Search papers, labs, and topics across Lattice.
Institute of Plasma Physics, Chinese Academy of Sciences
5
0
9
0
Achieve more reliable and interpretable virtual cell perturbation predictions by combining knowledge-driven multimodal modeling with evidence retrieval.
By reformulating particle-in-cell simulations for matrix processing units and co-designing compute, layout, and communication, POLAR-PIC achieves up to 10.9x speedups and maintains 67.5% weak scaling efficiency on over 2 million cores, outperforming GPU-optimized baselines.
Ignoring modality uncertainty in multimodal sarcasm detection hurts performance, but modeling it with a Gaussian posterior and using it to regulate fusion achieves state-of-the-art results.
Achieve superior exposure correction by explicitly modelling and addressing spatially non-uniform degradations, a common but often overlooked challenge in real-world images.
Current multimodal role-playing agents often fail at nuanced emotional expression, exhibiting biases towards positive emotions and struggling with role generalization, as revealed by a new decoupled evaluation framework.
