Unverified50% confidenceFactExact time
A rank-one matrix stretches space in only one direction; when used as an update, the memory matrix changes along only one direction, and existing components in the other directions are unaffected.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFlattening a matrix and feeding it to an MLP results in a non-fixed input dimension due to variable row counts, and breaks permutation invariance.74% similarUnverifiedNever scale a node to exactly zero; a degenerate transform matrix disrupts culling and transform calculations, so scaling should be clamped to at least 0.02 and nearly flattened nodes hidden.69% similarUnverifiedThe eigenvalues of orthogonal matrix initialization have a modulus of 1, which neither grows nor decays exponentially under repeated matrix multiplication, effectively mitigating vanishing and exploding gradients.67% similarUnverifiedThe core insight of KronLSTM is that each correction proceeds along only a single direction of the matrix, minimizing interference.66% similarUnverifiedMatrix multiplication is typically compute-bound in high-batch inference scenarios, while autoregressive decoding with a batch size of 1 is almost invariably bandwidth-bound.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073426API
curl https://kongchang.com/api/v1/knowledge/claims/1073426MCP
get_claim(id=1073426)