Weights Of Neural Network Research Articles

Is Federated Learning Still Alive in the Foundation Model Era?

Federated learning (FL) has arisen as an alternative to collecting large amounts of data in a central place to train a machine learning (ML) model. FL is privacy-friendly, allowing multiple parties to collaboratively train an ML model without exchanging or transmitting their training data. For this purpose, an aggregator iteratively coordinates the training process among parties, and parties simply share with the aggregator model updates, which contain information pertinent to the model such as neural network weights. Besides privacy, generalization has been another key driver for FL: parties who do not have enough data to train a good performing model by themselves can now engage in FL to obtain an ML model suitable for their tasks. Products and real applications in the industry and consumer space have demonstrated the power of this learning paradigm. Recently, foundation models have taken the AI community by storm, promising to solve the shortage of labeled data. A foundation model is a powerful model that can be recycled for a variety of use cases by applying techniques such as zero-shot learning and full or parameter-efficient fine tuning. The premise is that the amount of data required to fine tune a foundation model for a new task is much smaller than fully training a traditional model from scratch. The reason why this is the case is that a good foundation model has already learned relevant general representations, and thus, adapting it to a new task only requires a minimal number of additional samples. This raises the question: Is FL still alive in the era of foundation models? In this talk, I will address this question. I will present some use cases where FL is very much alive. In these use cases, finding a foundation model with a desired representation is difficult if not impossible. With this pragmatic point of view, I hope to shed some light into a real use case where disparate private data is available in isolation at different parties and where labels may be located at a single party that doesn’t have any other information, making it impossible for a single party to train a model on its own. Furthermore, in some vertically-partitioned scenarios, cleaning data is not an option due to privacy-related reasons and it is not clear how to apply foundation models. Finally, I will also go over a few other requirements that are often overlooked, such as unlearning of data and its implications for the lifecycle management of FL and systems based on foundation models.

Optimising air quality prediction in smart cities with hybrid particle swarm optimization‐long‐short term memory‐recurrent neural network model

AbstractIn smart cities, air pollution is a critical issue that affects individual health and harms the environment. The air pollution prediction can supply important information to all relevant parties to take appropriate initiatives. Air quality prediction is a hot area of research. The existing research encounters several challenges that is, poor accuracy and incorrect real‐time updates. This research presents a hybrid model based on long‐short term memory (LSTM), recurrent neural network (RNN), and Curiosity‐based Motivation method. The proposed model extracts a feature set from the training dataset using an RNN layer and achieves sequencing learning by applying an LSTM layer. Also, to deal with the overfitting issues in LSTM, the proposed model utilises a dropout strategy. In the proposed model, input and recurrent connections can be dropped from activation and weight updates using the dropout regularisation approach, and it utilises a Curiosity‐based Motivation model to construct a novel motivational model, which helps in the reconstruction of long short‐term memory recurrent neural network. To minimise the prediction error, particle swarm optimisation is implemented to optimise the LSTM neural network's weights. The authors utilise an online Air Pollution Monitoring dataset from Salt Lake City, USA with five air quality indicators for comparison, that is, SO2, CO, O3, and NO2, to predict air quality. The proposed model is compared with existing Gradient Boosted Tree Regression, Existing LSTM, and Support Vector Machine based Regression Model. Experimental analysis shows that the proposed method has 0.0184 (Root Mean Square Error (RMSE)), 0.0082 (Mean Absolute Error), 2002*109 (Mean Absolute Percentage Error), and 0.122 (R2‐Score). The experimental findings demonstrate that the proposed LSTM model had RMSE performance in the prescribed dataset and statistically significant superior outcomes compared to existing methods.

