Pattern-Based Matrix Analysis Under Permutations
Organizations: NL
Abstract
Low-rank approximation captures one important notion of matrix simplicity: a matrix is simple if it is well described by a low-dimensional linear model. This paper studies a complementary notion. We consider matrices that become structurally simple only after permuting rows and columns, so that their mass concentrates in a prescribed geometric pattern such as a band, block-diagonal, arrowhead, or related form. We introduce a pattern distance that measures how close a matrix is to a given pattern after optimal reordering, giving a common framework for structural problems such as latent ordering, community detection, and bandwidth minimization. The problem is NP-hard in general; we derive an alternating algorithm based on Linear Assignment subproblems. We further introduce a pattern-plus-low-rank decomposition that combines permutation-structured fitting with a low-rank residual, whose permutation step is pattern fitting on the residual; its gains across pattern families persist at a matched parameter budget and are substantially reduced under a density-matched random-mask control. We evaluate the framework on synthetic recovery tasks, real correlation and graph adjacency matrices, and transformer attention heads, showing that a single universal algorithm recovers qualitatively different forms of matrix structure.
Figures & tables
| Gauss | Gauss | Sparse | Sparse | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| other | other | other | other | |||||||||||||
| RCM | F+gr. | FAQ | AltLAP | RCM | F+gr. | FAQ | AltLAP | RCM | F+gr. | FAQ | AltLAP | RCM | F+gr. | FAQ | AltLAP | |
| 30 | 0.03 | 0.84 | 0.60 | 0.87 | 0.07 | 0.98 | 0.84 | 0.98 | 0.23 | 0.45 | 0.73 | 0.67 | 0.41 | 0.78 | 0.88 | 0.91 |
| 100 | 0.05 | 0.37 | 0.51 | 0.69 | 0.09 | 0.72 | 0.66 | 0.87 | 0.06 | 0.09 | 0.49 | 0.45 | 0.16 | 0.29 | 0.66 | 0.69 |
| 200 | 0.03 | 0.14 | 0.46 | 0.61 | 0.08 | 0.33 | 0.61 | 0.75 | 0.02 | 0.03 | 0.36 | 0.30 | 0.09 | 0.10 | 0.52 | 0.53 |
| 500 | 0.03 | 0.03 | 0.40 | 0.52 | 0.06 | 0.08 | 0.57 | 0.61 | 0.00 | 0.00 | 0.20 | 0.15 | 0.03 | 0.02 | 0.33 | 0.34 |
Appendix figures & tables33 assets
Supplementary material from the paper’s appendix.
Appendix
| Pattern | Shape | Classical problem | in this paper |
| Order and locality | |||
| Diagonal † | Linear assignment (LAP) | one LAP §§ | |
| Banded ( ) † | Seriation, bandwidth min. [ Atkins et al., 1998 , Cuthill and McKee, 1969 ] | NP-hard at , (App. B.5 ) | |
| Anti-diagonal † | Reverse ordering (diagonal under column reversal) | one LAP §§ | |
| Cyclic band † | Circular seriation [ Recanati et al., 2018 ] | – | |
