the diversity of detection objects. In this paper, a high-precision autofocus method with laser illumination was proposed, termed laser split-image autofocus (LSA), which actively endows the detection scene with image features. The common non-learning-based and learning-based methods for LSA were quantitatively analyzed and evaluated. Furthermore, a lightweight comparative framework model for LSA, termed split-image comparison model (SCM), was proposed to further improve the focusing accuracy and robustness, and a realistic split-image dataset of sufficient size was made to train all models. The experiment showed LSA has better focusing performance than natural-image-based method. In addition, SCM has a great improvement in accuracy and robustness compared with previous learning and non-learning methods, with a mean focusing error of 0.317µm in complex scenes. Therefore, SCM is more suitable for industrial measurement.

QOCT-Net: A Physics-Informed Neural Network for Intravascular Optical Coherence Tomography Attenuation Imaging.

Intravascular optical coherence tomography (IVOCT) provides high-resolution, depth-resolved images of coronary arterial microstructure by acquiring backscattered light. Quantitative attenuation imaging is important for accurate characterization of tissue components and identification of vulnerable plaques. In this work, we proposed a deep learning method for IVOCT attenuation imaging based on the multiple scattering model of light transport. A physics-informed deep network named Quantitative OCT Network (QOCT-Net) was designed to recover pixel-level optical attenuation coefficients directly from standard IVOCT B-scan images. The network was trained and tested on simulation and in vivo datasets. Results showed superior attenuation coefficient estimates both visually and based on quantitative image metrics. The structural similarity, energy error depth and peak signal-to-noise ratio are improved by at least 7%, 5% and 12.4%, respectively, compared with the state-of-the-art non-learning methods. This method potentially enables high-precision quantitative imaging for tissue characterization and vulnerable plaque identification.

BALLORG: State-of-the-art Image Restoration using Block-augmented Lagrangian and Low-rank Gradients

In this paper, we propose a blind image deblurring algorithm using block-augmented Lagrangian and low-rank priors (BALLORG) as a non-learning method that can give better results without the complexity of learning-based methods. The proposed algorithm achieves faster convergence within 20 iterations than conventional methods. Regularization priors are used in the form of gradients and sparse low-rank matrices, and recursive rank improvements result in better deblurring performance. The steepest descent in minimization is maintained through weight selection for penalty and regularization parameters. The block processing introduces local and global optimization, leading to better visual quality outputs. The proposed method has excellent performance in terms of the PSNR, SSIM, and FSIM matrix, which is on par with or better than that of other state-of-the-art learning and non-learning-based approaches.

A Deep Learning Method for Motion Artifact Correction in Intravascular Photoacoustic Image Sequence.

In vivo application of intravascular photoacoustic (IVPA) imaging for coronary arteries is hampered by motion artifacts associated with the cardiac cycle. Gating is a common strategy to mitigate motion artifacts. However, a large amount of diagnostically valuable information might be lost due to one frame per cycle. In this work, we present a deep learning-based method for directly correcting motion artifacts in non-gated IVPA pullback sequences. The raw signal frames are classified into dynamic and static frames by clustering. Then, a neural network named Motion Artifact Correction (MAC)-Net is designed to correct motion in dynamic frames. Given the lack of the ground truth information on the underlying dynamics of coronary arteries, we trained and tested the network using a computer-generated dataset. Based on the results, it has been observed that the trained network can directly correct motion in successive frames while preserving the original structures without discarding any frames. The improvement in the visual effect of the longitudinal view has been demonstrated based on quantitative evaluation of the inter-frame dissimilarity. The comparison results validated the motion-suppression ability of our method comparable to gating and image registration-based non-learning methods, while maintaining the integrity of the pullbacks without image preprocessing. Experimental results from in vivo intravascular ultrasound and optical coherence tomography pullbacks validated the feasibility of our method in the in vivo intracoronary imaging scenario.