FKeras: A Sensitivity Analysis Tool for Edge Neural Networks

Edge computation often requires robustness to faults, e.g., to reduce the effects of transient errors and to function correctly in high radiation environments. In these cases, the edge device must be designed with fault tolerance as a primary objective. FKeras is a tool that helps design fault-tolerant edge neural networks that run entirely on chip to meet strict latency and resource requirements. FKeras provides metrics that give a bit-level ranking of neural network weights with respect to their sensitivity to faults. FKeras includes these sensitivity metrics to guide efficient fault injection campaigns to help evaluate the robustness of a neural network architecture. We show how to use FKeras in the co-design of edge NNs trained on the high-granularity endcap calorimeter dataset, which represents high energy physics data, as well as the CIFAR-10 dataset. We use FKeras to analyze a NN’s fault tolerance to consider alongside its accuracy, performance, and resource consumption. The results show that the different NN architectures have vastly differing resilience to faults. FKeras can also determine how to protect neural network weights best, e.g., by selectively using triple modular redundancy on only the most sensitive weights, which reduces area without affecting accuracy.

Identification and Control of Flexible Joint Robots Based on a Composite-Learning Optimal Bounded Ellipsoid Algorithm and Prescribe Performance Control Technique

This paper presents an indirect adaptive neural network (NN) control algorithm tailored for flexible joint robots (FJRs), aimed at achieving desired transient and steady-state performance. To simplify the controller design process, the original higher-order system is decomposed into two lower-order subsystems using the singular perturbation technique (SPT). NNs are then employed to reconstruct the aggregated uncertainties. An adaptive prescribed performance control (PPC) strategy and a continuous terminal sliding mode control strategy are introduced for the reduced slow subsystem and fast subsystem, respectively, to guarantee a specified convergence speed and steady-state accuracy for the closed-loop system. Additionally, a composite-learning optimal bounded ellipsoid algorithm (OBE)-based identification scheme is proposed to update the NN weights, where the tracking errors of the reduced slow and fast subsystems are integrated into the learning algorithm to enhance the identification and tracking performance. The stability of the closed-loop system is rigorously established using the Lyapunov approach. Simulations demonstrate the effectiveness of the proposed identification and control schemes.

Predictor‐based collision‐free and connectivity‐preserving resilient formation control for multi‐agent systems under sensor deception attacks

AbstractMalicious attack is a potential threat to collision‐free and connectivity‐preserving formation. In this article, a predictor‐based collision‐free and connectivity‐preserving resilient formation control strategy is proposed for a class of multi‐agent systems under sensor deception attacks. The predictor states are constructed to replace original states in the control strategy, and a novel attack compensator is constructed to suppress sensor deception attacks. Prediction errors, instead of compromised errors, are introduced to update weights of radial basis function neural networks (RBFNNs). To achieve collision avoidance and connectivity preservation, a transformation function in logarithmic form is introduced. To avoid static and dynamic obstacles, an improved artificial potential function (APF) combined with their velocity information is constructed. Furthermore, to solve the local minimum problem in the combining of transformation function and APF, a virtual force is added. Based on the Lyapunov function, all closed‐loop signals are bounded and collision‐free and connectivity‐preserving formation is achieved. The simulation related to a group of quadrotors verifies the effectiveness of the proposed resilient control strategy.

Surface‐aware Mesh Texture Synthesis with Pre‐trained 2D CNNs

AbstractMesh texture synthesis is a key component in the automatic generation of 3D content. Existing learning‐based methods have drawbacks—either by disregarding the shape manifold during texture generation or by requiring a large number of different views to mitigate occlusion‐related inconsistencies. In this paper, we present a novel surface‐aware approach for mesh texture synthesis that overcomes these drawbacks by leveraging the pre‐trained weights of 2D Convolutional Neural Networks (CNNs) with the same architecture, but with convolutions designed for 3D meshes. Our proposed network keeps track of the oriented patches surrounding each texel, enabling seamless texture synthesis and retaining local similarity to classical 2D convolutions with square kernels. Our approach allows us to synthesize textures that account for the geometric content of mesh surfaces, eliminating discontinuities and achieving comparable quality to 2D image synthesis algorithms. We compare our approach with state‐of‐the‐art methods where, through qualitative and quantitative evaluations, we demonstrate that our approach is more effective for a variety of meshes and styles, while also producing visually appealing and consistent textures on meshes.

