Researcher, Qualcomm Inc, QualComm
1 paper at NeurIPS 2025
We show that multiplicative (bi-linear) hidden state transitions are a natural choice for representing state tracking behavior in linear recurrent networks.