We present a novel approach to view synthesis using multiplane images (MPIs). Building on recent advances in learned gradient descent, our algorithm generates an MPI from a set of sparse camera ...viewpoints. The resulting method incorporates occlusion reasoning, improving performance on challenging scene features such as object boundaries, lighting reflections, thin structures, and scenes with high depth complexity. We show that our method achieves high-quality, state-of-the-art results on two datasets: the Kalantari light field dataset, and a new camera array dataset, Spaces, which we make publicly available.
Recent legislation has led to interest in machine unlearning, i. e., removing specific training samples from a predictive model as if they never existed in the training dataset. Unlearning may also ...be required due to corrupted/adversarial data or simply a user's updated privacy requirement. For models which require no training (k-NN), simply deleting the closest original sample can be effective. But this idea is inapplicable to models which learn richer representations. Recent ideas leveraging optimization-based updates scale poorly with the model dimension d, due to inverting the Hessian of the loss function. We use a variant of a new conditional independence coefficient, L-CODEC, to identify a subset of the model parameters with the most semantic overlap on an individual sample level. Our approach completely avoids the need to invert a (possibly) huge matrix. By utilizing a Markov blanket selection, we premise that L-CODEC is also suitable for deep unlearning, as well as other applications in vision. Compared to alternatives, L-CODEC makes approximate unlearning possible in settings that would otherwise be infeasible, including vision models used for face recognition, person reidentification and NLP models that may require unlearning samples identified for exclusion. Code is available at https://github.com/vsingh-group/LCODEC-deep-unlearning
The theoretical developments of data -driven fault diagnosis methods have yielded fruitful achievements and significantly benefited industry practices. However, most methods are developed based on ...the assumption of data balance, which is incompatible with engineering scenarios. First, the normal state accounts for the majority of the equipment's lifespan; second, the probability of various faults varies, both of which result in an imbalance in the data. The consequence of data imbalance in intelligent fault diagnosis methods has attracted extensive attention from the research community, and a significant number of papers have been published. Nevertheless, a comprehensive review of achievements in this field is still missing, and the research perspectives have not been thoroughly investigated. To end this, we review and discuss all the research achievements in fault diagnosis under data imbalance in this survey, based on to the best of our knowledge. First, the existing imbalanced learning methods are classified into three categories: data processing methods, model construction methods, and training optimization methods. Then, the three methodologies are introduced and discussed in detail: the data processing method is to optimize the inputs of the intelligent fault diagnosis model so that the imbalance rate of the sample set involved in training is reduced; the model construction method is to design the structure and the features of the intelligent fault diagnosis model so that the model itself is resistant to the effects of imbalance; the training optimization method is an optimization of the training process for intelligent fault diagnosis models, raising the importance of the minority class in the training. Finally, this survey summarizes the prospects of the imbalanced learning problem in intelligent fault diagnosis, discusses the possible solutions, and provides some recommendations.
Importance Estimation for Neural Network Pruning Molchanov, Pavlo; Mallya, Arun; Tyree, Stephen ...
2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),
06/2019
Conference Proceeding
Odprti dostop
Structural pruning of neural network parameters reduces computational, energy, and memory transfer costs during inference. We propose a novel method that estimates the contribution of a neuron ...(filter) to the final loss and iteratively removes those with smaller scores. We describe two variations of our method using the first and second-order Taylor expansions to approximate a filter's contribution. Both methods scale consistently across any network layer without requiring per-layer sensitivity analysis and can be applied to any kind of layer, including skip connections. For modern networks trained on ImageNet, we measured experimentally a high (>93%) correlation between the contribution computed by our methods and a reliable estimate of the true importance. Pruning with the proposed methods led to an improvement over state-of-the-art in terms of accuracy, FLOPs, and parameter reduction. On ResNet-101, we achieve a 40% FLOPS reduction by removing 30% of the parameters, with a loss of 0.02% in the top-1 accuracy on ImageNet.
In this letter, a novel coding metasurface (CM) based on low Q resonators and fast optimization method is proposed to achieve wideband radar cross section (RCS) reduction of the microstrip antenna ...array (MAA) while maintaining its radiation properties. Theoretical analysis reveals the relationship between the Q value and the phase shift of the resonator, which indicates that the wideband control of the reflected wave can be achieved with the low Q resonator. Moreover, the fast optimization method based on the convolution theorem is proposed, which enhances the optimization efficiency of the coding matrix. Finally, the CM with two kinds of low Q resonators is constructed and applied to the MAA for wideband RCS reduction. The measured results indicate that the proposed low RCS MAA (LRMAA) can realize more than 10 dB RCS reduction in 5.8-21.5 GHz, which also demonstrates superior specular scattering suppression. In parallel, the radiation properties of the LRMAA are consistent with those of MAA. The proposed strategy achieves the high integration of radiation and scattering properties, which has potential applications in antenna systems of stealth platforms.
Efficient Neural Network Compression Kim, Hyeji; Khan, Muhammad Umar Karim; Kyung, Chong-Min
2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),
2019-June
Conference Proceeding
Odprti dostop
Network compression reduces the computational complexity and memory consumption of deep neural networks by reducing the number of parameters. In SVD-based network compression the right rank needs to ...be decided for every layer of the network. In this paper we propose an efficient method for obtaining the rank configuration of the whole network. Unlike previous methods which consider each layer separately, our method considers the whole network to choose the right rank configuration. We propose novel accuracy metrics to represent the accuracy and complexity relationship for a given neural network. We use these metrics in a non-iterative fashion to obtain the right rank configuration which satisfies the constraints on FLOPs and memory while maintaining sufficient accuracy. Experiments show that our method provides better compromise between accuracy and computational complexity/memory consumption while performing compression at much higher speed. For VGG-16 our network can reduce the FLOPs by 25% and improve accuracy by 0.7% compared to the baseline, while requiring only 3 minutes on a CPU to search for the right rank configuration. Previously, similar results were achieved in 4 hours with 8 GPUs. The proposed method can be used for lossless compression of a neural network as well. The better accuracy and complexity compromise, as well as the extremely fast speed of our method make it suitable for neural network compression.
