Search papers, labs, and topics across Lattice.
Affiliation:
3
0
6
Length generalization in transformers does not require double-exponential sample sizes: viewing inputs as compressed strings via power words slashes worst-case sample complexity down to polynomial.
Linear RNNs achieve transformer-like parallelization because they're essentially log-depth arithmetic circuits, while nonlinear RNNs are fundamentally limited by their ability to solve computationally harder problems.
Unlock formal verification for transformer-level programs by translating them into synchronous dataflow languages, enabling rigorous analysis and optimization.
