Skip to main content
Search

Hong Kong Polytechnic University

32 related articles

Subscribe to updates
  1. arXiv

    Changing only 5% of pixels blocks face cloning: SEAL holds identity similarity to 0.15 or less on CelebA-HQ and VGGFace2

    Adversarial perturbations added to photos can stop a diffusion model from learning a person's identity, but earlier methods altered the whole face densely, and the accumulated distortion became visible around facial features. SEAL confines the perturbation to 5% of spatial locations, jointly optimizing where it lands and its values with a differentiable top-k mask, and disrupts the latent variance of a frozen VAE encoder. On CelebA-HQ and VGGFace2, protected images reach above 41 dB PSNR, 0.97 SSIM and at most 0.07 LPIPS, while identity similarity after DreamBooth personalization stays at most 0.15.

  2. arXiv

    Rooftop PV modules mapped without per-module labels: MV-PVMapper recovers 7,710 of 7,711 modules across three plants

    On five-direction UAV oblique surveys at roughly 1 cm nadir ground sampling distance, MV-PVMapper couples the frozen vision foundation model SAM 3 with photogrammetric geometry to output georeferenced polygons for individual modules without task-specific annotation or weight updates. Across three operational rooftop PV plants holding 7,711 reference modules, two feedback rounds match 7,710 with no false positives at an IoU threshold of 0.5; macro-averaged AJI reaches 97.55%, against 60.73% for an orthomosaic segmentation baseline and 41.24% for a multi-view pixel-voting baseline.

  3. arXiv

    CADS scores an LLM proposer against the same screen with the LLM removed, and raw hit counts overstate its gain

    In sequential biological screens, LLM-proposed candidates are usually credited by hit counts, but the re-selection policy and random fallback also produce hits on their own. CADS runs the full loop beside an otherwise identical proposer-removed loop, certifies each discovery set separately, and reports the paired difference; in DepMap-backed simulations the no-LLM loop alone cleared the standard feedback gate, and origin counts inflated Qwen2.5-7B's net contribution from near zero to apparently positive.

  4. arXiv

    Janus turns red-blue adversarial search on deserialization defenses, cutting held-out attack success to 6.2% across 16 deployments

    Unsafe deserialization lets attacker-controlled data trigger chains of existing methods and reach code execution, while gateway, application and system controls each see only part of the attack. Janus runs a black-box attacker agent against an isolated replica to surface exploit variants, has a defender agent summarize each verified variant into security-relevant effects, and uses counterfactual replay to separate complementary coverage from redundant enforcement before jointly selecting and scoping controls. Across 16 deployments in six Java and PHP applications, held-out attack success falls to 6.2% versus 18.8%–55.0% for gateway, application, system and cross-layer baselines, while 98.4% of benign functionality tests pass.

  5. arXiv

    Tail calibration flags when a larger Best-of-N budget backfires, and code generation is where it shows

    Best-of-N inference returns the candidate with the highest proxy reward, so a larger budget reads the extreme tail of the reward distribution rather than its average ranking quality. The work writes the BoN value as a Beta-weighted integral of conditional true reward over within-prompt score quantiles, showing that BoN depends on scores only through their ranks and can decline only when the calibration curve in the tail turns downward. On public candidate pools, negative tail excess concentrates in code generation, and a cross-fitted tail statistic predicts whether a budget cap pays off on held-out data about as well as rank accuracy.

  6. arXiv

    Edge AI agents now choose their own next step: up to 26% higher accuracy and 28% lower resource use than fixed workflows

    Edge AI development has relied on developers hand-building fixed agent workflows that struggle with heterogeneous data modalities and diverse task pipelines. SE2 distills development into four atomic operations, retrieve, reason, code and test, and lets a coordinator agent pick the next one from the current context, while mobile runtime feedback and reinforcement fine-tuning let the agents keep evolving. On the IoTBench benchmark and three multimodal edge AI applications, its synthesized programs reach up to 26% higher accuracy and 28% lower resource consumption than eight state-of-the-art baselines.

  7. arXiv

    Same tools, different interface: agent behavior shifts, and repairing only the expression lifts six settings by up to 9.0 points

    With executable capabilities held fixed, presenting them as MCP tools or as CLI commands still sends language agents down different task trajectories. Built from a shared semantic action contract, the comparison spans six models and three interactive environments, and 94.3% of 2,004 paired episodes show structured divergence. DART, which repairs only the interface expression once a semantic decision is validated, raises benchmark score in all six benchmark–interface settings by up to 9.0 points and lowers MCP token cost in all three environments.

  8. arXiv

    Move the camera and the geometry features move too: splitting viewpoint out of the world state buys a 1.4-second forecast

    Driving 3D world models often treat geometry-foundation-model features as future states, yet those features shift with camera viewpoint, so the model must predict scene evolution and camera motion at once. GeoState compresses multi-view features into a view-decoupled scene state that a camera ray queries only at decoding, while camera motion is modeled separately. On Waymo it beats compared world models at every reported horizon and transfers zero-shot to KITTI and nuScenes.

  9. arXiv

    Looped reasoning stops keeping only the latest state: attention residuals hold solutions at nearly 25,000 layers

    Looped Transformers deepen reasoning by re-running a shared block at test time, but the usual carry-last rule lets later iterations overwrite correct intermediate results. InfiLoop replaces that rule with learned attention over recurrent outputs: a shared pseudo-query scores each candidate once, a learned exponential decay discounts older ones, and the aggregation has an exact streaming recurrence needing only a weighted-state vector and a normalization scalar per token. On Sudoku-Extreme a 7M-parameter model reaches 97.9% exact accuracy versus 94.2% for FPRM, and hits 90.9% at 1,872 layers where FPRM needs 3,744 layers for 90.1%.

  10. arXiv

    Reconfigurable optical links let AI data centers chase cheap, green power: 12.50% lower real-time grid cost and 2.724 points more local renewables

    Across five AI data centers with equal completed work and network resources, folding optical circuit switching into joint compute-electricity-network scheduling cut real-time grid purchase cost by 12.50% and raised local renewable utilization by 2.724 percentage points versus fixed optical operation.