arXiv subjects

Yuyu Fan

Publications and source records attributed to Yuyu Fan.

5 recordsLinked to original sources

arXiv2026

Subliminal Learning as Trait-Direction Drift: A Mechanism and Targeted Control under SFT Distillation

Beyond intended capabilities, model distillation can transfer hidden traits from a teacher. A teacher biased by a system prompt can generate semantically clean training data, such as numeric sequences, that still causes a downstream student to inherit the hidden preference, a phenomenon known as subliminal learning. Prior work has identified several parts of this process. How the signal builds up during training and produces behavioral transfer remains unclear, making targeted mitigation difficult. We propose and validate trait-direction drift as a mechanism for subliminal learning: biased generation creates measurable preference gaps in teacher data, and student-recognizable gaps induce trait-aligned updates during supervised fine-tuning that accumulate into behavioral transfer. Guided by this mechanism, we propose probe-space corridor regularization, a targeted defense that constrains drift along a calibrated trait direction during distillation. The method substantially reduces hidden-trait transfer, preserving task performance: for example, it lowers malicious-response transfer from 29.55% to 6.45% with low main-task accuracy cost, and consistently suppresses animal-preference transfer across the main Qwen setting. The preference-gap, training-trajectory, and intervention evidence links subliminal learning to trait-direction drift and motivates corridor regularization as a targeted control during distillation.

cs.LG↗

arXiv2026

Scaling Model-Generated Distillation Data Can Make Latent Teacher Traits More Recoverable

Scaling model-generated data is usually viewed as improving distillation: more examples should increase coverage, reduce noise, and produce stronger students. We show a second effect: larger datasets can make subtle teacher-specific signals easier to detect in the trained student, even when examples are off-task and never mention the trait. In a controlled setup inspired by subliminal learning, a teacher induced to express a target trait generates restricted off-task data, such as number-only completions. Students trained on different amounts of independent off-task data are evaluated in a separate domain, with matched no-trait controls isolating target-specific transfer. Our main finding is that larger independent datasets make the teacher's induced trait stand out more clearly in the student's later behavior. Other plausible traits may also strengthen with scale, but the target usually grows more. When the small-scale student already favors the target, scaling mainly amplifies that behavior; when it favors a related or salient alternative, more data can shift behavior toward the intended trait. Analyses of learned LoRA updates show a parallel trend. These effects appear across model families, trait types, multi-trait settings, and cross-model transfer. Our results suggest that scaling generated distillation data should be paired with trait-aware curation and evaluation, even when the data appears off-task or benign.

cs.LG↗

arXiv2026

TRIMS: Trajectory-Ranked Instruction Masked Supervision for Diffusion Language Models

Diffusion language models (DLMs) offer a promising path toward low-latency generation through parallel decoding, but their practical efficiency depends heavily on the decoding trajectory. In practice, this advantage often fails to fully materialize because standard training does not provide explicit supervision over token reveal order, creating a train-inference mismatch that leads to suboptimal decoding behavior. We propose Trajectory-Ranked Instruction Masked Supervision (TRIMS), a simple trajectory-guided supervised fine-tuning framework that injects trajectory supervision into standard Masked Diffusion Language Model (MDLM) training with minimal overhead. Instead of relying on costly DLM-based distillation, TRIMS uses lightweight signals from an autoregressive teacher to guide a trajectory-aware masking strategy, encouraging the model to learn more effective decoding orders. Experiments on LLaDA and Dream across math and coding benchmarks show that TRIMS significantly improves the accuracy-parallelism trade-off over both standard MDLM training and train-free acceleration baselines, while achieving competitive performance with prior distillation-based approaches at substantially lower training cost. Further analysis shows that TRIMS leads to better decoding trajectories, validating the effectiveness of trajectory-guided supervision for DLMs.

cs.CL↗

arXiv2025

SURE-Med: Systematic Uncertainty Reduction for Enhanced Reliability in Medical Report Generation

Automated medical report generation (MRG) holds great promise for reducing the heavy workload of radiologists. However, its clinical deployment is hindered by three major sources of uncertainty. First, visual uncertainty, caused by noisy or incorrect view annotations, compromises feature extraction. Second, label distribution uncertainty, stemming from long-tailed disease prevalence, biases models against rare but clinically critical conditions. Third, contextual uncertainty, introduced by unverified historical reports, often leads to factual hallucinations. These challenges collectively limit the reliability and clinical trustworthiness of MRG systems. To address these issues, we propose SURE-Med, a unified framework that systematically reduces uncertainty across three critical dimensions: visual, distributional, and contextual. To mitigate visual uncertainty, a Frontal-Aware View Repair Resampling module corrects view annotation errors and adaptively selects informative features from supplementary views. To tackle label distribution uncertainty, we introduce a Token Sensitive Learning objective that enhances the modeling of critical diagnostic sentences while reweighting underrepresented diagnostic terms, thereby improving sensitivity to infrequent conditions. To reduce contextual uncertainty, our Contextual Evidence Filter validates and selectively incorporates prior information that aligns with the current image, effectively suppressing hallucinations. Extensive experiments on the MIMIC-CXR and IU-Xray benchmarks demonstrate that SURE-Med achieves state-of-the-art performance. By holistically reducing uncertainty across multiple input modalities, SURE-Med sets a new benchmark for reliability in medical report generation and offers a robust step toward trustworthy clinical decision support.

cs.AI↗

arXiv2025

SafeWork-R1: Coevolving Safety and Intelligence under the AI-45$^{\circ}$ Law

We introduce SafeWork-R1, a cutting-edge multimodal reasoning model that demonstrates the coevolution of capabilities and safety. It is developed by our proposed SafeLadder framework, which incorporates large-scale, progressive, safety-oriented reinforcement learning post-training, supported by a suite of multi-principled verifiers. Unlike previous alignment methods such as RLHF that simply learn human preferences, SafeLadder enables SafeWork-R1 to develop intrinsic safety reasoning and self-reflection abilities, giving rise to safety `aha' moments. Notably, SafeWork-R1 achieves an average improvement of $46.54\%$ over its base model Qwen2.5-VL-72B on safety-related benchmarks without compromising general capabilities, and delivers state-of-the-art safety performance compared to leading proprietary models such as GPT-4.1 and Claude Opus 4. To further bolster its reliability, we implement two distinct inference-time intervention methods and a deliberative search mechanism, enforcing step-level verification. Finally, we further develop SafeWork-R1-InternVL3-78B, SafeWork-R1-DeepSeek-70B, and SafeWork-R1-Qwen2.5VL-7B. All resulting models demonstrate that safety and capability can co-evolve synergistically, highlighting the generalizability of our framework in building robust, reliable, and trustworthy general-purpose AI.

cs.AI↗