diff options
| author | ericmarin <maarin.eric@gmail.com> | 2026-03-31 16:43:47 +0200 |
|---|---|---|
| committer | ericmarin <maarin.eric@gmail.com> | 2026-04-01 15:08:27 +0200 |
| commit | 81d4d604aa43660b732b3538734a52d509d7c5df (patch) | |
| tree | e0341280c3c3f10752aab7fccb2ddd5ed795c889 /proof.md | |
| parent | d1b25fbde6b01529fd1bcfdd5778b6cb378eb865 (diff) | |
| download | vein-81d4d604aa43660b732b3538734a52d509d7c5df.tar.gz vein-81d4d604aa43660b732b3538734a52d509d7c5df.zip | |
refactored examples
Diffstat (limited to 'proof.md')
| -rw-r--r-- | proof.md | 12 |
1 files changed, 7 insertions, 5 deletions
@@ -38,13 +38,15 @@ By grouping the operations we get:  ### Flatten -Just identity mapping. +Just identity mapping because the wires are always Flatten. + ### MatMul -Equal to Gemm with  and . +Equal to Gemm with ,  and . ### Reshape -Just identity mapping. +Just identity mapping because the wires always Flatten. + ### Add ONNX Add node is defined as: @@ -553,8 +555,8 @@ Otherwise, there exists an active pair , the mathematical definition is preserved after any reduction step, it follows that ![[IN_{n+1}] = [IN_n]](https://latex.codecogs.com/svg.image?\inline&space;[\mathrm{IN_{n+1}}]=[\mathrm{IN_n}]). By the inductive hypothesis, ![[IN_{n+1}] = [NN]](https://latex.codecogs.com/svg.image?\inline&space;[\mathrm{IN_{n+1}}]=[\mathrm{NN}]). - By the principle of mathematical induction, the Interaction Net remains semantically equivalent to the original - Neural Network at every step of the reduction process. +By the principle of mathematical induction, the Interaction Net remains semantically equivalent to the original +Neural Network at every step of the reduction process. Since Interaction Nets are confluent, the reduced mathematical expression is unique regardless of order in which rules are applied. |