PTR-CNN for in-loop filtering in video coding

A deep learning method called PTR-CNN (Predicted frame with Transform unit partition and prediction Residual aided CNN) is proposed for in-loop filtering in video compression. To reduce the computational complexity of an end-to-end CNN in-loop filter, a non-learning method of reference frame selection is designed to select the highest quality frame based on the frame’s blurriness and smoothiness scores. The transform unit (TU) partition and the prediction residual (PR) of the current frame are used as extra inputs to the neural network as the filtering guidance. The selected similar and high quality reference frame (RF) and the current unfiltered frame (CUF) are input to a CNN based motion compensation module to generate a predicted frame (PF). Finally input the PF, the CUF, the CUF’s TU partition and the CUF’s PR into the main CNN to reconstruct the filtered frame. The model is implemented in Tensorflow and tested in HEVC and AV1. Experimental results show that the complexity of proposed PTR-CNN is less than SOTA CNN-based reference aided in-loop filtering methods and slightly outperforms their RD performance. The scheme introduces a complexity overhead of 7% on the encoder. In particular, for random access, the proposed model achieves 11.78% coding gain over HEVC with DBF/SAO off, while has a gain of 4.76% over HEVC with DBF/SAO on. Ablation study demonstrates that the RF contributes about 10% of the total gain, and the TU and PR contribute over 4% of the total one, proving the effectiveness of each module. Moreover, it is observed that the proposed method can restore detailed structures and textures and hence improve the subjective quality.

Estimation of Respiratory Rate from Breathing Audio.

The COVID-19 pandemic has fueled exponential growth in the adoption of remote delivery of primary, specialty, and urgent health care services. One major challenge is the lack of access to physical exam including accurate and inexpensive measurement of remote vital signs. Here we present a novel method for machine learning-based estimation of patient respiratory rate from audio. There exist non-learning methods but their accuracy is limited and work using machine learning known to us is either not directly useful or uses non-public datasets. We are aware of only one publicly available dataset which is small and which we use to evaluate our algorithm. However, to avoid the overfitting problem, we expand its effective size by proposing a new data augmentation method. Our algorithm uses the spectrogram representation and requires labels for breathing cycles, which are used to train a recurrent neural network for recognizing the cycles. Our augmentation method exploits the independence property of the most periodic frequency components of the spectrogram and permutes their order to create multiple signal representations. Our experiments show that our method almost halves the errors obtained by the existing (non-learning) methods. Clinical Relevance- We achieve a Mean Absolute Error (MAE) of 1.0 for the respiratory rate while relying only on an audio signal of a patient breathing. This signal can be collected from a smartphone such that physicians can automatically and reliably determine respiratory rate in a remote setting.

High-fidelity deconvolution for acoustic-resolution photoacoustic microscopy enabled by convolutional neural networks

Acoustic-resolution photoacoustic microscopy (AR-PAM) image resolution is determined by the point spread function (PSF) of the imaging system. Previous algorithms, including Richardson–Lucy (R–L) deconvolution and model-based (MB) deconvolution, improve spatial resolution by taking advantage of the PSF as prior knowledge. However, these methods encounter the problems of inaccurate deconvolution, meaning the deconvolved feature size and the original one are not consistent (e.g., the former can be smaller than the latter). We present a novel deep convolution neural network (CNN)-based algorithm featuring high-fidelity recovery of multiscale feature size to improve lateral resolution of AR-PAM. The CNN is trained with simulated image pairs of line patterns, which is to mimic blood vessels. To investigate the suitable CNN model structure and elaborate on the effectiveness of CNN methods compared with non-learning methods, we select five different CNN models, while R–L and directional MB methods are also applied for comparison. Besides simulated data, experimental data including tungsten wires, leaf veins, and in vivo blood vessels are also evaluated. A custom-defined metric of relative size error (RSE) is used to quantify the multiscale feature recovery ability of different methods. Compared to other methods, enhanced deep super resolution (EDSR) network and residual in residual dense block network (RRDBNet) model show better recovery in terms of RSE for tungsten wires with diameters ranging from 30 μm to 120 μm. Moreover, AR-PAM images of leaf veins are tested to demonstrate the effectiveness of the optimized CNN methods (by EDSR and RRDBNet) for complex patterns. Finally, in vivo images of mouse ear blood vessels and rat ear blood vessels are acquired and then deconvolved, and the results show that the proposed CNN method (notably RRDBNet) enables accurate deconvolution of multiscale feature size and thus good fidelity.

