The Hong Kong Polytechnic University · 16 records

dateidtitleauthorsvenue↑
2026-11-032611.02928Lab2Sim: Turning Real-World Experiment Videos into Interactive Virtual ExperimentsXusen Hei et al.arXiv4
2026-11-022610.35850OSWAM: An Effective World Action Model with One-Step Action GenerationGuowen Zhang et al.arXiv7
2026-09-302609.32488When Does Dense Retrieval Need Asymmetric Geometry? A Bias-Variance Theory of Shared and Dual ProjectionsMaojun Sun et al.arXiv18
2026-09-302609.34392Org-Agent: Beyond Personal Assistants Towards Organizational AgentsLuyao Zhuang et al.arXiv3
2026-09-292609.33437SMAT: Simple and Efficient Merge-Aware TrainingYanggan Gu et al.arXiv5
2026-09-292609.32821Routing Drift Alone Does Not Diagnose Failure in Merged MoE LLMsYuanyi Wang et al.arXiv5
2026-09-092609.07414RelightFormer: Feed-forward Generative Transformer for Multiview Object RelightingHejun Wang et al.arXiv2
2026-08-132608.10450Persistent Recursive Worlds Enable Autonomous Software EvolutionBeichen Huang et al.arXiv8
2026-06-182606.18322SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed BehaviorMingyue Cui et al.arXiv15
2026-06-162606.16519BadWorld: Adversarial Attacks on World ModelsLinghui Shen et al.arXiv16
2026-06-042605.29489Access Sets Matter: Budgeting Expert Reads for Scalable Weight-Space Model MergingYuanyi Wang et al.arXiv3
2026-06-012605.26844Not All Disagreement Is Learnable: Token Teachability in On-Policy DistillationYuanyi Wang et al.arXiv22
2026-05-192605.16882E-PMQ: Expert-Guided Post-Merge Quantization with Merged-Weight AnchoringWenjun Wang et al.arXiv1
2026-05-142605.13030FeatCal: Feature Calibration for Post-Merging ModelsYanggan Gu et al.arXiv4
2026-05-122605.09608Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-TrainingYuanyi Wang et al.arXiv51
2026-05-122509.24244Model Merging Scaling Laws in Large Language ModelsYuanyi Wang et al.arXiv42