Abstract. Natural disasters caused by heavy rainfall often cause huge loss of life and property. Hence, the task precipitation nowcasting is great importance. To solve this problem, several deep learning methods have been proposed to forecast future radar echo images, then predicted maps are converted distribution rainfall. The prevailing spatiotemporal sequence prediction apply a ConvRNN structure, which combines convolution recurrent neural network. Although achieve remarkable success,...

Automatic classification of sea ice and open water plays a vital role in climate change research, polar shipping, other applications. Many deep learning-based methods are proposed to automatically classify address this issue. Even though these have achieved remarkable success, the noise phenomenon SAR images still causes considerable limitations model performance. Meanwhile, existing ignore multi-scale global information from large-scale which tends produce misclassification. In paper, we...

As a key technology for maritime applications, trajectory prediction can effectively help ships reduce risks such as collisions and groundings at sea. Currently, although the combination of rich automatic identification system (AIS) data deep learning brings new possibilities ship prediction, is still hugely challenging due to complexity motion. In this paper, we improved model based on TrAISformer. On one hand, sparse multi-dimensional through dictionary coding, map it into probability...

Narrative reasoning relies on the understanding of eventualities in story contexts, which requires a wealth background world knowledge. To help machines leverage such knowledge, existing solutions can be categorized into two groups. Some focus implicitly modeling eventuality knowledge by pretraining language models (LMs) with eventuality-aware objectives. However, this approach breaks down structures and lacks interpretability. Others explicitly collect structured eventuality-centric graphs...

Recent advances in large language models (LLMs) have enhanced their ability to process long input contexts. This development is particularly crucial for tasks that involve retrieving knowledge from an external datastore, which can result inputs. However, recent studies show a positional bias LLMs, demonstrating varying performance depending on the location of useful information within sequence. In this study, we conduct extensive experiments investigate root causes bias. Our findings...

Tianhang Zhang, Lin Qiu, Qipeng Guo, Cheng Deng, Yue Zheng Chenghu Zhou, Xinbing Wang, Luoyi Fu. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Recent works show the effectiveness of cache-based neural coreference resolution models on long documents. These incrementally process a document from left to right and extract relations between mentions entities in cache, resulting much lower memory computation cost compared computing all parallel. However, they do not handle cache misses when high-quality are purged which causes wrong assignments leads prediction errors. We propose new hybrid that integrates two eviction policies capture...

Large language models (LLMs) have achieved remarkable success in NLP and multimodal tasks, among others. Despite these successes, two main challenges remain developing LLMs: (i) high computational cost, (ii) fair objective evaluations. In this paper, we report a solution to significantly reduce LLM training cost through growth strategy. We demonstrate that 101B-parameter with 0.31T tokens can be trained budget of 100K US dollars. Inspired by IQ tests, also consolidate an additional range...

Precipitation nowcasting plays an important role in our life. Many deep learning-based methods are proposed for precipitation by predicting radar echo sequence over the past years, and achieving better performance than traditional approaches. However, all of them based on a static model, which is trained offline learning does not adapt to real-time changing data. Recently, online incremental (OIL) has been dynamically update model continually new data preventing forgetting historical...

Abstract Precipitation forecasting plays an important role in disaster warning, agricultural production, and other fields. To solve this issue, some deep learning methods are proposed to forecast future radar echo images convert them into rainfall distributions. Prevailing spatiotemporal sequence prediction usually based on a ConvRNN structure that combines Convolutional Neural Network Recurrent Network. However, these existing ignore the image change prediction, which causes coherence of...

å›¾åƒè‡ªé€‚åº”æ»¤æ³¢æ˜¯éžçº¿æ€§çš„å›¾åƒå˜æ¢ï¼Œæœ‰å¹¿æ³›çš„åº”ç”¨åœºæ™¯ã€‚ä¼ ç»Ÿçš„å›¾åƒè‡ªé€‚åº”æ»¤æ³¢å™¨å‡æ˜¯ä¸“å®¶è®¾è®¡çš„ï¼Œå¦‚åŒè¾¹æ»¤æ³¢å™¨å’Œå½¢çŠ¶è‡ªé€‚åº”æ»¤æ³¢ç­‰ã€‚CNNä½œä¸ºç‰¹å¾æå–å’Œéžçº¿æ€§èƒ½åŠ›è¡¨è¾¾çš„æœ‰æ•ˆå·¥å ·ï¼Œå¯ç”¨äºŽå­¦ä¹ æž„é€ å›¾åƒè‡ªé€‚åº”æ»¤æ³¢å™¨ã€‚æœ¬æ–‡é¦–å...

Identifying anomalous human spatial trajectory patterns can indicate dynamic changes in mobility behavior with applications domains like infectious disease monitoring and elderly care. Recent advancements large language models (LLMs) have demonstrated their ability to reason a manner akin humans. This presents significant potential for analyzing temporal mobility. In this paper, we conduct empirical studies assess the capabilities of leading LLMs GPT-4 Claude-2 detecting behaviors from data,...

Introduction Clinicians iteratively adjust treatment approaches to improve outcomes but date, automatable for continuous learning of risk factors as these adjustments are made lacking. We combined a large-scale comprehensive real-world Learning Health System infrastructure (LHSI), with automated statistical profiling, visualization, and artificial intelligence (AI) approach test evidence-based discovery clinical three use cases: dysphagia, xerostomia, 3-year survival head neck cancer...

Objectives: When detecting changes in synthetic aperture radar (SAR) images, the quality of difference map has an important impact on detection results, and speckle noise image interferes with extraction change information. In order to improve accuracy SAR map, this paper proposes a method that combines popular deep neural network clustering algorithm.Methods: Firstly, was constructed, FFDNet architecture used retrain image, parameters better effect suppression were obtained. Then log ratio...

Large Language Models (LLMs) have shown impressive capabilities but also a concerning tendency to hallucinate. This paper presents RefChecker, framework that introduces claim-triplets represent claims in LLM responses, aiming detect fine-grained hallucinations. In an extractor generates from response, which are then evaluated by checker against reference. We delineate three task settings: Zero, Noisy and Accurate Context, reflect various real-world use cases. curated benchmark spanning NLP...

With the development of Human-AI Collaboration in Classification (HAI-CC), integrating users and AI predictions becomes challenging due to complex decision-making process. This process has three options: 1) autonomously classifies, 2) learning complement, where collaborates with users, 3) defer, defers users. Despite their interconnected nature, these options have been studied isolation rather than as components a unified system. In this paper, we address weakness novel HAI-CC methodology,...

Parameter-efficient fine-tuning (PEFT) methods typically assume that Large Language Models (LLMs) are trained on data from a single device or client. However, real-world scenarios often require these models private distributed across multiple devices. Federated Learning (FL) offers an appealing solution by preserving user privacy, as sensitive remains local devices during training. Nonetheless, integrating PEFT into FL introduces two main challenges: communication overhead and heterogeneity....

The combination of Oblivious RAM (ORAM) with Trusted Execution Environments (TEE) has found numerous real-world applications due to their complementary nature. TEEs alleviate the performance bottlenecks ORAM, such as network bandwidth and roundtrip latency, ORAM provides general-purpose protection for TEE against attacks exploiting memory access patterns. defining property this combination, which sets it apart from traditional designs, is its ability ensure that accesses, both inside outside...