The development of a scheduling strategy for an islanded microgrid (IMG) is critical for ensuring the system’s stability and economic efficiency. Traditional scheduling strategies for IMGs ...predominantly utilize centralized management by the microgrid central controller (MGCC), which introduces a vulnerability to a single point of failure. To address this limitation, this paper presents a two-layer energy management strategy for IMGs based on the improved alternating direction method of multipliers (ADMM) and inverse reinforcement learning (IRL). First, the framework of the proposed strategy, comprising a scheduling layer and a real-time dispatch layer, is outlined. Next, the problem formulation of the scheduling layer is analyzed, and the proposed IRL-based management strategy for the energy storage system (ESS) is presented. Then, a distributed optimization algorithm based on the improved ADMM is proposed for the management of controllable distributed generators (CDGs) in the real-time dispatch layer. Lastly, the case study demonstrates the efficacy of the proposed strategy in diminishing MGCC dependency. The comparative analysis indicates that the proposed strategy outperforms existing scheduling strategies in terms of cost-effectiveness when the forecast error exceeds 3%. Moreover, in contrast to existing scheduling strategies, the proposed strategy mitigates the risk associated with a single point of failure.
•A two-layer distributed energy management for islanded microgrid.•Inverse reinforcement learning based scheduling layer.•Distributed alternating direction method of multipliers real-time dispatch layer.•Reduce reliance on centralized controller and risk of single points of failure.
•Total twenty-eight meta-heuristic algorithms are introduced.•Various simulation results of different methods are discussed.•A comprehensive table summarizes all the meta-heuristic ...algorithms.•Several constructive recommendations are given for future development.
Accurate parameter identification is crucial for a precise PV cell modelling and analysis of characteristics of PV systems, while high nonlinearity of output I-V curve makes this problem extremely thorny. Hence, a large number of researches have aroused extensive interests in the past few years. Due to the rapid advancement of computer technology and swarm intelligence, various promising meta-heuristic algorithms have been proposed to further accelerate this trend. This paper aims to undertake a comprehensive review on meta-heuristic algorithms and related variants which have been applied on PV cell parameter identification. Particularly, these algorithms are classified into four categories, e.g., biology-based algorithms, physics-based algorithms, sociology-based algorithms and mathematics-based algorithms. Meanwhile, the evaluation criteria and identification performance of each algorithm are thoroughly addressed. Besides, in order to quantitatively evaluate and compare various algorithms, the identified PV parameters including the specific error and the simulated output I-V or P-V curves are provided at the end of each algorithm. Moreover, a comprehensive summary is also introduced to more specifically guide the readers to grasp and utilize these approaches. Lastly, based on the covered twenty-eight algorithms, conclusion presents some perspectives and recommendations for future development.
Display omitted
•Latest progress of inorganic heterogeneous co-catalytic Fenton systefms are reviewed.•Evolution of transition metal and non-metal heterogeneous cocatalysts are summarized.•Four ...optimization methods for inorganic heterogeneous co-catalysis are proposed.
Traditional Fenton reaction refers to the activation of H2O2 by Fe2+ to generate hydroxyl radicals, but Fe2+ is easily oxidized to Fe3+ and the reduction of Fe2+ is a difficult challenge. Recently, more and more scientists have been trying to resolve this problem by adding inorganic heterogeneous cocatalyst into the reaction and establish heterogeneous co-catalytic system. We reviewed the latest evolution of inorganic heterogeneous co-catalytic Fenton methods in wastewater treatment. The two distinct kinds of heterogeneous cocatalyst, transition metals (Single-atom, zero-valent metal, metal sulphide, and metal phosphide) and non-metal (carbon, boron, and phosphorus materials) cocatalysts are summarized, and the reaction mechanisms of each type are discussed, then the advantages and disadvantages of practical applications are assessed. Four optimization methods for the co-catalytic system are proposed, including material optimization, operational optimization, synergy with other advanced oxidation processes, and novel Fenton-like process construction. Finally, future research needs for Fenton-like systems are presented.
Photovoltaic (PV) cells are widely used for their clean and sustainable advantages, forcing researchers to accurately model their characteristics. The behavior of PV cells can be derived from their ...current–voltage characteristics, depending on their unknown circuit model parameters. Due to the simulation, evaluation, control, and optimization of PV systems, it is essential to accurately and reliably extract the parameters of PV models. However, because of the non-linear, multi-variable, and multi-modal characteristics, it is still a very challenging task. With the rapid development of intelligent computing, various meta-heuristic algorithms have been devoted to extracting the parameters of different PV models. The purpose of this paper is to comprehensively review the meta-heuristic algorithms and their related variants that have been used to extract the parameters of different PV models. Different from the existing research works, this paper presents a comprehensive review based on the reliability, robustness, computational resources, and time complexity of the algorithm. These features are essential to design an algorithm for efficient parameter extraction of PV models. Based on the conducted review, some useful recommendations are provided, which have important reference significance when designing the new parameter extraction methods of PV models and are of great significance for further improving the performance, control, and design of PV cells.
•Some common PV models and their parameter optimization problems are described.•Various meta-heuristic methods for parameter extraction of PV models are reviewed.•Important insights and future research directions are summarized.
PDF
