Skip to main content
Search

Southeast University

22 related articles

Subscribe to updates
  1. arXiv

    ForceWAM splits contact learning in two: demos set the goal, play data teach force response, reaching 95% on one real-robot task

    In contact-rich manipulation, expert demonstrations cover successful paths but say little about how force responds to different actions, while task-agnostic play records broader contact dynamics without a task goal. ForceWAM gives the task-compatible motion and six-axis force/torque future to a force world model trained on demonstrations, and gives the question of which actions realize that future to an inverse dynamics model trained on both demonstrations and play; at deployment, fresh wrench feedback corrects actions between world-model updates. Across force-regulated sliding, box flipping with and without disturbances, and ring placement, the full system reaches 85%, 95%, 85%, and 90% success, above the listed baselines.

  2. arXiv

    Web-GUI agents pay for every screenshot pixel: keeping 21% of visual tokens cuts latency 31% and holds 77.7% of performance

    Web-GUI agents locate buttons and links from screenshots, and high-resolution screenshots plus multi-turn histories create excessive visual tokens that raise inference cost. STP-GUI replaces independent token scoring with query-guided, spatially coherent window-level selection, trained through a lightweight selector while the vision-language backbone stays frozen. On Online-Mind2Web, retaining 21% of tokens preserves 77.7% of full-token performance while cutting inference latency by 31% and FLOPs by 57%; results on Android in the Wild suggest the method transfers to mobile GUI environments.

  3. arXiv

    Brain connectomics stops scanning every slice slowly: ZoomSEM rebuilds 16× detail from real low-resolution SEM scans

    Serial electron microscopy is the workhorse of brain connectomics, yet imaging a cubic millimeter of brain takes months to years. ZoomSEM trains a super-resolution model on real low- and high-resolution SEM image pairs and recovers 4×, 8× and up to 16× higher-resolution images from fast low-resolution scans while preserving the fine neuronal structures connectomic analysis needs, reaching up to 30-fold effective acceleration. Experiments across datasets of diverse biological samples show robust generalization under realistic low-resolution acquisition.

  4. arXiv

    UAV SAR drops the straight-and-level assumption: a diffusion prior plus motion compensation holds 21.3 dB at half physical resolution

    A UAV ISAC synthetic aperture radar imager casts imaging as an exact linear inverse problem, using a pretrained diffusion model as prior and the echoes as likelihood guidance to reconstruct super-resolution images under arbitrary known trajectories and unknown position errors.

  5. Low-dose CT denoising drops the clean-image pairs: align adjacent slices, and cone-beam CT works in image domain

    Low-dose CT denoising can now be trained from adjacent slices alone: a matching network estimates inter-slice deformation and corrects the denoiser output, reaching 40.08 dB PSNR on Mayo 2016 and, for the first time, image-domain self-supervised low-dose cone-beam CT denoising.

  6. arXiv

    Memory retrieval stops reading only whole-sentence vectors: reusing a frozen encoder's token states lifts Qwen3-4B average Recall@10 from 62.5 to 75.7

    Memory retrieval no longer compares only whole-memory vectors: sparse matching over token states a frozen encoder already computes, fused with whole-memory similarity, raises average Recall@10 on Qwen3-Embedding-4B from 62.5 to 75.7 without additional training.

  7. 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.

  8. arXiv

    CITA trains a comparative inference model with a Bayesian tool-graph simulator and LLM comparison judgments, consistently improving Tool F1 and task success across three tool-use benchmarks and multiple backbone LLMs

    The work proposes Comparative Inference for Tool-use Agents (CITA), which trains a Comparative Inference Model (CIM) from paired signals combining observed tool behavior, scalable supervision from a Bayesian tool-graph simulator, and semantic judgments from LLM-based comparison, so that the model estimates how likely a possible next tool invocation is to support final task success under the current context; across three tool-use benchmarks and multiple backbone LLMs, CITA consistently improves Tool F1 and task success, and additional analysis shows CIM learns accurate step-level value estimates for comparative tool choices.

  9. arXiv

    CIPO trains LLM agents with counterfactual branches to adaptively pick tool granularity, raising success and decision compression across three benchmarks and two backbones

    The work proposes CIPO: it first mines executable composite skills from successful tool-use trajectories via budget-constrained BPE-style merging, then augments GRPO by branching at the first eligible granularity decision of each base rollout and using the paired outcome difference as a bounded supplementary reward; across TOOLATHLON, TOUCAN, and TRAJECT-Bench with Qwen2.5-7B and Llama3.1-8B, CIPO attains the highest Tool F1 and TSR in all six settings with an average Comp./Raw of 0.845 versus 0.912 for GRPO + Skills, and the gain does not come from simply invoking skills more often.

  10. arXiv

    OG-OPSD selects KL direction by outcome correctness and truncates prefixes by entropy, lifting Qwen3 1.7B/4B/8B math averages to 44.4/64.9/66.4

    Analyzing on-policy self-distillation (OPSD) through the advantage formulation of RLVR, this work finds that a fixed divergence objective under-penalizes and over-rewards incorrect trajectories and that teacher supervision reliability varies with trajectory outcome and cumulative average teacher entropy; it proposes OG-OPSD, which applies forward KL across correct trajectories and reverse KL only within a prefix of incorrect trajectories whose cutoff is set by the last crossing of the cumulative average teacher entropy curves, consistently outperforming vanilla OPSD and multiple strong baselines on mathematical reasoning, multimodal reasoning, and out-of-distribution tasks across Qwen3-1.7B/4B/8B and Qwen3-VL-2B.