| Upper triangular † | Linear ordering / feedback arc set | – ¶ | |
| Mask / setting | Classical problem | Specialized methods compared | Where |
|---|---|---|---|
| Fixed , one permutation | Koopmans–Beckmann QAP | QAP-FAQ (as alternative solver) | § 5.1 |
| Fixed , two permutations | Bilinear assignment, graph matching | Birkhoff relaxations (discussed) | § 3 |
| Banded, exact support | Bandwidth / profile reduction | RCM, Gibbs–Poole–Stockmeyer | § 5.1 |
| Banded / Robinson, noisy | Seriation, latent ordering | Spectral and convex seriation | § 5.1 |
| Staircase band (rectangular) | Consecutive-ones, interval seriation | Barycentre, spectral, convex seriation | § 5.1 |
| Cyclic band | Circular seriation | Circular spectral embedding | App. E.6.2 |
| Symbol / term | Meaning |
|---|---|
| rows reordered by , columns by ; , a permutation mapping an original index to its position | |
| masking operator; is the off-pattern part | |
| off-pattern squared Frobenius mass at fixed permutations ( 1 ) | |
| its minimum over independent row and column permutations ( 2 ) | |
| , | minimum under a common permutation ( ); the value at a fixed |
| explained pattern variance |
| Mask, setting | Objective |
|---|---|
| full mask | identically |
| diagonal, | constant (permutation-invariant) |
| upper triangle, , symmetric | constant |
| upper triangle, , asymmetric | weight of the backward arcs (linear ordering) |
| diagonal or anti-diagonal, independent | one LAP (Lemma 1 ) |
| arrowhead, | exact: choice of the hub, |
| Regime | RCM | GPS | F+greedy | Spectral | Convex | FAQ | FAQ+AltLAP | AltLAP-r1 | AltLAP | |
|---|---|---|---|---|---|---|---|---|---|---|
| Gauss | 30 | 0.03/0.39 | 0.05/0.41 | 0.84/0.90 | 0.76/0.85 | 0.85/0.91 | 0.60/0.75 | 0.68/0.80 | 0.87/0.92 | 0.87/0.92 |
| 100 | 0.05/0.34 | 0.05/0.34 | 0.37/0.56 | 0.32/0.52 | 0.36/0.56 | 0.51/0.66 | 0.62/0.73 | 0.69/0.78 | 0.69/0.78 | |
| 200 | 0.03/0.30 | 0.04/0.31 | 0.14/0.38 | 0.15/0.39 | 0.14/0.38 | 0.46/0.61 | 0.58/0.70 | 0.61/0.72 | 0.61/0.72 | |
| 500 | 0.03/0.27 | 0.03/0.27 | 0.03/0.27 | 0.04/0.28 | 0.04/0.28 | 0.40/0.55 | 0.52/0.64 | 0.52/0.64 | 0.52/0.64 | |
| Gauss | 30 | 0.07/0.54 | 0.09/0.55 | 0.98/0.99 | 0.97/0.98 | 0.98/0.99 | 0.84/0.92 | 0.86/0.93 | 0.98/0.99 | 0.98/0.99 |
| 100 | 0.09/0.41 | 0.09/0.41 | 0.72/0.82 | 0.61/0.75 | 0.73/0.82 | 0.66/0.78 | 0.72/0.82 | 0.87/0.91 | 0.87/0.91 |
| Bandwidth | Profile | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Regime | identity | RCM | GPS | F+greedy | Convex | FAQ | AltLAP | RCM | GPS | AltLAP | |
| Gauss | 30 | 29 | 26 | 26 | 28 | 29 | 29 | 29 | 416 | 417 | 422 |
| 100 | 97 | 82 | 83 | 96 | 95 | 97 | 96 | 4251 | 4515 | 4466 | |
| 200 | 196 | 162 | 159 | 192 | 192 | 197 | 195 | 16809 | 17712 | 17592 | |