Online learning agents for cost-sensitive topical data acquisition from the web

Access to one of the richest data sources in the world, the web, is not possible without cost. Often, this cost is not taken into account in data acquisition processes. In this paper, we introduce the Learning Agents (LA) method for automatic topical data acquisition from the web with minimum bandwidth usage and the lowest cost. The proposed LA method uses online learning topical crawlers. The online learning capability makes the LA able to dynamically adapt to the properties of web pages during the crawling process of the target topic, and learn an effective combination of a set of link scoring criteria for that topic. That way, the LA resolves the challenge in the mechanism of combining the outputs of different criteria for computing the value of following a link, in the formerly approaches, and increases the efficiency of the crawlers. A version of the LA method is implemented that uses a collection of topical content analyzers for scoring the links. The learning ability in the implemented LA resolves the challenge of the unclear appropriate size of link contexts for pages of different topics. Using standard metrics in empirical evaluation indicates that when non-learning methods show inefficiency, the learning capability of LA significantly increases the efficiency of topical crawling, and achieves the state of the art results.

Quantitative endoscopic photoacoustic tomography using a convolutional neural network.

Endoscopic photoacoustic tomography (EPAT) is a catheter-based hybrid imaging modality capable of providing structural and functional information of biological luminal structures, such as coronary arterial vessels and the digestive tract. The recovery of the optical properties of the imaged tissue from acoustic measurements achieved by optical inversion is essential for implementing quantitative EPAT (qEPAT). In this paper, a convolutional neural network (CNN) based on deep gradient descent is developed for qEPAT. The network enables the reconstruction of images representing the spatially varying absorption coefficient in cross-sections of the tubular structures from limited measurement data. The forward operator reflecting the mapping from the absorption coefficient to the optical deposition due to pulsed irradiation is embedded into the network training. The network parameters are optimized layer by layer through the deep gradient descent mechanism using the numerically simulated data. The operation processes of the forward operator and its adjoint operator are separated from the network training. The trained network outputs an image representing the distribution of absorption coefficients by inputting an image that represents the optical deposition. The method has been tested with computer-generated phantoms mimicking coronary arterial vessels containing various tissue types. Results suggest that the structural similarity of the images reconstructed by our method is increased by about 10% in comparison with the non-learning method based on error minimization in the case of the same measuring view.

Un-supervised learning for blind image deconvolution via Monte-Carlo sampling

Deep learning has been a powerful tool for solving many inverse imaging problems. The majority of existing deep-learning-based solutions are supervised on an external dataset with many blurred/latent image pairs. Recently, there has been an increasing interest on developing dataset-free deep learning methods for image recovery without any prerequisite on external training dataset, including blind deconvolution. This paper aims at developing an un-supervised learning method for blind image deconvolution, which does not call any training sample yet provides very competitive performance. Based on the re-parametrization of latent image using a deep network with random weights, this paper proposed to approximate the maximum-a posteriori estimator of the blur kernel using the Monte-Carlo (MC) sampling method. The MC sampling is efficiently implemented by using dropout and random noise layer, which does not require conjugate model as traditional variational inference does. Extensive experiments on popular benchmark datasets for blind image deconvolution showed that the proposed method not only outperformed existing non-learning methods, but also noticeably outperformed existing deep learning methods, including both supervised and un-supervised ones.

Rethinking Panoptic Segmentation in Remote Sensing: A Hybrid Approach Using Semantic Segmentation and Non-Learning Methods

This letter proposes a novel method to obtain panoptic predictions by extending the semantic segmentation task with a few non-learning image processing steps, presenting the following benefits: (1) annotations do not require a specific format (e.g., COCO); (2) fewer parameters (e.g., single loss function and no need for object detection parameters); and (3) a more straightforward sliding windows implementation for large image classification (still unexplored for panoptic segmentation). Semantic segmentation models do not individualize touching objects, as their predictions can merge, i.e., a single polygon represents many targets. Our method overcomes this problem by isolating the objects using borders on the polygons that may merge. The data preparation requires generating a 1-pixel border, and for unique object identification, we create a list with the isolated polygons, attribute a different value to each one, and use the expanding border (EB) algorithm for those with borders. Although any semantic segmentation model applies, we used the U-Net with three backbones (EfficientNet-B5, EfficientNet-B3, and EfficientNet-B0). The results show that (1) the EfficientNet-B5 had the best results with 70% mIoU; (2) the EB algorithm presented better results for better models; (3) the panoptic metrics show a high capability of identifying things and stuff with 65 Panoptic Quality; and (4) the sliding windows on a 2560x2560 pixel area has shown promising results, in which the ratio of merged objects by correct predictions was lower than 1% for all classes.

