Lately, the practice of utilizing task-specific fine-tuning has been implemented to improve the performance of large language models (LLM) in subsequent tasks. Through the integration of diverse LLMs, the overall competency of LLMs is significantly boosted. Nevertheless, traditional ensemble methods are notably memory-intensive, necessitating the simultaneous loading of all specialized models into GPU memory. To address the inefficiency, model merging strategies have emerged, merging all LLMs into one model to reduce the memory footprint during inference. Despite these advances, model merging often leads to parameter conflicts and performance decline as the number of experts increases. Previous methods to mitigate these conflicts include post-pruning and partial merging. However, both approaches have limitations, particularly in terms of performance and storage efficiency when merged experts increase. To address these challenges, we introduce Channel Merging, a novel strategy designed to minimize parameter conflicts while enhancing storage efficiency. This method clusters and merges channel parameters based on their similarity to form several groups offline. By ensuring that only highly similar parameters are merged within each group, it significantly reduces parameter conflicts. During inference, we can instantly look up the expert parameters from the merged groups, preserving specialized knowledge. Our experiments demonstrate that Channel Merging consistently delivers high performance, matching unmerged models in tasks like English and Chinese reasoning, mathematical reasoning, and code generation. Moreover, it obtains results comparable to model ensemble with just 53% parameters when used with a task-specific router.

The emergence of 3D Gaussian Splatting (3DGS) has recently sparked a renewed wave of dense visual SLAM research. However, current methods face challenges such as sensitivity to artifacts and noise, sub-optimal selection of training viewpoints, and a lack of light global optimization. In this paper, we propose a dense SLAM system that tightly couples 3DGS with ORB features. We design a joint optimization approach for robust tracking and effectively reducing the impact of noise and artifacts. This involves combining novel geometric observations, derived from accumulated transmittance, with ORB features extracted from pixel data. Furthermore, to improve mapping quality, we propose an adaptive Gaussian expansion and regularization method that enables Gaussian primitives to represent the scene compactly. This is coupled with a viewpoint selection strategy based on the hybrid graph to mitigate over-fitting effects and enhance convergence quality. Finally, our approach achieves compact and high-quality scene representations and accurate localization. GSORB-SLAM has been evaluated on different datasets, demonstrating outstanding performance. The code will be available.

Large pre-trained models (LPMs), such as LLaMA and GLM, have shown exceptional performance across various tasks through fine-tuning. Although low-rank adaption (LoRA) has emerged to cheaply fine-tune these LPMs on downstream tasks, their deployment is still hindered by the vast model scale and computational costs. Neural network pruning offers a way to compress LPMs. However, the current pruning methods designed for LPMs are not compatible with LoRA. This is due to their utilization of unstructured pruning on LPMs, impeding the merging of LoRA weights, or their dependence on the gradients of pre-trained weights to guide pruning, which can impose significant memory overhead. To this end, we propose LoRAPrune, a new framework that delivers an accurate, compact model for efficient inference in a highly memory-effective manner. Specifically, we first design a LoRA-guided pruning criterion, which uses the weights and gradients of LoRA, rather than the gradients of pre-trained weights for importance estimation. We then propose a structured iterative pruning procedure, to remove redundant channels and heads. Extensive experimental results demonstrate the superior performance of our LoRAPrune over existing approaches on the LLaMA series models. For instance, at a 50% compression rate, LoRAPrune outperforms LLM-Pruner by a perplexity reduction of 8.0 on WikiText2 and 16.05 on PTB datasets, while concurrently reducing memory usage by 52.6%. The code will be released after review.

One-shot Neural architecture search (One-shot NAS) has been proposed as a time-efficient approach to obtain optimal subnet architectures and weights under different complexity cases by training only once. However, the subnet performance obtained by weight sharing is often inferior to the performance achieved by retraining. In this paper, we investigate the performance gap and attribute it to the use of uniform sampling, which is a common approach in supernet training. Uniform sampling concentrates training resources on subnets with intermediate computational resources, which are sampled with high probability. However, subnets with different complexity regions require different optimal training strategies for optimal performance. To address the problem of uniform sampling, we propose ShiftNAS, a method that can adjust the sampling probability based on the complexity of subnets. We achieve this by evaluating the performance variation of subnets with different complexity and designing an architecture generator that can accurately and efficiently provide subnets with the desired complexity. Both the sampling probability and the architecture generator can be trained end-to-end in a gradient-based manner. With ShiftNAS, we can directly obtain the optimal model architecture and parameters for a given computational complexity. We evaluate our approach on multiple visual network models, including convolutional neural networks (CNNs) and vision transformers (ViTs), and demonstrate that ShiftNAS is model-agnostic. Experimental results on ImageNet show that ShiftNAS can improve the performance of one-shot NAS without additional consumption. Source codes are available at https://github.com/bestfleer/ShiftNAS.

Robot navigation in dynamic environments shared with humans is an important but challenging task, which suffers from performance deterioration as the crowd grows. In this paper, multi-subgoal robot navigation approach based on deep reinforcement learning is proposed, which can reason about more comprehensive relationships among all agents (robot and humans). Specifically, the next position point is planned for the robot by introducing history information and interactions in our work. Firstly, based on subgraph network, the history information of all agents is aggregated before encoding interactions through a graph neural network, so as to improve the ability of the robot to anticipate the future scenarios implicitly. Further consideration, in order to reduce the probability of unreliable next position points, the selection module is designed after policy network in the reinforcement learning framework. In addition, the next position point generated from the selection module satisfied the task requirements better than that obtained directly from the policy network. The experiments demonstrate that our approach outperforms state-of-the-art approaches in terms of both success rate and collision rate, especially in crowded human environments.

Rotated object detection is a challenging task in aerial images as the object in aerial images are displayed in arbitrary directions and usually densely packed. Although considerable progress has been made, there are still challenges that existing regression-based rotation detectors suffer the problem of discontinuous boundaries, which is directly caused by angular periodicity or corner ordering. In this paper, we propose a simple effective framework to address the above challenges. Instead of directly regressing the five parameters (coordinates of the central point, width, height, and rotation angle) or the four vertices, we use the area ratio of parallelogram (ARP) to accurately describe a multi-oriented object. Specifically, we regress coordinates of center point, height and width of minimum circumscribed rectangle of oriented object and three area ratios {\lambda}_1, {\lambda}_2 and {\lambda}_3. This may facilitate the offset learning and avoid the issue of angular periodicity or label points sequence for oriented objects. To further remedy the confusion issue nearly horizontal objects, we employ the area ratio between the object and its horizontal bounding box (minimum circumscribed rectangle) to guide the selection of horizontal or oriented detection for each object. We also propose a rotation efficient IoU loss (R-EIoU) to connect the horizontal bounding box with the three area ratios and improve the accurate for the rotating bounding box. Experimental results on three remote sensing datasets including HRSC2016, DOTA and UCAS-AOD and scene text including ICDAR2015 show that our method achieves superior detection performance compared with many state-of-the-art approaches. The code and model will be coming with paper published.