| 500 | 494 | 405 | 399 | 489 | 490 | 496 | 492 | 104377 | 109910 | 110834 | |
| Gauss | 30 | 29 | 26 | 26 | 28 | 29 | 29 | 29 | 416 | 417 | 423 |
| Pattern | input | SVD-dof | SVD-best | RPCA | pattern only | PatLR | PatLR-rnd | oracle | |
|---|---|---|---|---|---|---|---|---|---|
| planted pattern, SNR 4.0 | |||||||||
| Banded | 0.035 | 0.75 | 0.12 | 0.75 | 0.10 | 0.76 | 0.76 | 0.55 | 0.88 |
| Block | 0.250 | 0.75 | 0.34 | 0.75 | 0.21 | 0.92 | 0.91 | 0.54 | 0.91 |
| Arrowhead | 0.015 | 0.75 | 0.34 | 0.75 | 0.19 | 0.78 | 0.78 | 0.36 | 0.78 |
| Checkerboard | 0.500 | 0.75 | 0.53 | 0.75 | 0.34 | 0.87 | 0.86 | 0.60 | 0.86 |
| Block+band | 0.327 | 0.75 | 0.41 | 0.75 | 0.25 | 0.91 | 0.89 | 0.55 | 0.89 |
| Pattern | SVD-dof | pattern only | PatLR, input order | PatLR | PatLR, planted order |
|---|---|---|---|---|---|
| Banded | 0.10 | 0.73 | 0.29 | 0.73 | 0.86 |
| Block | 0.41 | 0.96 | 0.31 | 0.96 | 0.96 |
| Arrowhead | 0.28 | 0.48 | 0.74 | 0.74 | 0.73 |
| Checkerboard | 0.68 | 0.97 | 0.54 | 0.97 | 0.97 |
| Block+band | 0.50 | 0.96 | 0.38 | 0.97 | 0.96 |
| Anti-diagonal | 0.21 | 0.81 | 0.14 | 0.83 | 0.92 |
| Pattern | input | SVD- | SVD-dof | RPCA | BM | BM-dof | PatLR | PatLR-rnd | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Banded | 0.035 | 7 | 51 | 0.57 | 0.05 | 0.10 | 0.07 | 0.04 | 0.27 | 0.73 | 0.52 |
| Block | 0.250 | 30 | 106 | 0.92 | 0.06 | 0.41 | 0.18 | 0.04 | 0.42 | 0.96 | 0.61 |
| Arrowhead | 0.015 | 5 | 42 | 0.13 | 0.28 | 0.28 | 0.05 | 0.27 | 0.43 | 0.74 | 0.29 |
| Checkerboard | 0.500 | 63 | 146 | 0.96 | 0.06 | 0.68 | 0.31 | 0.04 | 0.62 | 0.97 | 0.73 |
| Block+band | 0.327 | 40 | 120 | 0.93 | 0.06 | 0.51 | 0.22 | 0.04 | 0.42 | 0.97 | 0.64 |
| Anti-diagonal | 0.102 | 14 | 73 | 0.79 | 0.06 | 0.21 | 0.13 | 0.04 | 0.26 | 0.83 | 0.57 |
| SNR 4 | SNR 1 | |||||||
| planted mask | Oracle | Adaptive | SVD | exact / cont. | Oracle | Adaptive | SVD | exact / cont. |
| pattern only | ||||||||
| block | 0.86 | 0.86 | 0.75 | 100 | 0.44 | 0.46 | 0.01 | 100 |
| block | 0.89 | 0.89 | 0.75 | 40 / 60 | 0.60 | 0.60 | 0.01 | 45 / 55 |
| block | 0.91 | 0.91 | 0.75 | 100 | 0.68 | 0.68 | 0.02 | 100 |
| block | 0.92 | 0.92 | 0.75 | 100 | 0.75 | 0.74 | 0.03 | 95 / 5 |
| , | Spectral | Convex | FAQ | AltLAP-r1 | vs spectral [CI], seeds / | vs convex [CI], seeds / | |
|---|---|---|---|---|---|---|---|
| 6 | 100, 0.05 | 0.981 | 0.992 | 0.269 | 1.000 | [ ], 20/0 | [ ], 20/0 |
| 6 | 100, 0.1 | 0.943 | 0.975 | 0.357 | 0.986 | [ ], 20/0 | [ ], 17/3 |
| 6 | 100, 0.2 | 0.824 | 0.905 | 0.293 | 0.861 | [ ], 17/3 | [ ], 9/11 |
| 6 | 100, 0.3 | 0.673 | 0.700 | 0.225 | 0.722 | [ ], 18/2 | [ ], 15/5 |
| 6 | 300, 0.05 | 0.811 | 0.861 | 0.159 | 0.828 | [ ], 18/2 | [ ], 8/12 |