Antenna modeling based on meta-heuristic intelligent algorithms and neural networks

As wireless communication technology continues to advance, the antenna, as an essential front-end device in radio communication system, is surrounded by more and more complex electromagnetic wave environments with increasing variety, resulting in greater demand for antennas and higher design requirements. While the traditional antenna design methods suffered from the disadvantage of low design efficiency, a powerful tool for accelerating antenna design is the modelling of antennas with neural networks. Aiming to enhance the modeling accuracy of neural network, multiple novel meta-heuristic swarm intelligent algorithms are introduced and part of them are modified for the purpose of applying to optimizing network’s weights and biases so as to raise the antenna model`s prediction precision on the basis of neural network. Specifically, the intelligent algorithms and their improvement directions include the strategy of optimizing weights and biases for neural networks with seagull optimization algorithm, optimizing the weights and biases of neural network with the improved butterfly algorithm fused with reverse learning, and the artificial rabbit algorithm optimizing the neural network weights and biases. In addition, two intelligent optimization algorithms that are already more mature: the particle swarm algorithm and the genetic algorithm are added to compare with the above three algorithms. The accuracy of neural network prediction before and after the optimisation of neural network by seagull algorithm, the butterfly algorithm incorporating reverse learning, the artificial rabbit algorithm, the particle swarm algorithm, and the genetic algorithm are got through the results respectively. The results of the experiments displayed that the neural network optimized of the improved butterfly algorithm incorporating reverse learning has a prediction accuracy of 99.69 % with stable results, the optimised neural network prediction accuracy of the seagull algorithm reaches 99.51%, and the optimised neural network prediction accuracy of the artificial rabbit algorithm is 99.49 %. The remaining two traditional algorithms optimized neural network accuracy is 83.1 % and 99.43 % respectively. Therefore, the improved butterfly algorithm incorporating reverse learning is the most effective among these three new algorithms applied to the field of antenna prediction. Moreover, the running time of the network optimized by different algorithms is quite distinct, among which the neural network optimized by the improved butterfly algorithm incorporating reverse learning takes the shortest time, which increases the prediction efficiency of the network by more than 70%. In summary, the application of the fused reverse learning improved butterfly algorithm in optimizing neural network predictions yields the shortest processing time and highest accuracy. This not only enables faster and more precise antenna design but also holds greater significance for the field of antenna design and analysis.

A Metro Passenger Flow Forecasting Model Based On Time-series Evolving Interaction Graph Network

Passenger flow forecasting is an important task in metro operation management. In order to achieve more accurate metro passenger flow forecasting, this paper proposes a metro passenger flow forecasting model based on time-series evolution interaction graph. First, by introducing two kinds of inter-station interaction graphs, namely connectivity graph and temporal correlation graph, to capture the potential interaction relationship among metro passenger flow stations. Then, by using the time-series evolving graph, the weights of the graph convolutional neural network are dynamically evolved in time series. Finally, taking Suzhou Metro as an example, the short-term passenger flow of the metro is forecasted. The experimental results show that the Root Mean Squared Error (RMSE) of this model is 34.17, the Mean Absolute Error (MAE) is 16.35, the R-Squared is 0.94, the Mean Absolute Percent Error (MAPE) is 0.21. All evaluation metrics are better than the baseline models, thus verifying the effectiveness and applicability of the metro passenger flow forecasting model based on the time series evolution interaction graph.

An analytical adaptive optimal control approach without solving HJB equation for nonlinear systems with input constraints

AbstractThis paper presents an analytical method to solve the optimal control problem for affine nonlinear systems with unknown drift dynamics. A new non‐quadratic cost function over an infinite horizon is presented that considers input constraints and includes the cost of the feed‐forward component of the control law. The mean value theorem for vector‐valued functions has been used to derive an integral form of this theorem. Based on this theorem, a rigorous proof is provided demonstrating that the cost function can be converted into another form. In the presence of input constraints, this converted form enables extracting the optimal control solution without solving the HJB equation. Additionally, unknown nonlinearity effects in drift dynamics are compensated in the control input. This is accomplished by estimating the unknown drift dynamics via an adaptive neural network (NN) approach. It is proven that the states and weights of NN are uniformly ultimately bounded based on a Lyapunov technique. The necessary and sufficient conditions are provided that ensure the optimality of the infinite horizon optimal control problem with a discount factor. As a result, it is demonstrated that the proposed approach satisfies the optimality criteria. To evaluate the effectiveness of the proposed approach, simulation examples are provided.

