To diagnose the transport-layer fault in RDMA for ensuring the reliability of data center networks (DCN), the existing RDMA pingmesh (R-Pingmesh) technology often requires manual analysis to identify the root causes of faults. However, the fault identification faces challenges in few-shot training for rare faults and prominent fault-normal data imbalance. To address the challenges, this paper proposes an LSTMbased fault classifier: The model pre-training phase captures general fault patterns to alleviate the difficulty of few-shot training for rare faults. To deploy the model in the target DCN, it uses supervised fine-tuning (SFT) with the data obtained from the network to adapt to the different network topological characteristics. Through such training, this classifier enables finegrained RDMA transport-layer fault diagnosis across multiple network topologies, achieving accurate classification of various fault types and fine-grained localization from “abnormal paths” to “fault segments”. This collaborative design of pre-training and topology adaptation achieves key challenges of few-shot learning under data distribution imbalance for accurate distinction and localization of RDMA transport-layer faults. Our experiments show that this method performs excellently on datasets from different typical DCN, providing an effective technical solution for the automated diagnosis of RDMA transport-layer faults.

Electrocardiogram (ECG), a key bioelectrical time-series signal, is crucial for assessing cardiac health and diagnosing various diseases. Given its time-series format, ECG data is often incorporated into pre-training datasets for large-scale time-series model training. However, existing studies often overlook its unique characteristics and specialized downstream applications, which differ significantly from other time-series data, leading to an incomplete understanding of its properties. In this paper, we present an in-depth investigation of ECG signals and establish a comprehensive benchmark, which includes (1) categorizing its downstream applications into four distinct evaluation tasks, (2) identifying limitations in traditional evaluation metrics for ECG analysis, and introducing a novel metric; (3) benchmarking state-of-the-art time-series models and proposing a new architecture. Extensive experiments demonstrate that our proposed benchmark is comprehensive and robust. The results validate the effectiveness of the proposed metric and model architecture, which establish a solid foundation for advancing research in ECG signal analysis.

The image-based visual servoing (IBVS) of manipulators is important for intelligent manipulation using visual feedbacks. While the traditional IBVS methods for manipulators require the knowledge of the depth information in the interaction matrix, in this article, we propose a novel IBVS method for manipulators without depth estimation by leveraging the property of the associated image Jacobian. Because of a novel transformation, the IBVS problem is converted into a convex optimization problem subject to the kinematic constraint, joint constraints, and other constraints that are not explicitly related to the depth information. The problem is then solved by developing a recurrent neural network of global asymptotic convergence, and a dynamic neural control law without depth estimation emerges for the IBVS of manipulators. The theoretical guarantee and simulation results are provided to show the efficacy of the proposed method.

Dynamic facial expression recognition (DFER) is a rapidly developing field that focuses on recognizing facial expressions in video sequences. However, the complex temporal modeling caused by noisy frames, along with the limited training data significantly hinder the further development of DFER. Previous efforts in this domain have been limited as they tackled these issues separately. Inspired by recent advances of pretrained vision-language models (e.g., CLIP), we propose to leverage it to jointly address the two limitations in DFER. Since the raw CLIP model lacks the ability to model temporal relationships and determine the optimal task-related textual prompts, we utilize DFER-specific domain knowledge, including characteristics of temporal correlations and relationships between facial behavior descriptions at different levels, to guide the adaptation of CLIP to DFER. Specifically, we propose enhancements to CLIP's visual encoder through the design of a hierarchical video encoder that captures both short- and long-term temporal correlations in DFER. Meanwhile, we align facial expressions with action units through prior knowledge to construct semantically rich textual prompts, which are further enhanced with visual contents. Furthermore, we introduce a class-aware consistency regularization mechanism that adaptively filters out noisy frames, bolstering the model's robustness against interference. Extensive experiments on three in-the-wild dynamic facial expression datasets demonstrate that our method outperforms the state-of-the-art DFER approaches. The code is available at https://github.com/liliupeng28/DK-CLIP.

High-frequency trading proposes new challenges to classical portfolio selection problems. Especially, the timely and accurate solution of portfolios is highly demanded in financial market nowadays. This article makes progress along this direction by proposing novel neural networks with softmax equalization to address the problem. To the best of our knowledge, this is the first time that softmax technique is used to deal with equation constraints in portfolio selections. Theoretical analysis shows that the proposed method is globally convergent to the optimum of the optimization formulation of portfolio selection. Experiments based on real stock data verify the effectiveness of the proposed solution. It is worth mentioning that the two proposed models achieve 5.50% and 5.47% less cost, respectively, than the solution obtained by using MATLAB dedicated solvers, which demonstrates the superiority of the proposed strategies.

