Skip to main content
Search

Soochow University

4 related articles

Subscribe to updates
  1. arXiv

    Expert shape breaks free from routing granularity: near-square experts plus layerwise capacity allocation top nine downstream tasks at 8B

    Latent mixture-of-experts decouples expert input dimension from residual width, making expert shape a tunable design choice; at 3.4B and 8B scales, near-square experts with capacity allocated by layerwise intrinsic dimension reach lower validation loss than matched uniform configurations.

  2. arXiv

    Poisoned fine-tuning can be repaired at two layers: harmful score falls from 11.54 to 0.08

    In fine-tuning-as-a-service, malicious fine-tuning can weaken a model's refusal behavior while keeping its downstream task performance. SLDR replaces layer-wise safety diagnostics with a signed sensitivity spectrum, trains a LoRA recovery adapter only on the two layers with the highest and lowest sensitivity, and uses a representation-based maliciousness score at inference to decide when to activate it. On Llama3.1/SST2, the average harmful score drops from 11.54 to 0.08 while downstream accuracy matches undefended fine-tuning, and the harmful score stays near zero up to a poisoning ratio of 0.9.

  3. arXiv

    HDPO has LLMs list candidate solution outlines before selecting one to reason through, reportedly improving both reasoning and candidate-solution diversity

    The work proposes Hint-Guided Diversified Policy Optimization (HDPO), which lets a model first list all potential candidate solution outlines as hints and then select the most reliable one for further reasoning; the method has two stages, Cold Start for Structured Reasoning and Hint-Guided Diversified Reinforcement Learning, to incentivize diverse and reliable solutions along a "propose-select-think" trajectory, and experimental results are reported to show that HDPO effectively boosts LLM reasoning and enhances the diversity of candidate solutions as well as the model's ability to identify reliable solutions.

  4. Network meta-analysis of 10 phase III trials and over 8,000 advanced gastric cancer patients ranks trastuzumab and tislelizumab highest for delaying quality-of-life deterioration

    This network meta-analysis pooled 10 phase III randomized controlled trials with over 8,000 patients with unresectable advanced gastric cancer, ranked first-line immune checkpoint inhibitors and targeted agents by SUCRA on time to deterioration across EORTC QLQ-C30, EORTC QLQ-STO22 and EQ-5D domains, and used an exploratory minimum distance criterion to integrate overall survival with health-related quality of life, finding that trastuzumab (HER2-positive population) and tislelizumab (largely biomarker-unselected population) ranked most favorably across most quality-of-life domains, while the composite metric remains exploratory, lacks uncertainty estimates, and should not serve as primary evidence for clinical decision-making.

You’re all caught up