Near-Optimal Tracking Control of Partially Unknown Discrete-Time Nonlinear Systems Based on Radial Basis Function Neural Network

This paper proposes an optimal tracking control scheme through adaptive dynamic programming (ADP) for a class of partially unknown discrete-time (DT) nonlinear systems based on a radial basis function neural network (RBF-NN). In order to acquire the unknown system dynamics, we use two RBF-NNs; the first one is used to construct the identifier, and the other one is used to directly approximate the steady-state control input, where a novel adaptive law is proposed to update neural network weights. The optimal feedback control and the cost function are derived via feedforward neural network approximation, and a means of regulating the tracking error is proposed. The critic network and the actor network were trained online to obtain the solution of the associated Hamilton–Jacobi–Bellman (HJB) equation within the ADP framework. Simulations were carried out to verify the effectiveness of the optimal tracking control technique using the neural networks.

Thermophysical dynamics of magnetohydrodynamic electroosmotic two phase mixed convective flow in vertical annulus: neuro computation

ABSTRACT The thermophysical dynamics of electro – magnetohydrodynamic mixed convective electrically conducting fluid flow in the vertical cylindrical annulus are explored with electric and magnetic fields. The flow and heat transfer properties are explicitly examined in viscous fluid and electroosmotic flow regions. The coupling between an external pressure gradient, electroosmosis, and an extended electromagnetic field can control fluid flow and heat transfer. The governing equations associated with the physical phenomenon are addressed with the Galerkin finite element method. Eloquently, the analysis revealed that, in the absence of a lateral electric field, the flow velocity decelerates as the Hartmann number increases, which in turn ultimately, causes the Nusselt number to rise. Since a lateral electric field is involved, the flow field, temperature field, and Nusselt number profiles are presented in two regions. The artificial neural network model discusses the thermal transport phenomenon in both regions. The artificial neural network models regression, performance plots, weights, and bias are presented. The research outcome can be utilized to design exquisite and efficacious electromagnetic devices, particularly within a particular range of thermal physical properties.

Optimization Research on Provincial Users' Participation in Regional Electricity Spot Market Bidding Based on Hierarchically Gated Recurrent Neural Network

The creation of strategic pricing models is crucial in the highly competitive centralized energy markets in order to maximize revenues. In order to achieve maximum profit, market participants must act intelligently while bidding on electrical energy purchases or sales. Notwithstanding the many restructuring changes of the Chinese electricity sector, market participants are nonetheless exposed to risks associated with volatility in market prices and uncertainty over demand behavior. The data are collected from New South Wales electricity market. Data-adaptive Gaussian average filtering (DAGAF) is used to pre-process the data during the pre-processing step. Hierarchically Gated Recurrent Neural Networks (HGRNNs) are successfully used to classify both long-term and spot transactions. The neural network's weight parameter is optimized by the Lotus Effect Optimization Algorithm (LEA), which enhances the HGRNN. The RESMB-HGRNN-LEA proposed is implemented on the Python working platform. To calculate the suggested approach, performance measures including accuracy, precision, sensitivity, computation time, and recall were looked at.In comparison to the existing technique, the proposed RESMB-HGRNN-LEA method yields better results in terms of accuracy (16.65% and 18.85%), sensitivity (16.34% and 12.23%), precision (14.89% and 16.89%), and computing time (82.37% and 94.47%).

