Add a recurrent associative matrix to each selected transformer layer so recent key-value relationships can be retrieved without retaining every past token or performing gradient updates. The matrix uses input-dependent retention and write gates, but retrieval is always performed from the pre-write state, preventing the current target from leaking into its own prediction. Frobenius-norm clipping makes the recurrent memory bounded and provides a direct stability control.
Replace fixed-budget token or patch pruning with greedy selection that combines a teacher-derived relevance score and Gaussian-process mutual information. Select an item when it is both relevant and non-redundant, and stop when the largest remaining information gain falls below a calibrated threshold instead of retaining a fixed number of items.
Generate memory slots or attention keys by applying learned unitary transformations to one seed vector instead of storing every slot independently. Select transformations sequentially when they add a sufficiently new direction and reject phase-equivalent or highly coherent candidates, targeting a well-conditioned near-Parseval frame.
Use the number of tokens participating in an attention operation to adapt the projected query/key dimension according to the sharp Johnson–Lindenstrauss bound. Short attention windows receive a dimension limited by n-1, while longer windows use a logarithmic dimension, avoiding a fixed large head dimension when the current attention set contains too few points to require it.
Replace uniform or random landmark selection in low-rank attention with approximate greedy pivoted QR selection on the attention matrix. At each step, select the key whose residual attention column has the largest norm, orthogonalize it against previously selected columns, and stop when the residual energy falls below a target tolerance. Use the selected columns as an adaptive attention basis and reconstruct the full attention output through a small coefficient matrix.
Compress a trained graph-attention model's edge-dependent logits or gates in the line-graph Fourier domain. Smooth edge values become concentrated in low-frequency coefficients, allowing low-frequency coefficients to retain more precision while high-frequency residuals use fewer bits or are discarded.
Use the dimension-specific relation A_3=0 to remove all intermediate channels transforming as the third exterior power of the two-dimensional vector representation. In tensor-product attention or equivariant MLPs, this is an exact algebraic pruning rule rather than approximate low-rank compression.
Convert attention scores into binary incidence set systems at one or several score thresholds, then group queries that induce the same sampled-key pattern. Compute the expensive key-value aggregation once per pattern and reuse it for all queries in the group. The method is exact for the thresholded routing component and approximates dense attention when queries have a small number of stable high-weight neighborhoods.
Use the sharp exponential tail bound to set a local clipping threshold from a desired exceedance probability. Instead of globally clipping activations at a fixed value or percentile, clip each local window at its minimum plus B log(e/delta), where delta is the tolerated fraction of clipped entries.
Replace one-step greedy landmark selection in Nyström attention or kernel compression with a restricted pairwise-lookahead rule. The lookahead is motivated by the paper's explicit obstruction: a signed triangle can make individual column gains exhibit increasing rather than diminishing returns, so the best next column need not belong to the best pair.