Fengxiang Wang

LC-QAT achieves superior performance in 2-bit quantization with just a fraction of the training data, setting a new standard for data-efficient model optimization.

Current audio-visual generation models struggle to maintain coherence and alignment when scaling to minute-long content, a problem exposed by the new LongAV-Compass benchmark.

Achieve faster and more accurate remote sensing interpretation by intelligently pruning visual tokens based on task-specific semantic and geometric importance, without any training.

Robots can now nimbly respond to new audio-visual commands in real-time, thanks to a meta-RL approach that bypasses the sensory processing bottleneck.

PReD leaps ahead by creating the first foundation model to close the loop on perception, recognition, and decision-making for electromagnetic signals.

MLLMs can now reliably interpret electromagnetic signals even in noisy environments, thanks to a new training framework and benchmark designed specifically for this challenging domain.

MLLMs struggle to effectively zoom into relevant details in ultra-high-resolution remote sensing imagery, but a new staged training framework can teach them when and where to focus for substantial accuracy gains.