Singing melody extraction from polyphonic music is a complex but important task in music information retrieval. Most existing models struggle with adaptive multi-scale feature extraction and long-range dependency modeling, often resulting in common artifacts such as octave errors and pitch fragmentation. To address this, we propose the Cross-Scale Transformer Network (CSTNet), a novel deep learning architecture that dynamically captures and fuses multi-scale temporal and spectral features. The CSTNet introduces two key components: the Multi scale Time-Frequency Aggregation (MF-TFA) module for adaptive local feature representation across resolutions, and Feature Cross Transformer Block (FCTB), which enables explicit cross scale interaction through a global channel-wise cross-attention mechanism. Experimental results show that our proposed method outperforms five compared state-of-the-art methods, achieving overall accuracy scores of 87.3%, 90.5%, and 77.4% on the ADC 2004, MIREX 05, and MEDLEY DB datasets, respectively. Visualization results demonstrate that CSTNet effectively reduces octave errors.

Kullback-Leibler (KL) divergence plays a central role in hypothesis testing. It gives a measure of the statistical distance between two probability distributions. In the distributed detection problem, it is used as a design criterion in the absence of the information regarding the fusion center's (FC) decision rule: The local sensor decision rules are designed to maximize the KL divergence between the distributions of quantized messages sent to the FC under alternative and null hypotheses. In decision making tasks involving humans, subjective perception of probability values due to behavioral biases needs to be taken into account. In this letter, the notion of behavioral KL divergence is proposed. The statistical distance between two distributions is computed based on the perceived values of the probabilities, which are obtained from the actual probabilities using the probability weighting function employed in prospect theory. It is proved that the behavioral KL divergence between the distributions of the quantized decision at the output of a detector under both hypotheses is maximized by either the Neyman-Pearson (NP) rule or flipped Neyman-Pearson (FNP) rule for any fixed false alarm probability. Based on this result, it is also established that under a constraint on the average perceived false alarm probability, the average behavioral KL divergence is maximized by time-sharing between at most two single-threshold likelihood-ratio tests, each of which is either an NP or an FNP rule. The theoretical results are supported by numerical examples.

Even though recent studies have demonstrated significant improvements in 3D human mesh reconstruction, most methods still face challenges under occlusions. To mitigate this issue, multi-view human mesh reconstruction methods have been proposed, which leverage additional visual cues by aggregating the complementary information. However, in scenes with close human interactions, the identity of each person is hard to be preserved across different views, thus leading to incorrect merging of visual features between individuals. To address this limitation, we propose an identity-aware fusion scheme for interactive human mesh reconstruction. The key idea is to maintain the identity of each person across different views based on the shape feature from the anchor view, which is determined by the highest visibility of each person. This helps the model to distinguish features encoded from each person during the aggregation process, even under close interactions with severe occlusions, leading to the reliable reconstruction of 3D human meshes. Furthermore, we design a weighting scheme to adaptively fuse joints belonging to the same identity across different views while allowing the model to focus more on visible joints. Experimental results on benchmark datasets show that the proposed method efficiently improves the performance of interactive human mesh reconstruction.

Factor graphs are fundamental to large-scale estimation in robotics and computer vision, but their underlying normal equations incur significant computational costs. To overcome this, we introduce an Extended Sparse Linear System (ESLS) that seamlessly integrates state variables with auxiliary measurements. This framework fundamentally restructures the normal equations via algebraic-topological isomorphism and structured marginalization, achieving intrinsic computational optimization. The ESLS features a skew-symmetric block-structured matrix that is isomorphic to the factor graph topology, ensuring that algebraic marginalization directly implements graphical inference. We prove that classical normal equations are a special case of ESLS marginalization through Schur complements. Moreover, we characterize novel theoretical properties of the ESLS, derive closed-form self- and cross-information updates, and formalize how information flows to adjacent graph elements during Schur complement operations. Overall, the ESLS establishes a unified and computationally efficient algebraic foundation for large-scale inference, bridging graphical models and numerical linear algebra in a principled way.

Deep functionality hiding is an emerging technique that embeds confidential or sensitive functions within seemingly benign deep learning models (DLMs), which perform ordinary machine learning tasks. This enables such models to execute covert tasks while remaining undetected. Despite the rapid progress in deep functionality hiding, countermeasures remain unexplored. In this paper, we propose Distribution Offset Analysis (DOA), a novel method for detecting hidden functionalities in DLMs. Our key insight is that the weight distribution of a benign DLM typically follows a Gaussian distribution, whereas a container DLM with hidden functionalities exhibits notable statistical deviations from this Gaussian pattern. In our methodology, we first compute the distributional distance (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.,</i> offsets) between the model's weights and an ideal Gaussian distribution. We then fuse these offsets with weight features into a unified representation, which is subsequently used to train a meta-classifier for hidden functionality detection. Through extensive experiments, we demonstrate the effectiveness of the proposed DOA method, which achieves an average detection rate of over 87% against existing state-of-the-art deep functionality hiding techniques.

Adaptive modulation and coding (AMC) systems require the transmission of control signals, thereby reducing overall system transmission efficiency. The channel coding blind recognition technique is key to solving this problem. This paper proposes a reduced-complexity method for blind recognition of low-density parity-check (LDPC) coding parameters within a given candidate set, thereby enhancing the work efficiency of AMC systems. This paper applies a method based on code rate for classification evaluation, circumventing superfluous calculations for several candidate parity-check matrices. Furthermore, computational complexity is reduced by applying the offset min-sum algorithm (OMSA) to the parity-check stage. Subsequently, the Z-score is used to measure the difference between the actual data and the theoretical distribution. Compared with the best existing recognition methods, the proposed algorithm offers clear advantages in computational complexity and is virtually identical in recognition performance.

Eye tracking has been widely utilized for the applications in disease diagnosis, health management and rehabilitation treatment. However, blinking may influence the performance of pupil localization dramatically. To address this problem, we propose an efficient algorithm termed space-adaptive eye tracking that can detect pupil coordinates and blink event jointly with high precision. The algorithm uses a shared network to extract features of templates and search images. Feature similarity is then compared through two task-specific branches to obtain both pupil coordinates and existence confidence. On this basis, to solve the problem of confidence decline caused by the pupil moving to the edge of the active space with a large deformation when shooting with a near-eye camera, the algorithm divides the pupil active space and records the templates of each sub-region, which can be used to select appropriate templates for the edge regions. By using this space-adaptive method, the confidence will only decrease when blinking, thus enabling the detection of blinking behavior using the confidence. Experiments have shown that the proposed method is superior to the existing method.

Perception and cognition are core processes that transform external sensory signals into internal representations for knowledge construction and understanding, and in visually impaired individuals, this transformation is reorganized through auditory and tactile feedback. To explain how perceptual information evolves into stable cognitive representations under limited sensory bandwidth, this study proposes Phase Transition Hypothesis of Perception and Cognition. The proposed hypothesis models the perceptual–cognitive process as a dynamic phase transition, in which sensory information evolves from fragmented perception into organized cognition. To counteract perceptual bias induced by information collapse, the Perceptual Dynamic Optimization Mechanism adaptively regulates sensory deviations to stabilize the perceptual–cognitive transition, whereas the Cognitive Potential Model, derived from the Free-Energy Principle, elucidates how stable and self-organizing cognition emerges from this dynamic process. A cognitive simulation system and a blind writing navigation experiment are conducted to validate the hypothesis. Experiments demonstrate the proposed adaptive correction of perceptual bias and the phase transition mechanism from perception to cognition.