Soumi Chattopadhayay

MACH reveals that preserving modality conflict is crucial for accurately interpreting multimodal intent, challenging the conventional wisdom of enforcing alignment among signals.

PRIME redefines multimodal intent recognition by enabling the restoration of unreliable inputs, leading to improved robustness without sacrificing performance on clean data.