Multi-agent path finding (MAPF) is a safety-critical scenario where the goal is to secure collision-free trajectories from initial to desired locations. However, due to system complexity and uncertainty, integrating learning-based controllers with MAPF is challenging and cannot theoretically guarantee the safety of the learned controllers. In response, our study proposes a verified safe multi-agent neural control (VSMANC) approach for MAPF, focusing on the unified training of Decentralized Control Barrier Functions (DCBF) and controllers to enhence safety. VSMANC enables all agents to concurrently learn controllers and DCBFs using a unified loss function designed to maximize safety, adhere to standard control policies, and incorporate path-finding-related heuristics. We also propose a formal verification-guided retraining process to both verify the properties of the learned DCBFs and generate counterexamples for retraining, thereby providing a verified safety guarantee. We validate our approach through shape formation experiments and UAV simulations, demonstrating significant improvements in safety and effectiveness in complex multi-agent environments.

In image recognition, improving the interpretability of the recognition model can help people understand the model better and increase the trust of human beings for model prediction. The prototype-based interpretable model is a self-explanatory image recognition model that simulates the evidence reasoning used in human recognition. Each prototype is evidence that contains category features, which can help in determining the image category. Based on the prototype-based model, this paper introduces a deep interpretable network architecture called the high-quality prototypical part network (HQProtoPNet). Compared to existing work, this paper adds random erasing to enhance the picture, helping to improve prototype generation and increase model prediction. The multiple scale conversion operation is also introduced and the similarity calculation is improved to make the prototype have multiscale information and matching ability. Furthermore, the accuracy of HQProtoPNet can reach or even exceed the accuracy of several black-box models. Additionally, due to the improvement in the quality of the prototype, the model's prediction accuracy is improved by stacking without reducing the interpretability of the stacked model, which gives the model real stackability.

Pedicle screw placement is a standard but technically demanding operation. Improper screw placement can cause nerve damage and postoperative complications. Usually, the computed tomography (CT) image of the patient’s spine is analyzed in joint surgical planning, and next the surgeons complete the path planning manually for screw placement, which is an error-prone, time-consuming, and labour-intensive process. This article aims to realize the automatic lumbar landmarks localization and segmentation for pedicle screw placement. For this purpose, we propose a coarse-to-fine framework based on deep learning. First, the lumbar part is automatically extracted from the 3D CT image of the patient’s spine, by using a CNN to predict the lumbar vertebrae centroid landmarks. Then another network is used to predict for each lumbar vertebra the midsagittal and pedicle landmarks critical to screw planning. These landmarks are used to construct bounding boxes for the five lumbar vertebrae, which are then cropped separately and rotated horizontally to train a segmentation network. Using the landmarks and vertebral body segmentation mask, we can localize the plane and initial path of screw placement, as well as the screw geometry parameters. Experimental results show that our framework can locate the lumbar vertebrae landmarks and segment the body with high accuracy, and plan an initial screw path as critical assistant information for orthopaedic clinicians. Moreover, this approach also facilitates automatic optimization of screw paths and has potential application value in preoperative planning automation.

Abstract We provide a new approach to synthesize controllers for nonlinear continuous dynamical systems with control against safety properties. The controllers are based on neural networks (NNs). To certify the safety property we utilize barrier functions, which are represented by NNs as well. We train the controller-NN and barrier-NN simultaneously, achieving a verification-in-the-loop synthesis. We provide a prototype tool nncontroller with a number of case studies. The experiment results confirm the feasibility and efficacy of our approach.

This paper presents an approach of safety verification based on neural networks for continuous dynamical systems which are modeled as a system of ordinary differential equations. We adopt the deductive verification methods based on barrier certificates. These are functions over the states of the dynamical system with certain constraints the existence of which entails the safety of the system under consideration. We propose to represent the barrier function by neural networks and provide a comprehensive synthesis framework. In particular, we devise a new type of activation functions, i.e., Bent-ReLU, for the neural networks; we provide sampling based approaches to generate training sets and formulate the loss functions for neural network training which can capture the essence of barrier certificate; we also present practical methods to check a learnt candidate barrier certificate against the criteria of barrier certificates as a formal guarantee. We implement our approaches via proof-of-concept experiments with encouraging results.

In this paper, we propose an adaptive variable exponent level set method based on Legendre polynomials for object segmentation in complex visual environment. First, we use a set of Legendre basis functions to approximate the region intensity, which enable us to accommodate heterogeneous objects. Second, an improved function is presented to update exponent adaptively and ensures the image gradient information embedding into the model easily. The proposed method is robust to low contrast, blurred boundaries, noise and the 10-cation of initial contour, and sufficient in handling large scale intensity variations. Experimental results demonstrate that the proposed method can achieve relatively high segmentation accuracy and less computational time.

In this work, the expectation of wavelet transform coefficients is used for estimating an image sharpness. It's based on the observation that the greater the probability of big detail coefficients, the more pixels appear sharply, and consequently, the sharper the image. Specifically, an input image is firstly decomposed into three directional sub-bands by a separable discrete wavelet transform. Then these directional sub-bands are viewed as three random variables, and their expectations are computed. Finally, The proposed sharpness index is the weighted sum of three expectations. The experiments show that, despite its simplicity, the proposed sharpness index is competitive with the current best-performance techniques for no-reference image sharpness estimation.

This paper proposes a novel method for estimating visual saliency based on a typical agreement that image saliency depends mainly on local and global contrast from various feature channels. We compute the contrast between image patches on different low-level feature maps which are generated by color space conversion and support value transform. To obtain the representative measurement effectively, we calculate the dissimilarity in a reduced dimensional principal component space. In addition, our method may be easily extended for more conspicuous feature channels in an efficient manner. Experimental results on two public available human eye fixation datasets demonstrate that our method outperforms other seven state-of-the-art saliency models.

In this paper, we consider an extended concept of invariant for polynomial dynamical systems (PDSs) with domain and initial condition, and establish a sound and complete criterion for checking semi-algebraic invariants (SAIs) for such PDSs. The main idea is encoding relevant dynamical properties as conditions on the high order Lie derivatives of polynomials occurring in the SAI. A direct consequence of this criterion is a relatively complete method of SAI generation based on template assumption and semi-algebraic constraint solving. Relative completeness means if there is an SAI in the form of a predefined template, then our method can indeed find one.