Design a Platform for Sharing Educational Resources for Student Based on Dual Attention Graph Convolutional Networks

A key component of raising educational standards is always the sharing of educational resources throughout higher education institutions. The development of digital and network infrastructure among colleges and universities can propel the sharing and exploitation of excellent educational resources to new heights. Nevertheless, there is a lack of an ideal platform for resource selection and integration in the current crop of information-based teaching tools. To overcome this drawback, the methodology used in this study was described in this research in order to classify student and teacher data more precisely. Ideological and Political Education (IPE) is the source of the data collection. The Modified Hamilton filter is used to preprocess the data during the pre-processing step. The excellent course videos, classroom performance, and interactions are all successfully classified by Dual Attention Graph Convolutional Networks (DAGCN).The neural network's weight parameter is optimized using the Lotus Effect Optimization Algorithm (LEA) to improve the DAGCN. The suggested DAGNN-LEA used with the MATLAB platform. The suggested approach was calculated using performance measures such as accuracy, precision, sensitivity, F-score, computation time, and recall. In comparison to the existing method, the suggested DAGNN-LEA method yields better results in terms of high accuracy 16.65%, 18.85%, and 17.89%, high sensitivity 16.34, 12.23%, and 18.54%, high precision 14.89%, 16.89%, and 18.23%, and low computing time 82.37%, 94.47%, and 87.76%.

Integral reinforcement learning-based dynamic event-triggered safety control for multiplayer Stackelberg–Nash games with time-varying state constraints

In this article, an integral reinforcement learning-based dynamic event-triggered safety control scheme is proposed to tackle the multiplayer Stackelberg–Nash games (MSNG) problem in continuous-time nonlinear systems with time-varying state constraints. Initially, a new barrier function (BF) is introduced by integrating traditional barrier functions with a novel smooth function to address the challenge of time-varying state constraints. Then, the constrained MSNG system is transformed into an unconstrained system through the application of the state transformation technique, which helps to characterize hierarchical decision problems as MSNG problems with the leader and followers. Meanwhile, the integral reinforcement learning (IRL) technique is also applied to ease the demand for precise system dynamics. Moreover, a new dynamic event-triggered control (DETC) mechanism is designed, resulting in coupled dynamic event-triggered Hamilton–Jacobi (HJ) equations. A single critic neural network (NN) is constructed to learn the optimal control laws for the leader and followers. By employing Lyapunov theory, the integral reinforcement learning-based dynamic event-triggered safety control method ensures that the system state and the critic NN weight errors are uniformly ultimately bounded (UUB). Finally, two simulation examples are provided to demonstrate the effectiveness of the proposed algorithm.

Reinforcement Learning for Dual-Control Aircraft Six-Degree-of-Freedom Attitude Control with System Uncertainty

This article proposes a near-optimal control strategy based on reinforcement learning, which is applied to the six-degree-of-freedom (6-DoF) attitude control of dual-control aircraft. In order to solve the problem that the existing reinforcement learning is difficult to apply to the high-dimensional multiple-input multiple-output (MIMO) systems, the Long Short-Term Memory (LSTM) neural network is introduced to replace the polynomial network in the adaptive dynamic programming (ADP) technique. Meanwhile, based on the Lyapunov method, a novel online adaptive updating law of LSTM neural network weights is given, and the stability of the system is verified. In the simulation process, the algorithm proposed in this article is applied to the six-degree-of-freedom attitude control problem of dual-control aircraft with system uncertainty. The simulation results show that the algorithm can achieve near-optimal control.

Approximate optimal and safe coordination of nonlinear second-order multirobot systems with model uncertainties

This paper investigates the approximate optimal coordination for nonlinear uncertain second-order multi-robot systems with guaranteed safety (collision avoidance) Through constructing novel local error signals, the collision-free control objective is formulated into an coordination optimization problem for nominal multi-robot systems. Based on approximate dynamic programming technique, the optimal value functions and control policies are learned by simplified critic-only neural networks (NNs). Then, the approximated optimal controllers are redesigned using adaptive law to handle the effects of robots’ uncertain dynamics. It is shown that the NN weights estimation errors are uniformly ultimately bounded under proper conditions, and safe coordination of multiple robots can be achieved regardless of model uncertainties. Numerical simulations finally illustrate the effectiveness of the proposed controller.

