Explore the vast range of titles available.

In this paper, we propose a new approach to train a deep neural network with multiple intermediate auxiliary classifiers, branching from it. These ‘multi-exits’ models can be used to reduce the inference time by performing early exit on the intermediate branches, if the confidence of the prediction is higher than a threshold. They rely on the assumption that not all the samples require the same amount of processing to yield a good prediction. In this paper, we propose a way to train jointly all the branches of a multi-exit model without hyper-parameters, by weighting the predictions from each branch with a trained confidence score. Each confidence score is an approximation of the real one produced by the branch, and it is calculated and regularized while training the rest of the model. We evaluate our proposal on a set of image classification benchmarks, using different neural models and early-exit stopping criteria.

This paper challenges the prevailing assumption that Mixture of Experts (MoE) consistently improves computational efficiency through a systematic evaluation of MoE variants in Transformer models. We implement and compare three approaches: basic MoE, top-k routing, and capacity-factored routing, each progressively addressing load-balancing challenges. Our experiments reveal critical trade-offs between performance and efficiency: while MoE models maintain validation performance comparable to baselines, they require significantly longer training times (a 50% increase) and demonstrate reduced inference speeds (up to 56% slower). Analysis of routing behavior shows that even with load-balancing techniques, expert utilization remains unevenly distributed. These findings provide empirical evidence that MoE’s computational benefits are highly dependent on model scale and task characteristics, challenging common assumptions about sparse architectures and offering crucial guidance for adaptive neural architecture design across different computational constraints.

In this paper, for the Steklov-Lamé eigenvalue problem, we propose a multigrid discretization scheme of discontinuous Galerkin method based on the shifted-inverse iteration. Based on the existing a priori error estimates, we give the error estimates for the proposed scheme and prove that the resulting approximations can achieve the optimal convergence order when the mesh sizes fit into some relationships. Finally, we combine the multigrid scheme and adaptive procedure to present some numerical examples which indicate that our scheme are locking-free and efficient for computing Steklov-Lamé eigenvalues.

The computational cost of evaluating a neural network usually only depends on design choices such as the number of layers or the number of units in each layer and not on the actual input. In this work, we build upon deep Residual Networks (ResNets) and use their properties to design a more efficient adaptive neural network building block. We propose a new architecture, which replaces the sequential layers with an iterative structure where weights are reused multiple times for a single input image, reducing the storage requirements drastically. In addition, we incorporate an adaptive computation module that allows the network to adjust its computational cost at run time for each input sample independently. We experimentally validate our models on image classification, object detection and semantic segmentation tasks and show that our models only use their full capacity for the hardest input samples and are more efficient on average.

Collaborative recommender systems allow personalization for e-commerce by exploiting similarities and dissimilarities among customers' preferences. We investigate the use of association rule mining as an underlying technology for collaborative recommender systems. Association rules have been used with success in other domains. However, most currently existing association rule mining algorithms were designed with market basket analysis in mind. Such algorithms are inefficient for collaborative recommendation because they mine many rules that are not relevant to a given user. Also, it is necessary to specify the minimum support of the mined rules in advance, often leading to either too many or too few rules; this negatively impacts the performance of the overall system. We describe a collaborative recommendation technique based on a new algorithm specifically designed to mine association rules for this purpose. Our algorithm does not require the minimum support to be specified in advance. Rather, a target range is given for the number of rules, and the algorithm adjusts the minimum support for each user in order to obtain a ruleset whose size is in the desired range. Rules are mined for a specific target user, reducing the time required for the mining process. We employ associations between users as well as associations between items in making recommendations. Experimental evaluation of a system based on our algorithm reveals performance that is significantly better than that of traditional correlation-based approaches.

In this study, a dynamic stochastic resonance (DSR)-based technique in spatial domain has been proposed for the enhancement of dark- and low-contrast images. Stochastic resonance (SR) is a phenomenon in which the performance of a system (low-contrast image) can be improved by addition of noise. However, in the proposed work, the internal noise of an image has been utilised to produce a noise-induced transition of a dark image from a state of low contrast to that of high contrast. DSR is applied in an iterative fashion by correlating the bistable system parameters of a double-well potential with the intensity values of a low-contrast image. Optimum output is ensured by adaptive computation of performance metrics – relative contrast enhancement factor (F), perceptual quality measures and colour enhancement factor. When compared with the existing enhancement techniques such as adaptive histogram equalisation, gamma correction, single-scale retinex, multi-scale retinex, modified high-pass filtering, edge-preserving multi-scale decomposition and automatic controls of popular imaging tools, the proposed technique gives significant performance in terms of contrast and colour enhancement as well as perceptual quality. Comparison with a spatial domain SR-based technique has also been illustrated.

Zero Trust Network Access (ZTNA) has emerged as a fundamental paradigm for securing cloud-native and distributed computing environments. However, existing ZTNA implementations remain largely limited by static policy enforcement and opaque machine-learning-based anomaly detection mechanisms, which often lack contextual adaptability, policy awareness, and interpretable decision-making capabilities. These limitations create significant challenges in dynamic multi-cloud environments where access behavior continuously evolves and security decisions must be both accurate and explainable. To address these challenges, this study proposes Cognitive ZTNA framework, a unified neuro-symbolic trust enforcement framework that integrates transformer-based behavioral trust modeling with ontology-guided symbolic reasoning. The proposed architecture enables continuous trust evaluation by combining behavioral access patterns with explicit policy semantics through a hybrid trust fusion mechanism. This design allows the system to capture long-range behavioral dependencies while maintaining policy-compliant and interpretable access control decisions. The framework is evaluated using the CloudZT-Bench-2025 dataset, comprising 4.2 million cross-platform access events derived from enterprise security telemetry, AWS CloudTrail logs, and simulated adversarial scenarios. Experimental results demonstrate that Cognitive ZTNA achieves Precision = 0.96, Recall = 0.93, and F1-score = 0.95, significantly outperforming rule-based and machine-learning baselines while reducing the false positive rate to 0.03. In addition, the system maintains real-time feasibility with an average decision latency of 24 ms and explanation latency below 5 ms, while achieving 92% analyst-rated explanation sufficiency. These findings demonstrate that integrating behavioral intelligence with symbolic policy reasoning enables adaptive, interpretable, and policy-aware Zero Trust enforcement. The proposed framework therefore provides a practical foundation for next-generation ZTNA systems capable of supporting secure, transparent, and context-aware access control in modern cloud environments.

Various finite element families for the Brinkman flow (or Stokes–Darcy flow) are tested numerically. Particularly, the effect of small permeability is studied. The tested finite elements are the MINI element, the Taylor–Hood element, and the stabilized equal order methods. The numerical tests include both a priori analysis and adaptive methods.

The Rabinovich system, describing the process of interaction between waves in plasma, is considered. It is shown that the Rabinovich system can exhibit a hidden attractor in the case of multistability as well as a classical self-excited attractor. The hidden attractor in this system can be localized by analytical/numerical methods based on the continuation and perpetual points. The concept of finite-time Lyapunov dimension is developed for numerical study of the dimension of attractors. A conjecture on the Lyapunov dimension of self-excited attractors and the notion of exact Lyapunov dimension are discussed. A comparative survey on the computation of the finite-time Lyapunov exponents and dimension by different algorithms is presented. An adaptive algorithm for studying the dynamics of the finite-time Lyapunov dimension is suggested. Various estimates of the finite-time Lyapunov dimension for the hidden attractor and hidden transient chaotic set in the case of multistability are given.