Skip to main content
Search

Adelaide University

7 related articles

Subscribe to updates
  1. arXiv

    History Forcing teaches few-step video models to remember: a 1.3B student reaches 79.82 on 60-second VBench-Long

    Few-step autoregressive video diffusion models gradually lose subject and scene information over long rollouts, and existing distillation targets do not constrain distant history separately. History Forcing builds a context residual from the difference between one autoregressive teacher's predictions with and without extra history, then weights it into the DMD or OPSD distillation target. A 1.3B student reaches a VBench quality score of 85.07 for 15-second generation and a VBench-Long aggregate of 79.82 for 60-second generation, and raises PlayWorld Evolution Quality from 2.35 to 2.70 when the camera leaves and returns.

  2. arXiv

    SP-DocReader cuts Qwen3-VL-4B Vary-600K character error rate from 2.40 to 1.11 with difference-aware self-play

    SP-DocReader is a self-play framework for document OCR that aligns reference and generated readings with a longest common subsequence, applies contrastive relative scoring and direct negative log-likelihood supervision only at unmatched positions, and trains only the OCR module while the visual encoder and language backbone stay frozen; across Qwen3-VL-4B and InternVL-3.5-4B, three self-play rounds reduce Vary-600K character error rate relative to two-epoch supervised fine-tuning and improve DocBank, IIT-CDIP, DocVQA and InfographicVQA scores.

  3. arXiv

    SPL raises task completion in all 60 staggered-participation MARL comparisons, averaging a 14.1% gain

    The authors formalize staggered participation (SP), a cooperative MARL setting where agents contribute to one objective over temporally offset participation windows and earlier participants leave task-relevant information for later ones, and propose SPL, a training-time augmentation combining prospective acquisition supervision for earlier agents with outcome-supervised receiver learning for later agents; across 60 MPE/RWARE backbone comparisons SPL shows higher observed mean task completion in every case, averaging a 14.1% difference, with gains extending to eight-agent teams and a physics-based UAV–UGV setting in Isaac Lab.

  4. arXiv

    Removing residual connections in the DDT condition encoder and fusing multi-depth features lets DDT-RFE improve four visual understanding tasks and ImageNet FID with fewer encoder blocks

    The authors remove the residual connections around Self-Attention and MLP in the condition encoder of the Decoupled Diffusion Transformer (DDT), stabilize training with zero-initialized output gates and a mixed orthogonal attention initialization, and form the encoder output by aggregating the input patch embedding with the outputs of all encoder blocks; under ImageNet-1k training at the stated resolution, patch size 2, 80 epochs and batch size 1024, DDT-RFE improves overall over DDT on image classification, semantic segmentation, object discovery and semantic correspondence while using fewer encoder blocks (12 vs. 20 at L, 14 vs. 22 at XL), and lowers ImageNet generation FID from 8.98 to 7.68 at L and from 7.27 to 6.81 at XL.

  5. arXiv

    No Model Required: Text Entropy Rate Filtering Yields +42% Unique Trigrams, +30% Vocabulary, and -19% Repetition in Six-Generation QLoRA Iterative Fine-Tuning

    The work introduces a non-parametric Kontoyiannis entropy rate estimator h_k, computed entirely from raw text with no model of any kind, as a training-data filter; in a six-generation QLoRA collapse experiment on Llama-3.1-8B, h_k-filtering yields +42% unique trigrams, +30% vocabulary, and -19% repetition (all p<0.001), whereas logprob-based filtering provides no significant text-diversity benefit on any metric (p>0.23); h_k is further validated as a cross-domain entropy proxy (beta=0.924, R2=0.746) and collapse detector (rho=+0.454, p<0.0001) across 4 domains, 2 temperatures, 2 generator-scorer model pairs, and 1,520 generated documents.

  6. arXiv

    DCSD decouples credit direction from magnitude in self-distillation, topping 11 benchmarks and lifting math reasoning by 8.45 points over base models

    The work introduces Decoupled Credit Self-Distillation (DCSD), which uses belief-margin probing to set each reasoning step's credit direction and marginal information gain to quantify its contribution magnitude, leaving the privileged teacher to allocate credit only within steps; across 11 mathematical and multimodal reasoning benchmarks DCSD achieves the best overall scores against GRPO, OPSD, RLSD and RLCSD, improving overall score by 8.45 points on mathematical reasoning and 7.01 points on multimodal reasoning over base models, while correcting credit direction for about 6% of tokens and reducing token credit magnitude by roughly 1.5 times.

  7. arXiv

    TextCSP combines sub-region-aware prompts with soft cascade decoding to reach 87.0% average Dice and 4.81 mm HD95 on TextBraTS

    The work proposes TextCSP, a hierarchical text-guided brain tumor segmentation framework built on the TextBraTS baseline with three components: a text-modulated soft cascade decoder that predicts WT→TC→ET in a coarse-to-fine manner, sub-region-aware prompt tuning that uses learnable soft prompts with a LoRA-adapted BioBERT encoder to generate branch-specialized text representations, and text-semantic channel modulators that convert those representations into channel-wise refinement signals; on the TextBraTS dataset it reaches 87.0% average Dice and 4.81 mm average HD95, improving over the previous best TextBraTS by 1.7% and about 6% (0.32 mm) respectively, with consistent gains across all three sub-regions.

You’re all caught up