diff options
Diffstat (limited to '')
| -rw-r--r-- | notes.norg | 9 |
1 files changed, 3 insertions, 6 deletions
@@ -1,19 +1,16 @@ @document.meta title: Neural Network Equivalence -description: WIP tool to prove NNEQ using Interaction Nets as pre-processor +description: WIP tool to prove NNEQ using Interaction Nets as pre-processor fo my Batchelor's Thesis authors: ericmarin categories: research created: 2026-03-14T09:21:24 -updated: 2026-03-14T18:34:04 +updated: 2026-03-17T11:18:11 version: 1.1.1 @end * TODO - (?) Scalability %Maybe done? I have increased the limits of Inpla, but I have yet to test% - - ( ) Soundness of translated NN - ~~ Define the semantic of the Agents (give a mathematical definition) - ~~ Prove that a Layer L and the Inpla translation represent the same function - ~~ Prove that each Interaction Rules preserve the mathematical semantic of the output + - (x) Soundness of translated NN - ( ) Compatibility with other types of NN - ( ) Comparison with other tool ({https://github.com/NeuralNetworkVerification/Marabou}[Marabou], {https://github.com/guykatzz/ReluplexCav2017}[Reluplex]) - ( ) Add Range agent to enable ReLU optimization |