The learning with errors (LWE) search problem is thought to be a suitable framework for post-quantum cryptography, because time required to achieve an exact solution for it is not known to be less than exponential. There exists an LWE decision problem which can be reduced from the LWE search problem, so this paper applies a support vector machine (SVM) as an adversary for LWE decision. To deal with data that is not linearly separable, the non-linear RBF kernel was used. A grid search among hyperparameters was applied, leading to an accuracy confidence interval of (0.51623915, 0.51735885) with a significance level of 0.99, which represents an advantage of greater than 1.623%.

The full report is in the report.pdf file, the Python code is in the generation.py file, and the results database is in the results.db file.