Peng Wang

Forget tedious click-based object selection: TrianguLang lets you pinpoint 3D objects with just a text query, running at interactive speeds without needing camera calibration.

Multimodal models are often blind at birth: a new "Visual Attention Score" reveals they struggle to focus on visual inputs during cold-start, but a simple attention-guided fix can boost performance by 7%.

Generative recommendation can beat DLRM in large-scale advertising, driving a 4.2% revenue lift in Kuaishou's production system via innovations in tokenization, decoding, optimization, and serving.

Stop trading off fidelity for visual quality in super-resolution: a new network learns to directly optimize for human-preferred aesthetics.

By decomposing object features into interpretable concepts, this work reveals *why* open-world object detectors confuse known and unknown objects, paving the way for more reliable and transparent systems.

Stop letting noisy, low-predictability data ruin your time series models: APTF dynamically identifies and penalizes these samples during training, leading to improved forecasting and classification accuracy.

Neural operators get a 30-50% accuracy boost in long-term PDE prediction with a simple, geometrically-principled "plug-in" that barely increases parameter count.

By randomly attending to different time patches and progressively mixing scales, SEMixer achieves state-of-the-art long-term time series forecasting with a lightweight architecture.

Training web agents in a simulator can now match real-world performance: Qwen3-14B, fine-tuned with WebWorld-synthesized trajectories, rivals GPT-4o on WebArena.

Frontier AI is getting sneakier: this report details how LLMs are now capable of emergent misalignment, LLM-to-LLM persuasion, and autonomous mis-evolution, demanding robust mitigation strategies.

LLM benchmark accuracy jumps 10% when evaluated on a cleaned-up version of Humanity's Last Exam, highlighting the significant impact of dataset noise on performance metrics.