Two-player nonlinear Stackelberg differential game via off-policy integral reinforcement learning

For the nonzero-sum games, players have taken different strategies to achieve the Nash equilibrium. However, regarding hierarchical optimization and asymmetric information, Nash equilibrium is ineffective. The Stackelberg game provides us with a leader–follower strategy to cope with this problem. This paper solved two-player continuous-time nonlinear Stackelberg differential games with unknown system dynamics using an off-policy integral reinforcement learning (IRL) technique. A two-level hierarchy optimal control problem is the Stackelberg differential game, and locating the open-loop Stackelberg equilibrium is equivalent to solving the coupled Hamiltonian-Jacobi (HJ) equations. First, the optimal strategies for the leader and the follower are derived. Then, the closed-loop system’s asymptotic stability of optimal control solution is proved. Based on the policy iteration (PI) algorithm, an off-policy IRL algorithm is developed to solve the Stackelberg game for completely unknown system dynamics. The control strategies and cost functions of the leader and follower are approximated using neural networks (NNs). It is demonstrated that NNs weight errors of off-policy algorithm are uniformly ultimately bounded (UUB). Lastly, two simulation examples demonstrate that the proposed learning algorithm is capable of obtaining two-player Stackelberg equilibrium strategies.

Guidelines for the Regularization of Gammas in Batch Normalization for Deep Residual Networks

L 2 regularization for weights in neural networks is widely used as a standard training trick. In addition to weights, the use of batch normalization involves an additional trainable parameter γ, which acts as a scaling factor. However, L 2 regularization for γ remains an undiscussed mystery and is applied in different ways depending on the library and practitioner. In this article, we study whether L 2 regularization for γ is valid. To explore this issue, we consider two approaches: (1) variance control to make the residual network behave like an identity mapping and (2) stable optimization through the improvement of effective learning rate. Through two analyses, we specify the desirable and undesirable γ to apply L 2 regularization and propose four guidelines for managing them. In several experiments, we observed that applying L 2 regularization to applicable γ increased 1% to 4% classification accuracy, whereas applying L 2 regularization to inapplicable γ decreased 1% to 3% classification accuracy, which is consistent with our four guidelines. Our proposed guidelines were further validated through various tasks and architectures, including variants of residual networks and transformers.

Research on bronze wine vessel classification using improved SSA-CBAM-GNNs.

This article proposes an advanced classification algorithm for bronze drinking utensils, taking into account the complexity of their cultural characteristics and the challenges of dynasty classification. The SSA-CBAM-GNNs algorithm integrates the Sparrow Search Algorithm (SSA), Spatial and Spectral Attention (CBAM) modules, and Graph Neural Networks (GNNs). The CBAM module is essential for optimizing feature extraction weights in graph neural networks, while SSA enhances the weighted network and expedites the convergence process. Experimental results, validated through various performance evaluation indicators, illustrate the outstanding performance of the improved SSA-CBAM-GNNs algorithm in accurately identifying and classifying cultural features of bronze drinking utensils. Comparative experiments confirm the algorithm's superiority over other methods. Overall, this study proposes a highly efficient identification and classification algorithm, and its effectiveness and excellence in extracting and identifying cultural features of bronze drinking utensils are experimentally demonstrated.

Co-design of SLC/MLC FeFET-based highly error-tolerant low-power CiM and strong lottery ticket hypothesis-based algorithm

This study co-designs single-level cell (SLC) mask and multilevel cell (MLC) weight twin FeFET devices and a strong lottery ticket hypothesis (SLTH)-based neural network (NN) algorithm to achieve highly error-tolerant low-power Computation-in-Memory (CiM). The SLC mask FeFET masks or transfers the NN weight stored in the MLC weight FeFET, and the masked NN weight reduces the CiM power consumption. The proposed SLC mask FeFETs, which are trained, and MLC weight FeFETs, in which V TH are uniformly randomized, achieve 87% inference accuracy against 10-year data retention and read disturb. The SLC mask FeFETs show 86% inference accuracy even at 2000 endurance cycles. In addition, shared-bottom-select-gate (BSG) SLTH CiM and common-mask SLTH CiM for the NN convolutional layer are proposed to reduce the CiM area by sharing BSG and mask FeFET. Moreover, NN weight mapping schemes for SLTH CiM are proposed. The proposed mapping schemes show a tradeoff between inference accuracy and CiM area. One of the schemes reduces the CiM area by 45% with a 9.1% accuracy loss.