arXiv subjects

Hung-Chun Hsu

Publications and source records attributed to Hung-Chun Hsu.

5 recordsLinked to original sources

arXiv2026

LODESTAR: Robust Entropy-Based Answer Selection in Retrieval-Augmented Generation for Question Answering -- Directing Frozen-LLM Entropy with a Reinforcement-Learned Prompt Polarizer under Misleading Passages

Predictive-distribution entropy is a strong answer-selection rule in retrieval-augmented generation (RAG) for question answering: across five QA benchmarks, selecting the answer a frozen respondent LLM produces with the lowest answer-token entropy lifts mean $F_1$ from 0.4769 to 0.5148 over the retriever's top-ranked passage, without gold answers. Yet this rule, which prior entropy-based selectors adopt, fails: a misleading passage makes the respondent confidently wrong, driving entropy down where the uncertainty signal looks most trustworthy. The failure comes from the passage the respondent reads, and the context it is read in is an input we can intervene on. We introduce LODESTAR: to our knowledge the first method to score a text intervention by the uncertainty it induces in a third-party frozen respondent, compared within one question. LODESTAR uses reinforcement learning (GRPO) to train, once and offline, a polarizer -- a short fixed natural-language string inserted into the respondent's prompt and never into its weights, directing entropy so that entropy-based answer selection stays robust to misleading passages; training labels are built from gold answers and two LLM judges, and inference reads neither. With every competing selector under the same frozen respondent and candidate pools on 5,008 questions, LODESTAR attains the highest mean $F_1$ of any inference-ready selector (0.5339), the highest macro exact match (0.4136), and the highest GPT-4o judge score of the frozen-respondent configurations judged (0.6435); its three-seed mean wins all 70 $F_1$ cells against fourteen published configurations and is paired-significant on $F_1$ against every one. The gain holds in-domain on NQ-Open and out-of-domain over SQuAD, TriviaQA, EntityQuestions and WebQuestions. Ablating the polarizer shows it is what makes the respondent read a misleading passage less often (26.0% vs 30.3%).

cs.CL↗

arXiv2025

CFDA & CLIP at TREC iKAT 2025: Enhancing Personalized Conversational Search via Query Reformulation and Rank Fusion

The 2025 TREC Interactive Knowledge Assistance Track (iKAT) featured both interactive and offline submission tasks. The former requires systems to operate under real-time constraints, making robustness and efficiency as important as accuracy, while the latter enables controlled evaluation of passage ranking and response generation with pre-defined datasets. To address this, we explored query rewriting and retrieval fusion as core strategies. We built our pipelines around Best-of-$N$ selection and Reciprocal Rank Fusion (RRF) strategies to handle different submission tasks. Results show that reranking and fusion improve robustness while revealing trade-offs between effectiveness and efficiency across both tasks.

cs.IR↗

arXiv2025

Test-Time Scaling Strategies for Generative Retrieval in Multimodal Conversational Recommendations

The rapid evolution of e-commerce has exposed the limitations of traditional product retrieval systems in managing complex, multi-turn user interactions. Recent advances in multimodal generative retrieval -- particularly those leveraging multimodal large language models (MLLMs) as retrievers -- have shown promise. However, most existing methods are tailored to single-turn scenarios and struggle to model the evolving intent and iterative nature of multi-turn dialogues when applied naively. Concurrently, test-time scaling has emerged as a powerful paradigm for improving large language model (LLM) performance through iterative inference-time refinement. Yet, its effectiveness typically relies on two conditions: (1) a well-defined problem space (e.g., mathematical reasoning), and (2) the model's ability to self-correct -- conditions that are rarely met in conversational product search. In this setting, user queries are often ambiguous and evolving, and MLLMs alone have difficulty grounding responses in a fixed product corpus. Motivated by these challenges, we propose a novel framework that introduces test-time scaling into conversational multimodal product retrieval. Our approach builds on a generative retriever, further augmented with a test-time reranking (TTR) mechanism that improves retrieval accuracy and better aligns results with evolving user intent throughout the dialogue. Experiments across multiple benchmarks show consistent improvements, with average gains of 14.5 points in MRR and 10.6 points in nDCG@1.

cs.IR↗

arXiv2024

A GAN Approach for Node Embedding in Heterogeneous Graphs Using Subgraph Sampling

Graph neural networks (GNNs) face significant challenges with class imbalance, leading to biased inference results. To address this issue in heterogeneous graphs, we propose a novel framework that combines Graph Neural Network (GNN) and Generative Adversarial Network (GAN) to enhance classification for underrepresented node classes. The framework incorporates an advanced edge generation and selection module, enabling the simultaneous creation of synthetic nodes and edges through adversarial learning. Unlike previous methods, which predominantly focus on homogeneous graphs due to the difficulty of representing heterogeneous graph structures in matrix form, this approach is specifically designed for heterogeneous data. Existing solutions often rely on pre-trained models to incorporate synthetic nodes, which can lead to optimization inconsistencies and mismatches in data representation. Our framework avoids these pitfalls by generating data that aligns closely with the inherent graph topology and attributes, ensuring a more cohesive integration. Evaluations on multiple real-world datasets demonstrate the method's superiority over baseline models, particularly in tasks focused on identifying minority node classes, with notable improvements in performance metrics such as F-score and AUC-PRC score. These findings highlight the potential of this approach for addressing critical challenges in the field.

cs.LG↗

arXiv2020

NINEPINS: Nuclei Instance Segmentation with Point Annotations

Deep learning-based methods are gaining traction in digital pathology, with an increasing number of publications and challenges that aim at easing the work of systematically and exhaustively analyzing tissue slides. These methods often achieve very high accuracies, at the cost of requiring large annotated datasets to train. This requirement is especially difficult to fulfill in the medical field, where expert knowledge is essential. In this paper we focus on nuclei segmentation, which generally requires experienced pathologists to annotate the nuclear areas in gigapixel histological images. We propose an algorithm for instance segmentation that uses pseudo-label segmentations generated automatically from point annotations, as a method to reduce the burden for pathologists. With the generated segmentation masks, the proposed method trains a modified version of HoVer-Net model to achieve instance segmentation. Experimental results show that the proposed method is robust to inaccuracies in point annotations and comparison with Hover-Net trained with fully annotated instance masks shows that a degradation in segmentation performance does not always imply a degradation in higher order tasks such as tissue classification.

eess.IV↗