1 June 2009 in Computers & Mathematics with Applications
2 July 2024 in Complex & Intelligent Systems
Large language models have achieved outstanding performance on various downstream tasks with their advanced understanding of natural language and zero-shot capability. However, they struggle with knowledge constraints, particularly in tasks requiring complex reasoning or extended logical sequences. These limitations can affect their performance in question answering by leading to inaccuracies and hallucinations. This paper proposes a novel framework called KnowledgeNavigator that leverages large language models on knowledge graphs to achieve accurate and interpretable multi-hop reasoning. Especially with an analysis-retrieval-reasoning process, KnowledgeNavigator searches the optimal path iteratively to retrieve external knowledge and guide the reasoning to reliable answers. KnowledgeNavigator treats knowledge graphs and large language models as flexible components that can be switched between different tasks without additional costs. Experiments on three benchmarks demonstrate that KnowledgeNavigator significantly improves the performance of large language models in question answering and outperforms all large language models-based baselines.
10 January 2018 in IEEE Access
Smartphone-based pedestrian dead-reckoning (PDR) has become promising in indoor localization since it locates users with a smartphone only. However, existing PDR approaches are still facing the problem of accumulated localization errors due to the low-cost noisy sensors and complicated human movements. This paper presents a novel PDR indoor localization algorithm combined with online sequential extreme learning machine (OS-ELM). By analyzing the process of PDR localization, we first formulate the process of PDR localization as an approximation function. And then, we design a sliding-window based scheme for preprocessing the obtained inertial sensor data and thus to generate the feature dataset. At last, we propose OS-ELM based PDR algorithm to address pedestrian’s localization problem. Due to fact of universal approximation capability and extreme learning speed within OS-ELM, our algorithm can adapt to localization environment dynamically and reduce the localization errors to a lower scale. Additionally, by taking the pedestrian’s movement habits into the process of extreme learning, our algorithm can predict the pedestrian’s position regardless of holding postures. To evaluate the performance of the proposed algorithm, we implement OS-ELM based PDR on a real android-based smartphone and compare it with the state-ofthe- art approaches. Extensive experiment results demonstrate the effectiveness of the proposed algorithm in various different postures and the practicability in indoor localization.
29 June 2017 in IEEE Access
In recent years, with the development of cloud computing technology, the size of a data center is expanding rapidly. To minimize the energy consumption of a data center, we propose an energy-efficient virtual resource dynamic integration (VRDI) method. In the proposed VRDI method, first, by monitoring the load patterns of the physical machines (PMs) and the corresponding thresholds of PMs calculated using the statistical data, we propose a PM selection algorithm to find a set of PMs, which should be integrated. Furthermore, we propose a virtual machine (VM) selection algorithm based on minimum migration policy to select the VMs that are deployed on the integrated PMs. Finally, to solve the target VM placement, we propose a VM placement algorithm based on an improved genetic algorithm. Using the encoding, crossover and mutation operations of the genetic algorithm, we obtain an effective solution for the VM placement problem. The experiments show that the proposed VRDI method can reduce the energy consumption of data center and ensure the quality of service of the cloud applications developed on the VMs.
2 April 2025 in Complex & Intelligent Systems
With the widespread application of large language models (LLMs) in natural language processing (NLP), hallucinations have become a significant impediment to their effective use of LLMs in industry applications. To address this challenge, we integrate existing hallucination detection and mitigation methods into a unified hallucination detection and mitigation framework. The framework consists of four main components: output parser, reference parser, fact verifier, and mitigator. These components collectively consolidate various hallucination detection and mitigation methods. Within this unified framework, we introduce the hierarchical semantic piece (HSP) for hallucination detection and mitigation. The HSP method extracts multi-granularity semantic pieces from both the reference material and the generated text. Sentence-level semantic pieces encapsulate global semantic information, while entity-level semantic pieces handle local semantic information. This method verifies the consistency between the generated text and the reference text at corresponding granularities, thereby enhancing the effectiveness of hallucination detection and mitigation. Experimental results show that the HSP method is very effective in detecting and mitigating hallucinations and shows lower computational resource consumption. Our method has great potential and promises for industry applications that rely on professionalism and reliability.