Academic

GraRe: Grasp Candidate Re-Ranking for Frozen 6-DoF Grasp Detectors

Existing 6-DoF grasp detectors often rank candidates by detector confidence, which can be poorly aligned with grasp quality. GraRe re-ranks candidates from frozen detectors using candidate attributes, shell-stratified local geometry, and object context, fused by a Transformer. Experiments on GraspNet-1Billion with three frozen detectors improve Average AP by up to 13.60 points, while real-robot tests demonstrate robust grasping in cluttered scenes.

Construction and Study of an Explainable Logical Reasoning Dataset

Logical reasoning is crucial for machine understanding, yet existing benchmarks lack detailed explanations of reasoning processes. This work introduces the Chinese-English Explainable Logical Reasoning (Ex-LoR) dataset, containing 3,411 problems with explanations across six reasoning categories. It defines question-answering and explanation-generation tasks, and experiments with several language models show that current systems still struggle to answer reliably and produce reasonable explanations.

BERT-Enhanced Retrieval Tool for Homework Plagiarism Detection System

This paper presents a GPT-3.5-based method for generating 32,927 paired plagiarism-detection examples covering diverse plagiarism strategies. A Faiss-and-BERT retrieval system identifies plagiarism efficiently and accurately, achieving 98.86% accuracy, 98.90% precision, 98.86% recall, and 0.9888 F1 in experiments. A user-friendly demo platform supports uploading a text library for interactive analysis.