Tao Wang

Winning systems in the AT-ADD challenge achieved over 90% accuracy in detecting all types of audio deepfakes, showcasing the potential of innovative detection strategies.

Pair-Aware Discriminative Reasoning in UMER reveals critical distinctions between semantically similar candidates, elevating retrieval accuracy in multimodal tasks.

PILOT's proactive framework boosts recommendation system performance by over 40% in search efficiency and achieves up to 1.60% gains in core metrics, all without human oversight.

Delays in cloud-side inference can be effectively compensated, retaining over 80% performance in imitation-learning tasks even under significant latency.

PhysDGM not only generates high-fidelity synthetic time-series data but also boosts predictive performance by up to 48% while slashing data collection costs by an order of magnitude.

Achieving high-quality document restoration without manual prompts, DocPure outperforms traditional methods by leveraging degradation-aware structural insights.

Re-ranking control alone boosts key performance metrics by over 2%, but extending it to fine ranking unlocks even greater gains without sacrificing system stability.

A new benchmark reveals that even the best LLMs lag significantly behind human experts in reviewing national standards, but structured coordination can bridge this gap.

Legacy data is only useful for upgraded robots after reaching a critical competence threshold, revealing a surprising three-phase pattern in transfer learning.

Recovery routing can outperform escalation strategies by leveraging execution feedback, achieving a higher solve rate at only 35% of the typical recovery cost.

Users can now intuitively grasp a robot's inferred goals through its motion, reducing control effort and enhancing collaboration.

AgentX can autonomously iterate on recommendation algorithms, outpacing human-driven processes and fundamentally changing how we approach system development.

Unified vision-language perception in MLLMs is not just an evolution; it’s a critical leap toward achieving artificial general intelligence.

V-Zero achieves fine-grained visual reasoning without any annotated answer labels, outperforming traditional methods in both speed and accuracy.

Spatial reasoning can be transformed from isolated frame predictions to dynamic scene understanding, significantly boosting performance in multi-view and video tasks.

Flux-Guard achieves a breakthrough by enabling effective face editing that simultaneously thwarts face recognition systems without compromising image quality.

Removing gold answer strings from rewritten contexts can cause F1 scores to plummet by up to 64 points, underscoring their critical role in retrieval-augmented QA performance.

Finally, a watermarking method exists that can be embedded directly into the weights of open-source neural speech generation models, enabling proactive copyright protection.