End-to-end backpropagation remains the dominant training paradigm in deep learning, yet it suffers from inherent drawbacks, including update locking, high GPU memory consumption, and limited biological plausibility. Supervised local learning alleviates these issues by dividing the network into multiple blocks and training each block independently with an auxiliary network. However, gradient isolation also weakens the influence of downstream representations on earlier blocks, often resulting in a clear accuracy gap to end-to-end training. We propose Momentum Auxiliary Network++ (MAN++), a scalable framework that improves supervised local learning via a lightweight parameter-space transfer between adjacent blocks. MAN++ employs the exponential moving average (EMA) of parameters from adjacent blocks to propagate contextual information across the network. To address feature mismatches arising from direct EMA parameter transfer, we introduce a learnable scaling bias, which compensates feature statistics mismatch and stabilizes the transfer. Extensive experiments on image classification, object detection, and semantic segmentation across multiple architectures illustrate that MAN++ achieves accuracy on par with end-to-end training while substantially reducing GPU memory usage. These results position MAN++ as a practical and effective alternative to conventional backpropagation, offering new insights into scalable supervised local learning for vision tasks.

Deep neural networks (DNNs) typically employ an end-to-end (E2E) training paradigm which presents several challenges, including high GPU memory consumption, inefficiency, and difficulties in model parallelization during training. Recent research has sought to address these issues, with one promising approach being local learning. This method involves partitioning the backbone network into gradient-isolated modules and manually designing auxiliary networks to train these local modules. Existing methods often neglect the interaction of information between local modules, leading to myopic issues and a performance gap compared to E2E training. To address these limitations, we propose the Multilaminar Leap Augmented Auxiliary Network (MLAAN). Specifically, MLAAN comprises Multilaminar Local Modules (MLM) and Leap Augmented Modules (LAM). MLM captures both local and global features through independent and cascaded auxiliary networks, alleviating performance issues caused by insufficient global features. However, overly simplistic auxiliary networks can impede MLM's ability to capture global information. To address this, we further design LAM, an enhanced auxiliary network that uses the Exponential Moving Average (EMA) method to facilitate information exchange between local modules, thereby mitigating the shortsightedness resulting from inadequate interaction. The synergy between MLM and LAM has demonstrated excellent performance. Our experiments on the CIFAR-10, STL-10, SVHN, and ImageNet datasets show that MLAAN can be seamlessly integrated into existing local learning frameworks, significantly enhancing their performance and even surpassing end-to-end (E2E) training methods, while also reducing GPU memory consumption.

Most earlier researches on talking face generation have focused on the synchronization of lip motion and speech content. However, head pose and facial emotions are equally important characteristics of natural faces. While audio-driven talking face generation has seen notable advancements, existing methods either overlook facial emotions or are limited to specific individuals and cannot be applied to arbitrary subjects. In this paper, we propose a novel one-shot Talking Head Generation framework (SPEAK) that distinguishes itself from the general Talking Face Generation by enabling emotional and postural control. Specifically, we introduce Inter-Reconstructed Feature Disentanglement (IRFD) module to decouple facial features into three latent spaces. Then we design a face editing module that modifies speech content and facial latent codes into a single latent space. Subsequently, we present a novel generator that employs modified latent codes derived from the editing module to regulate emotional expression, head poses, and speech content in synthesizing facial animations. Extensive trials demonstrate that our method ensures lip synchronization with the audio while enabling decoupled control of facial features, it can generate realistic talking head with coordinated lip motions, authentic facial emotions, and smooth head movements. The demo video is available: https://anonymous.4open.science/r/SPEAK-8A22.

