TAPAS

# TAPAS: Datasets for Learning the Learning with Errors Problem

## About this Data
AI-powered attacks on Learning with Errors (LWE)—an important hard math problem in post-quantum cryptography—rival or outperform "classical" attacks on LWE under certain parameter settings. Despite the promise of this approach, a dearth of accessible data limits AI practitioners' ability to study and improve these attacks. Creating LWE data for AI model training is time- and compute-intensive and requires significant domain expertise. To fill this gap and accelerate AI research on LWE attacks, we propose the TAPAS datasets, a **t**oolkit for **a**nalysis of **p**ost-quantum cryptography using **A**I **s**ystems. These datasets cover several LWE settings and can be used off-the-shelf by AI practitioners to prototype new approaches to cracking LWE.

The table below gives an overview of the datasets provided in this work:
| n | log q | omega | rho | # samples |
|--------|-----------|----------|--------|------------|
| 256 | 20 | 10 | 0.4284 | 400M |
| 512 | 12 | 10 | 0.9036 | 40M |
| 512 | 28 | 10 | 0.6740 | 40M |
| 512 | 41 | 10 | 0.3992 | 40M |
| 1024 | 26 | 10 | 0.8600 | 40M |

## Usage

These datasets are intended to be used in conjunction with the code at: https://github.com/facebookresearch/LWE-benchmarking

Download and unzip the .tar.gz files into a directory with enough storage. For the datasets split into different chunks, concatenate all the files into one data.prefix file after unzipping.

Then, follow the instructions in this [README](https://github.com/facebookresearch/LWE-benchmarking/blob/main/README.md) to generate the full sets of LWE pairs and train AI models on this data.

Due to storage constraints, we only provide 40M of the n=256 data here on huggingface. The rest can be found at this directory (append filenames ranging from chunk_ab.tar.gz to chunk_aj.tar.gz to download): http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/

Here are the exact links to each remaining section of the n=256 data (each link has 40M examples):
[1](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ab.tar.gz)
[2](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ac.tar.gz)
[3](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ad.tar.gz)
[4](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ae.tar.gz)
[5](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_af.tar.gz)
[6](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ag.tar.gz)
[7](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ah.tar.gz)
[8](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ai.tar.gz)
[9](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_aj.tar.gz)

# TAPAS: 用于学习容错学习问题的数据集

## 关于本数据集
基于人工智能的容错学习（Learning with Errors, LWE）攻击——后量子密码学中的一类核心困难数学问题——在部分参数配置场景下，其性能已可媲美甚至超越传统的LWE经典攻击方法。尽管该技术路径前景可观，但可用数据集的匮乏极大限制了AI从业者研究与优化此类攻击的能力。为LWE模型训练生成专属数据集不仅耗时耗力，还需要深厚的领域专业知识储备。为填补这一研究空白，加速LWE攻击相关的AI研究进程，我们推出了TAPAS数据集，即借助人工智能系统开展后量子密码学分析（**t**oolkit for **a**nalysis of **p**ost-quantum cryptography using **A**I **s**ystems）的工具集。本数据集覆盖多种LWE参数配置场景，AI从业者可直接开箱即用，以此为基础原型化开发破解LWE的全新方案。

下表为本工作提供的数据集概览：
| 参数n | log q | omega | rho | 样本总量 |
|--------|-----------|----------|--------|------------|
| 256 | 20 | 10 | 0.4284 | 4亿 |
| 512 | 12 | 10 | 0.9036 | 4000万 |
| 512 | 28 | 10 | 0.6740 | 4000万 |
| 512 | 41 | 10 | 0.3992 | 4000万 |
| 1024 | 26 | 10 | 0.8600 | 4000万 |

## 使用方式
本数据集需配合下述代码仓库使用：https://github.com/facebookresearch/LWE-benchmarking
请将下载的.tar.gz文件解压至具备足够存储空间的目录中。若数据集被拆分为多个分块，解压完成后需将所有文件合并为一个名为data.prefix的文件。
随后，请遵循此[README文档](https://github.com/facebookresearch/LWE-benchmarking/blob/main/README.md)中的说明，生成完整的LWE样本对集，并基于此数据训练AI模型。

受限于存储空间，我们仅在Hugging Face平台上提供了n=256配置下的4000万条数据。其余数据可通过以下目录下载（需将文件名范围从chunk_ab.tar.gz至chunk_aj.tar.gz追加至目录路径后）：http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/
以下为n=256数据集各剩余分块的精准下载链接（每个链接对应4000万条样本）：
[1](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ab.tar.gz)
[2](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ac.tar.gz)
[3](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ad.tar.gz)
[4](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ae.tar.gz)
[5](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_af.tar.gz)
[6](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ag.tar.gz)
[7](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ah.tar.gz)
[8](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_ai.tar.gz)
[9](http://dl.fbaipublicfiles.com/large_objects/lwe-benchmarking/n256_logq20/chunk_aj.tar.gz)