Retrieval Meets Reasoning: Dynamic In-Context Editing for Long-Text Understanding

Current Large Language Models (LLMs) face inherent limitations due to their pre-defined context lengths, which impede their capacity for multi-hop reasoning within extensive textual contexts. While existing techniques like Retrieval-Augmented Generation (RAG) have attempted to bridge this gap by sourcing external information, they fall short when direct answers are not readily available. We introduce a novel approach that re-imagines information retrieval through dynamic in-context editing, inspired by recent breakthroughs in knowledge editing. By treating lengthy contexts as malleable external knowledge, our method interactively gathers and integrates relevant information, thereby enabling LLMs to perform sophisticated reasoning steps. Experimental results demonstrate that our method effectively empowers context-limited LLMs, such as Llama2, to engage in multi-hop reasoning with improved performance, which outperforms state-of-the-art context window extrapolation methods and even compares favorably to more advanced commercial long-context models. Our interactive method not only enhances reasoning capabilities but also mitigates the associated training and computational costs, making it a pragmatic solution for enhancing LLMs' reasoning within expansive contexts.

Multi-Level Feature Aggregation and Recursive Alignment Network for Real-Time Semantic Segmentation

Real-time semantic segmentation is a crucial research for real-world applications. However, many methods lay particular emphasis on reducing the computational complexity and model size, while largely sacrificing the accuracy. In some scenarios, such as autonomous navigation and driver assistance system, accuracy and speed are equally important. To tackle this problem, we propose a novel Multi-level Feature Aggregation and Recursive Alignment Network (MFARANet), aiming to achieve high segmentation accuracy at real-time inference speed. We employ ResNet-18 as the backbone to ensure efficiency, and propose three core components to compensate for the reduced model capacity due to the shallow backbone. Specifically, we first design Multi-level Feature Aggregation Module (MFAM) to aggregate the hierarchical features in the encoder to each scale to benefit subsequent spatial alignment and multi-scale inference. Then, we build Recursive Alignment Module (RAM) by combining the flow-based alignment module with recursive upsampling architecture for accurate and efficient spatial alignment between multi-scale score maps. Finally, the Adaptive Scores Fusion Module (ASFM) is proposed to adaptively fuse multi-scale scores so that the final prediction can favor objects of multiple scales. Comprehensive experiments on three benchmark datasets including Cityscapes, CamVid and PASCAL-Context show the effectiveness and efficiency of our method. In particular, we achieve a better balance between speed and accuracy than state-of-the-art real-time methods on Cityscapes and CamVid datasets. Code is available at: https://github.com/Yanhua-Zhang/MFARANet.