Lambertian Model-Based Normal Guided Depth Completion for LiDAR-Camera System

Depth completion is an essential task for the dense scene reconstruction on light detection and ranging (LiDAR)-camera system. Learning-based method achieves precise depth completion results on specific data sets. However, for the general outdoor scenes with insufficient labeled data sets, an efficient nonlearning method is still required. In this letter, from the geometrical constraint between depth and normal, a novel nonlearning normal guided depth completion method is proposed. For the objects in the outdoor scene, local brightness normal (LBN) constraint is derived from the Lambertian model. It is used to recover dense normal from RGB image and sparse normal. After that, we present a pipeline for depth completion with the guidance of dense normal. Extensive experiments on the KITTI depth completion data set demonstrate that our method achieves smaller root mean squared error (RMSE) than current nonlearning methods.

Learned cardinality estimation

Cardinality estimation is core to the query optimizers of DBMSs. Non-learned methods, especially based on histograms and samplings, have been widely used in commercial and open-source DBMSs. Nevertheless, histograms and samplings can only be used to summarize one or few columns, which fall short of capturing the joint data distribution over an arbitrary combination of columns, because of the oversimplification of histograms and samplings over the original relational table(s). Consequently, these traditional methods typically make bad predictions for hard cases such as queries over multiple columns, with multiple predicates, and joins between multiple tables. Recently, learned cardinality estimators have been widely studied. Because these learned estimators can better capture the data distribution and query characteristics, empowered by the recent advance of (deep learning) models, they outperform non-learned methods on many cases. The goals of this paper are to provide a design space exploration of learned cardinality estimators and to have a comprehensive comparison of the SOTA learned approaches so as to provide a guidance for practitioners to decide what method to use under various practical scenarios.

A Surface Geometry Model for LiDAR Depth Completion

LiDAR depth completion is a task that predicts depth values for every pixel on the corresponding camera frame, although only sparse LiDAR points are available. Most of the existing state-of-the-art solutions are based on deep neural networks, which need a large amount of data and heavy computations for training the models. In this letter, a novel non-learning depth completion method is proposed by exploiting the local surface geometry that is enhanced by an outlier removal algorithm. The proposed surface geometry model is inspired by the observation that most pixels with unknown depth have a nearby LiDAR point. Therefore, it is assumed those pixels share the same surface with the nearest LiDAR point, and their respective depth can be estimated as the nearest LiDAR depth value plus a residual error. The residual error is calculated by using a derived equation with several physical parameters as input, including the known camera intrinsic parameters, estimated normal vector, and offset distance on the image plane. The proposed method is further enhanced by an outlier removal algorithm that is designed to remove incorrectly mapped LiDAR points from occluded regions. On KITTI dataset, the proposed solution achieves the best error performance among all existing non-learning methods and is comparable to the best self-supervised learning method and some supervised learning methods. Moreover, since outlier points from occluded regions is a commonly existing problem, the proposed outlier removal algorithm is a general preprocessing step that is applicable to many robotic systems with both camera and LiDAR sensors.

