Yuhang Jiang

The paper introduces PInVerify, an offline embodied benchmark for Active Instance Verification (AIV), a task requiring agents to actively select viewpoints to confirm if a candidate object matches a fine-grained natural-language description.

The paper demonstrates that in Mamba-2, single-bucket probes can detect a large functional signature (detection layer) that is not fully responsible for the actual computation (execution layer), challenging the assumption that representational similarity implies functional equivalence.

The paper demonstrates that the location and nature of state encoding in sequence models are not fixed architectural traits but are highly dependent on the specific task, showing that the encoding profile can reverse across different tasks and architectures.