| 6 | 300, 0.1 | 0.543 | 0.549 | 0.165 | 0.554 | [ ], 14/6 | [ ], 8/12 |
| , | Spectral | Convex | ||||
|---|---|---|---|---|---|---|
| 6 | 100, 0.1 | 0.943 | 0.975 | 0.981 | 0.986 | 0.983 |
| 6 | 100, 0.3 | 0.673 | 0.700 | 0.710 | 0.722 | 0.726 |
| 6 | 300, 0.1 | 0.543 | 0.549 | 0.547 | 0.554 | 0.553 |
| 6 | 300, 0.3 | 0.250 | 0.284 | 0.255 | 0.256 | 0.261 |
| 12 | 100, 0.1 | 0.983 | 0.988 | 0.989 | 0.991 | 0.972 |
| 12 | 100, 0.3 | 0.921 | 0.939 | 0.967 | 0.972 | 0.913 |
| Method | ||||
|---|---|---|---|---|
| Sort by marginals | 0.05 / 0.04 / 582 | 0.05 / 0.05 / 638 | 0.06 / 0.06 / 734 | 0.07 / 0.06 / 837 |
| Barycentre reordering | 0.73 / 0.66 / 295 | 0.61 / 0.51 / 448 | 0.53 / 0.42 / 608 | 0.46 / 0.36 / 732 |
| Spectral seriation | 0.90 / 0.87 / 203 | 0.84 / 0.66 / 447 | 0.67 / 0.44 / 659 | 0.49 / 0.37 / 790 |
| Convex seriation | 0.92 / 0.92 / 163 | 0.85 / 0.85 / 312 | 0.71 / 0.69 / 550 | 0.50 / 0.53 / 718 |
| Spectral + 2-SUM local search | 0.92 / 0.88 / 181 | 0.85 / 0.83 / 332 | 0.71 / 0.61 / 570 | 0.49 / 0.53 / 721 |
| Staircase mask, AltLAP (cold) | 0.99 / 1.00 / 66 | 0.99 / 0.99 / 164 | 0.98 / 0.97 / 335 | 0.95 / 0.91 / 502 |
| Setting | noise | convex | fixed mask | re-estimated | noise | convex | fixed mask | re-estimated |
|---|---|---|---|---|---|---|---|---|
| , (as Table 14 ) | 179 | 0.86 / 317 | 1.00 / 166 | 1.00 / 165 | 540 | 0.51 / 716 | 0.92 / 496 | 0.93 / 494 |
| , | 161 | 0.80 / 328 | 0.98 / 155 | 0.98 / 153 | 483 | 0.49 / 669 | 0.77 / 461 | 0.77 / 451 |
| , | 161 | 0.88 / 247 | 1.00 / 145 | 1.00 / 145 | 483 | 0.61 / 574 | 0.97 / 430 | 0.97 / 430 |
| irregular start profile | 179 | 0.71 / 334 | 0.85 / 188 | 0.85 / 184 | 540 | 0.51 / 695 | 0.80 / 510 | 0.80 / 507 |
| interval lengths | 179 | 0.86 / 304 | 0.96 / 189 | 0.97 / 182 | 540 | 0.53 / 704 | 0.86 / 515 | 0.85 / 510 |
| Noise | Spectral (linear) | Circ. spectral | Circ. spectral AltLAP | AltLAP, cyclic (cold) | vs circ. spectral [CI] | seeds / |
|---|---|---|---|---|---|---|
| 0.3 | 0.372 | 0.963 | 0.965 | 0.958 | [ ] | 6/13 |
| 0.5 | 0.393 | 0.903 | 0.934 | 0.938 | [ ] | 13/7 |
| 0.7 | 0.332 | 0.782 | 0.879 | 0.885 | [ ] | 18/2 |
| 1.0 | 0.203 | 0.431 | 0.471 | 0.469 | [ ] | 11/9 |
| Degree sort | Spectral | AltLAP, nested mask | Degree sort AltLAP | |
|---|---|---|---|---|
| 0.0 | 0.994 / 0 | 0.586 / 1277 | 0.994 / 0 | 0.994 / 0 |
| 0.02 | 0.967 / 134 | 0.573 / 1382 | 0.965 / 129 | 0.963 / 131 |
| 0.05 | 0.942 / 301 | 0.544 / 1658 | 0.932 / 299 | 0.925 / 301 |
| 0.1 | 0.906 / 574 | 0.548 / 1712 | 0.881 / 586 | 0.862 / 587 |
| 0.15 | 0.876 / 819 | 0.507 / 1931 | 0.817 / 843 | 0.806 / 842 |
| Bethe H. | Non-backtr. | Reg. sp. | Leiden | Swap only | Ours | BH ours | vs BH [CI] | seeds / | ||