Lane detection is the cornerstone of autonomous driving. Although existing methods have achieved promising results, there are still limitations in addressing challenging scenarios such as abnormal weather, occlusion, and curves. These scenarios with low visibility usually require to rely on the broad information of the entire scene provided by global semantics and local texture information to predict the precise position and shape of the lane lines. In this paper, we propose a Global Semantic Enhancement Network for lane detection, which involves a complete set of systems for feature extraction and global features transmission. Traditional methods for global feature extraction usually require deep convolution layer stacks. However, this approach of obtaining global features solely through a larger receptive field not only fails to capture precise global features but also leads to an overly deep model, which results in slow inference speed. To address these challenges, we propose a novel operation called the Global feature Extraction Module (GEM). Additionally, we introduce the Top Layer Auxiliary Module (TLAM) as a channel for feature distillation, which facilitates a bottom-up transmission of global features. Furthermore, we introduce two novel loss functions: the Angle Loss, which account for the angle between predicted and ground truth lanes, and the Generalized Line IoU Loss function that considers the scenarios where significant deviations occur between the prediction of lanes and ground truth in some harsh conditions. The experimental results reveal that the proposed method exhibits remarkable superiority over the current state-of-the-art techniques for lane detection.Our codes are available at:https://github.com/crystal250/GSENet.

BACKGROUND: With the continuous advances in third-generation sequencing technology and the increasing affordability of next-generation sequencing technology, sequencing data from different sequencing technology platforms is becoming more common. While numerous benchmarking studies have been conducted to compare variant-calling performance across different platforms and approaches, little attention has been paid to the potential of leveraging the strengths of different platforms to optimize overall performance, especially integrating Oxford Nanopore and Illumina sequencing data. RESULTS: We investigated the impact of multi-platform data on the performance of variant calling through carefully designed experiments with a deep learning-based variant caller named Clair3-MP (Multi-Platform). Through our research, we not only demonstrated the capability of ONT-Illumina data for improved variant calling, but also identified the optimal scenarios for utilizing ONT-Illumina data. In addition, we revealed that the improvement in variant calling using ONT-Illumina data comes from an improvement in difficult genomic regions, such as the large low-complexity regions and segmental and collapse duplication regions. Moreover, Clair3-MP can incorporate reference genome stratification information to achieve a small but measurable improvement in variant calling. Clair3-MP is accessible as an open-source project at: https://github.com/HKU-BAL/Clair3-MP . CONCLUSIONS: These insights have important implications for researchers and practitioners alike, providing valuable guidance for improving the reliability and efficiency of genomic analysis in diverse applications.

Accurate identification of genetic variants from family child-mother-father trio sequencing data is important in genomics. However, state-of-the-art approaches treat variant calling from trios as three independent tasks, which limits their calling accuracy for Nanopore long-read sequencing data. For better trio variant calling, we introduce Clair3-Trio, the first variant caller tailored for family trio data from Nanopore long-reads. Clair3-Trio employs a Trio-to-Trio deep neural network model, which allows it to input the trio sequencing information and output all of the trio's predicted variants within a single model to improve variant calling. We also present MCVLoss, a novel loss function tailor-made for variant calling in trios, leveraging the explicit encoding of the Mendelian inheritance. Clair3-Trio showed comprehensive improvement in experiments. It predicted far fewer Mendelian inheritance violation variations than current state-of-the-art methods. We also demonstrated that our Trio-to-Trio model is more accurate than competing architectures. Clair3-Trio is accessible as a free, open-source project at https://github.com/HKU-BAL/Clair3-Trio.

SUMMARY: Circular consensus sequencing reads are promising for the comprehensive detection of structural variants (SVs). However, alignment-based SV calling pipelines are computationally intensive due to the generation of complete read-alignments and its post-processing. Herein, we propose a SKeleton-based analysis toolkit for Structural Variation detection (SKSV). Benchmarks on real and simulated datasets demonstrate that SKSV has an order of magnitude of faster speed than state-of-the-art SV calling approaches; moreover, it achieves higher F1 scores for various types of SVs. AVAILABILITY AND IMPLEMENTATION: SKSV is available from https://github.com/ydLiu-HIT/SKSV. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.