Multi-class object recognition is a critical capability for an intelligence robot to perceive its environment. In this paper, a new approach consisting of a number of modular neural networks is proposed to recognize multiple classes of objects for a robotic system. The population of the modular neural networks depends on the class number of the objects to be recognized and each modular network only focuses on learning one object class. For each modular neural network, both the bottom-up (sensory-driven) and top-down (expectation-driven) pathways are fused together, and a supervised learning algorithm is applied to update corresponding weights of both pathways. Furthermore, two different training strategies are evaluated: positive-only training and positive-and-negative training. Experiments on visual image recognition demonstrate the efficiencies of the proposed approach and the corresponding training strategies.

To improve the efficiency of multi-object recognition in complex scenes, a hybrid system is proposed to learn the concurrencies and spatial relationships among different objects, and to apply such relationships for better recognitions. The hybrid system includes a bottom-up saliency map to generate regions of interest (ROIs), an independent classifiers to recognize these ROIs based on object appearances, and a what-where-memory (WWM) to cast the top-down knowledge of object relationships to help the recognitions provided by independent classifiers. The WWM learns not only the concurrencies but also the spatial layouts of different classes, which can filter out the classes that unlikely appear, and distinguish the correct class from ambiguous classes provided by independent classifiers. Experiments of multi-object recognition on two well-known image datasets demonstrate the efficiency of the proposed system.

In this paper, an artificial neural network model is built up with two pathways: bottom-up sensory-driven pathway and top-down expectation-driven pathway, which are fused to train the neural network for visual object recognition. During the supervised learning process, the bottom-up pathway generates hypotheses as network outputs. Then target label will be applied to update the bottom-up connections. On the other hand, the hypotheses generated by the bottom-up pathway will produce expectations on the sensory input through the top-down pathway. The expectations will be constrained by the real data from the sensory input which can be used to update the top-down connections accordingly. This two-pathway based neural network can also be applied to semi-supervised learning with both labeled and unlabeled data, where the network is able to generate hypotheses and corresponding expectations. Experiments on visual object recognition suggest that the proposed neural network model is promising to recover the object for the cases with missing data in sensory inputs.

This paper presents a new object tracking algorithm by using the particle swarm optimization (PSO), which is a bio-inspired population-based searching algorithm. Firstly the potential solutions of the problem are projected into a state space called solution space where every point in the space presents a potential solution. Then a group of particles are initialized and start searching in this solution space. The swarm particles search for the best solution within this solution space using the Particle Swarm Optimization (PSO) algorithm. An accumulative histogram of the object appearance is applied to build up the fitness function for the interested object pattern. Eventually the swarming particles driven by the fitness function converge to the optimal solution. Experimental results demonstrate that the proposed PSO method is efficient and robust in visual object tracking under dynamic environments

This paper presents a new object tracking algorithm that embeds swarming particles into generic particle filter framework to achieve more robustness and flexibility. Firstly a group of particles associated with potential solutions are initialized in a high-dimensional space. Then particle swarm optimization (PSO) is used to drive particles flying. The object is tracked when the particles reach convergence. This PSO-based algorithm contains resample, similarity measure, and integration together such that the degeneracy problem of particle filter can be avoided. Furthermore, a multiple feature model is proposed for object description to enhance the tracking accuracy and efficiency. The proposed algorithm is independent with specific objects and can be used for any free-selected object tracking. Some experimental results demonstrate efficiency and robustness of the algorithm.

This paper presents a robust tracking algorithm using an adaptive tracking window associated with five parameters, where the parameters of the tracking window are optimized by a particle swarm optimization (PSO) algorithm. Basically, the optimization of a tracking window is transformed into a searching algorithm in a five-dimension feature space, which constrains the possibilities of the window. Particles associated with different parameters fly around the searching space independently, while they are sharing information from the society and adjust their behaviors to achieve the global optimization, which means the most optimized parameters for the tracking window. Appearance histogram is employed to calculate the fitness function for particles, where the distance between histograms is measured by histogram intersection. Estimated people motion is utilized to expedite the convergence of particles. Experimental results of people tracking demonstrate that the algorithm is efficient, robust, and adaptive to various rigid and non-rigid people motions