|---|---|---|---|---|---|---|---|---|---|---|
| 1.3 | 8 | 0.42 | 0.38 | 0.41 | 0.13 | 0.05 | 0.15 | 0.47 | [ ] | 0/20 |
| 1.3 | 16 | 0.43 | 0.39 | 0.35 | 0.15 | 0.08 | 0.22 | 0.51 | [ ] | 1/19 |
| 1.3 | 32 | 0.47 | 0.39 | 0.40 | 0.13 | 0.11 | 0.35 | 0.57 | [ ] | 2/18 |
| 1.6 | 8 | 0.73 | 0.72 | 0.73 | 0.61 | 0.13 | 0.70 | 0.78 | [ ] | 11/9 |
| 1.6 | 16 | 0.74 | 0.74 | 0.75 | 0.71 | 0.32 | 0.82 | 0.82 | [ ] | 20/0 |
| 1.6 | 32 | 0.78 | 0.78 | 0.78 | 0.83 | 0.58 | 0.86 | 0.86 | [ ] | 18/2 |
| Spectral | Reg. sp. | Bethe H. | Non-b. | Leiden | Squared | KL (lin.) | AltLAP-lin | Spec. ours | vs spectral [CI] | seeds / | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.30 | 0.403 | 0.028 | 0.064 | 0.052 | 0.033 | 0.003 | 0.324 | 0.356 | 0.556 | [ ] | 7/13 |
| 0.35 | 0.680 | 0.073 | 0.119 | 0.102 | 0.054 | 0.004 | 0.807 | 0.827 | 0.838 | [ ] | 20/0 |
| 0.40 | 0.850 | 0.154 | 0.218 | 0.179 | 0.092 | 0.006 | 0.927 | 0.932 | 0.932 | [ ] | 19/1 |
| 0.50 | 0.975 | 0.482 | 0.560 | 0.445 | 0.354 | 0.018 | 0.994 | 0.995 | 0.995 | [ ] | 18/1 |
| 0.60 | 0.996 | 0.821 | 0.814 | 0.813 | 0.866 | 0.050 | 1.000 | 1.000 | 1.000 | [ ] | 7/0 |
| Ours, mask sizes | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Planted sizes | Bethe H. | Reg. spectral | Leiden | KL (true) | BH ours (true) | equal | true | estimated | |
| equal | 1.6 | 0.74 | 0.75 | 0.74 | 0.32 | 0.79 | 0.82 | 0.82 | 0.80 |
| equal | 2.0 | 0.92 | 0.92 | 0.95 | 0.94 | 0.95 | 0.95 | 0.95 | 0.94 |
| 1:1:2:2 | 1.6 | 0.59 | 0.54 | 0.52 | 0.66 | 0.52 | 0.49 | 0.84 | 0.65 |
| 1:1:2:2 | 2.0 | 0.89 | 0.85 | 0.93 | 0.92 | 0.68 | 0.59 | 0.96 | 0.89 |
| Planted sizes | Spectral | KL-lin (true) | Squared loss (true) | equal | true | estimated | |||
| Method | Karate ( ) | Davis ( ) | Polbooks ( ) | Football ( ) |
|---|---|---|---|---|
| Spectral clustering | 0.882 | 0.000 | 0.465 | 0.813 |
| Regularized spectral | 0.882 | 0.000 | 0.624 | 0.869 |
| Bethe Hessian | 0.882 | 1.000 | 0.659 | 0.897 |
| Non-backtracking | 0.882 | 1.000 | 0.405 | 0.906 |
| Leiden | 0.465 | 0.405 | 0.621 | 0.807 |
| AltLAP, equal blocks | 0.772 | 0.777 | 0.455 | 0.739 |
| Method | target ER | block ARI | within-block |
| Spectral clustering | 0.057 | 1.000 | 0.09 |
| Fiedler ordering | 0.050 | 0.839 | 0.10 |
| Two-stage (cluster, then order) | 0.115 | 1.000 | 0.45 |
| QAP-FAQ (composite mask) | 0.137 | 0.181 | 0.18 |
| AltLAP (composite mask) | 0.171 | 0.781 | 0.20 |
| AltLAP + boundary estimation | 0.173 | 0.877 | 0.20 |
| target ER | within-block | ER, convex inside | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| oracle | pipel. | Alt | rel. | [CI] | pipel. | Alt | [CI] | pipel. | Alt | |||