Deep Reinforcement Learning Based Dynamic Spectrum Competition in Green Cognitive Virtualized Networks

This paper examines the optimal spectrum competing strategy for a virtual network operator in cognitive cellular networks with energy-harvesting base stations. In the scenario for this study, multiple cognitive virtual network operators (CVNOs) obtain spectrum resources from a mobile network operator via spectrum sensing and leasing in order to provide data services to their subscribers. Compared to traditional spectrum leasing via long-term contract, spectrum acquired by sensing is usually cheaper but is unreliable due to the stochastic activities of the licensed users. The CVNOs need to determine the optimal sensing and leasing amount to satisfy the needs of subscribers while guaranteeing a low leasing cost. We aim to find an efficient spectrum sensing and leasing scheme for a CVNO in order to maximize its utility in the long run. The problem is first formulated as the framework of a sequential decision process considering the dynamics of users’ activities, spectrum prices, and harvested energy. We then develop a deep reinforcement learning algorithm that uses deep neural networks as function approximators so the CVNO can learn the optimal decision policy by interacting with the environment. We analyze the performance of our proposed scheme through extensive simulations. The experiment results show that the proposed mechanism can significantly improve the CVNO’s long-term benefit compared to other learning and non-learning methods.

Research on Recognition Effect of DSCN Network Structure in Hand-Drawn Sketch.

With the rapid development of image recognition technology, freehand sketch recognition has attracted more and more attention. How to achieve good recognition effect in the absence of color and texture information is the key to the development of freehand sketch recognition. Traditional nonlearning classical models are highly dependent on manual selection features. To solve this problem, a neural network sketch recognition method based on DSCN structure is proposed in this paper. Firstly, the stroke sequence of the sketch is drawn; then, the feature is extracted according to the stroke sequence combined with neural network, and the extracted image features are used as the input of the model to construct the time relationship between different image features. Through the control experiment on TU-Berlin dataset, the results show that, compared with the traditional nonlearning methods, HOG-SVM, SIFT-Fisher Vector, MKL-SVM, and FV-SP, the recognition accuracy of DSCN network is improved by 15.8%, 10.3%, 6.0%, and 2.9%, respectively. Compared with the classical deep learning model, Alex-Net, the recognition accuracy is improved by 5.6%. The above results show that the DSCN network proposed in this paper has strong ability of feature extraction and nonlinear expression and can effectively improve the recognition accuracy of hand-painted sketches after introducing the stroke order.

Improving Real-Time Position Estimation Using Correlated Noise Models

We provide algorithms for inferring GPS (Global Positioning System) location and for quantifying the uncertainty of this estimate in real time. The algorithms are tested on GPS data from locations in the Southern Hemisphere at four significantly different latitudes. In order to rank the algorithms, we use the so-called log-score rule. The best algorithm uses an Ornstein–Uhlenbeck (OU) noise model and is built on an enhanced Kalman Filter (KF). The noise model is capable of capturing the observed autocorrelated process noise in the altitude, latitude and longitude recordings. This model outperforms a KF that assumes a Gaussian noise model, which under-reports the position uncertainties. We also found that the dilution-of-precision parameters, automatically reported by the GPS receiver at no additional cost, do not help significantly in the uncertainty quantification of the GPS positioning. A non-learning method using the actual position measurements and employing a constant uncertainty does not even converge to the correct position. Inference with the enhanced noise model is suitable for embedded computing and capable of achieving real-time position inference, can quantify uncertainty and be extended to incorporate complementary sensor recordings, e.g., from an accelerometer or from a magnetometer, in order to improve accuracy. The algorithm corresponding to the augmented-state unscented KF method suggests a computational cost of , where is the dimension of the augmented state-vector and is an adjustable, design-dependent parameter corresponding to the length of “past values” one wishes to keep for re-evaluation of the model from time to time. The provided algorithm assumes . Hence, the algorithm is likely to be suitable for sensor fusion applications.

Practical sensorless aberration estimation for 3D microscopy with deep learning.

Estimation of optical aberrations from volumetric intensity images is a key step in sensorless adaptive optics for 3D microscopy. Recent approaches based on deep learning promise accurate results at fast processing speeds. However, collecting ground truth microscopy data for training the network is typically very difficult or even impossible thereby limiting this approach in practice. Here, we demonstrate that neural networks trained only on simulated data yield accurate predictions for real experimental images. We validate our approach on simulated and experimental datasets acquired with two different microscopy modalities and also compare the results to non-learned methods. Additionally, we study the predictability of individual aberrations with respect to their data requirements and find that the symmetry of the wavefront plays a crucial role. Finally, we make our implementation freely available as open source software in Python.