arXiv subjects

Ziyi Pan

Publications and source records attributed to Ziyi Pan.

4 recordsLinked to original sources

arXiv2026

Knowledge Injection Exists in MoE? Exploring Expert-Aware Contrast Decoding in MoE for Mitigating LLMs'Hallucinations

Existing LLM hallucination mitigation methods, including prompt engineering and model optimization, either hardly alter models'internal knowledge or have poor cross-domain generalization. Contrastive decoding mitigates hallucinations by using layer-wise differences in LLMs. However, prior studies only explore transformer-based models (e.g., GPT), ignoring other effective frameworks like mixture-of-experts (MoE) models. Since MoE alters the traditional transformer architecture, we conduct empirical studies to investigate whether similar layer-wise differences exist in MoEs. Our results show that they do not exist in MoE with shared experts; nevertheless, across different MoEs, higher layers exhibit distinct expert activation patterns between factual and non-factual outputs. Building on these, we propose EAACD, an expert-aware adaptive contrast decoding that uses expert differences in MoE's higher layers to mitigate hallucinations on QA tasks. EAACD splits high-layer experts into a higher-reliability group and several lower-reliability groups based on their confidence and consistency. It contrasts the higher-reliability group's prediction with each lower-reliability group's prediction to calibrate the model's original predictions. To strengthen this contrast, EAACD amplifies hallucinations from lower-reliability experts via attention and masking to provide stronger negative references. EAACD outperforms all baselines on four datasets.

cs.CL↗

arXiv2026

SongSong: A Time Phonograph for Chinese SongCi Music from Thousand of Years Away

Recently, there have been significant advancements in music generation. However, existing models primarily focus on creating modern pop songs, making it challenging to produce ancient music with distinct rhythms and styles, such as ancient Chinese SongCi. In this paper, we introduce SongSong, the first music generation model capable of restoring Chinese SongCi to our knowledge. Our model first predicts the melody from the input SongCi, then separately generates the singing voice and accompaniment based on that melody, and finally combines all elements to create the final piece of music. Additionally, to address the lack of ancient music datasets, we create OpenSongSong, a comprehensive dataset of ancient Chinese SongCi music, featuring 29.9 hours of compositions by various renowned SongCi music masters. To assess SongSong's proficiency in performing SongCi, we randomly select 85 SongCi sentences that were not part of the training set for evaluation against SongSong and music generation platforms such as Suno and SkyMusic. The subjective and objective outcomes indicate that our proposed model achieves leading performance in generating high-quality SongCi music.

cs.SD↗

arXiv2024

Artifact Correction in Magnetic Resonance Temperature Imaging for Laser Interstitial Thermotherapy with Multi-echo Acquisitions

In MRI-guided laser interstitial thermotherapy (MRgLITT), a signal void sometimes appears at the heating center of the measured temperature map. In neurosurgical MRgLITT treatments, cerebrospinal fluid pulsation (CSF), which may lead to temperature artifacts, also needs to be carefully managed. We find that signal loss in MR magnitude images can be one distinct contributor to the temperature imaging signal void. Therefore, this study aims to investigate this finding and more importantly. Also, this study intends to improve measurement accuracy by correcting CSF-induced temperature errors and employing a more reliable phase unwrapping algorithm. A gradient echo sequence with certain TE values for temperature imaging is used to quantify T2* variations during MRgLITT and to investigate the development of signal voids throughout the treatment. Informed by these findings, a multi-echo GRE sequence with appropriate TE coverage is employed. A multi-echo-based correction algorithm is developed to address the signal loss-induced temperature errors. A new phase unwrapping method and a new CSF pulsation correction approach are developed for multi-echo signal processing. The temperature imaging method is evaluated by gel phantom, ex-vivo, and in-vivo LITT heating experiments. T2* shortening during heating can be one important cause of the temperate imaging signal voids and this demands the multi-echo acquisition with varied TE values. The proposed multi-echo-based method can effectively correct signal loss-induced temperature errors and raise temperature estimation precision. The multi-echo thermometry in the in-vivo experiments shows smoother hotspot boundaries, fewer artifacts, and improved thermometry reliability. In the in-vivo experiments, the ablation areas estimated from the multi-echo thermometry also show satisfactory agreement with those determined from post-ablation MR imaging.

physics.med-ph↗

arXiv2024

Zero-resource Hallucination Detection for Text Generation via Graph-based Contextual Knowledge Triples Modeling

LLMs obtain remarkable performance but suffer from hallucinations. Most research on detecting hallucination focuses on the questions with short and concrete correct answers that are easy to check the faithfulness. Hallucination detections for text generation with open-ended answers are more challenging. Some researchers use external knowledge to detect hallucinations in generated texts, but external resources for specific scenarios are hard to access. Recent studies on detecting hallucinations in long text without external resources conduct consistency comparison among multiple sampled outputs. To handle long texts, researchers split long texts into multiple facts and individually compare the consistency of each pairs of facts. However, these methods (1) hardly achieve alignment among multiple facts; (2) overlook dependencies between multiple contextual facts. In this paper, we propose a graph-based context-aware (GCA) hallucination detection for text generations, which aligns knowledge facts and considers the dependencies between contextual knowledge triples in consistency comparison. Particularly, to align multiple facts, we conduct a triple-oriented response segmentation to extract multiple knowledge triples. To model dependencies among contextual knowledge triple (facts), we construct contextual triple into a graph and enhance triples' interactions via message passing and aggregating via RGCN. To avoid the omission of knowledge triples in long text, we conduct a LLM-based reverse verification via reconstructing the knowledge triples. Experiments show that our model enhances hallucination detection and excels all baselines.

cs.CL↗