| 3 eq. | 1 | 0.1 | 0.130 | 0.043 | 0.104 | +142% | [ ] | 0.31 | 0.32 | [ ] | 0.045 | 0.105 |
| 3 eq. | 1 | 0.3 | 0.086 | 0.025 | 0.084 | +239% | [ ] | 0.17 | 0.16 | [ ] | 0.026 | 0.084 |
| 3 eq. | 3 | 0.1 | 0.305 | 0.196 | 0.288 | +47% | [ ] | 0.71 | 0.76 | [ ] | 0.215 | 0.290 |
| 3 eq. | 3 | 0.3 | 0.214 | 0.100 | 0.200 | +100% | [ ] | 0.41 | 0.43 | [ ] | 0.100 | 0.200 |
| 3 uneq. | 1 | 0.1 | 0.116 | 0.036 | 0.091 | +154% | [ ] | 0.29 | 0.30 | [ ] | 0.038 | 0.093 |
| Head type | True mask | True mask selected | Labels selected | Margin |
|---|---|---|---|---|
| Local | band | 15/20 | band (15), block+band (3), block (1), block (1) | 0.03 |
| Topic | block | 12/20 | block (12), block+band (8) | 0.56 |
| Global | arrowhead | 20/20 | arrowhead (20) | 0.60 |
| Self | diagonal | 0/20 | band (20) | 0.02 |
| label share (%) | agreement over prompts | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | arrowhead [CI] | block- 2 | other blocks | bands | all | arrowhead | other | confident | null | scalars / maj. | |
| GPT-2 | 512 | 69.3 [69.0, 69.7] | 14.8 | 12.9 | 3.0 | 0.91 | 0.99 | 0.74 | 0.77 | 0.07 | 0.87 / 0.69 |
| 128 | 72.2 [71.7, 72.6] | 12.8 | 10.8 | 4.1 | 0.90 | 0.98 | 0.68 | 0.78 | 0.08 | 0.86 / 0.72 | |
| Qwen2.5-0.5B | 512 | 67.1 [66.5, 67.6] | 15.9 | 15.4 | 1.6 | 0.86 | 0.96 | 0.64 | 0.71 | 0.12 | 0.81 / 0.67 |
| 128 | 69.1 [68.3, 69.8] | 16.3 | 12.9 | 1.7 | 0.86 | 0.96 | 0.62 | 0.74 | 0.12 | 0.81 / 0.69 | |
| Qwen2.5-7B | 512 | 79.9 [79.4, 80.3] | 11.4 | 8.2 | 0.5 | 0.92 | 0.98 | 0.67 | 0.83 | 0.09 | 0.89 / 0.80 |
| arrowhead share | labels changed vs identity | |||||
|---|---|---|---|---|---|---|
| Model | identity | optimized | unified | unified [CI] | optimized | solver exact |
| GPT-2 | 0.70 | 0.70 | 0.69 | 3.5% [3, 4] | 19% | 99.9% |
| Qwen2.5-0.5B | 0.70 | 0.68 | 0.67 | 6.5% [6, 7] | 24% | 99.8% |
| Qwen2.5-7B | 0.61 | 0.80 | 0.80 | 23.4% [8, 39] | 34% | 99.6% |
| time | ER, | ER, | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Fiedler | per iter. | LAPs | total | FAQ | memory | AltLAP | FAQ | AltLAP | time | |
| 500 | 0.0 | 0.04 | 0.03 | 0.4 | 0.7 | 0.2 GB | 0.51 (0.80) | 0.57 | 0.51 (0.80) | 1.0 |
| 1000 | 0.1 | 0.14 | 0.10 | 1.5 | 4.7 | 0.4 GB | 0.44 (0.67) | 0.48 | 0.52 (0.80) | 1.6 |
| 2000 | 1.0 | 0.73 | 0.49 | 8.3 | 32.5 | 0.6 GB | 0.34 (0.50) | 0.38 | 0.55 (0.80) | 7.5 |
| 4000 | 10.9 | 5.4 | 3.1 | 65 | 249 | 1.3 GB | 0.24 (0.34) | 0.25 | 0.59 (0.80) | 46 |
| 8000 | 105 | 37 | 18 | 479 | 1936 | 3.2 GB | 0.15 (0.20) | 0.16 | 0.63 (0.80) | 334 |
Combinatorial Optimization
Combinatorial optimization finds the best feasible choices among discrete alternatives in routing, scheduling, graph problems, and resource allocation.
See all 413 papers