Add split learning sim_lia method (IEEE2024)

[bear-zd]

Type of change

• [x] Add new papers (Please tell us why you think this paper is awesome!)
• [ ] Fix the category of an existing paper/papers (Please tell us the reasons)
• [ ] Add a new tool/primitive/application with a new markdown page (Thank you! Also, please tell us more about this awesome thing!)

Description Here's the English translation of the PR content:

Added the sim_lia algorithm, based on the 2024 IEEE paper "Similarity-based Label Inference Attack against Training and Inference of Split Learning." https://ieeexplore.ieee.org/document/10411061

The method is based on cosine distance and Euclidean distance of smashed data and gradients. The attack can be distance-based or clustering-based, with the following options:

availabel_data_type = ["feature", "grad"]
availabel_attack_method = ["k-means", "distance"]
availabel_distance_metric = ["euclidean", "cosine"]
all_availabel_options = {
    "k-means": ["feature", "grad"],
    "distance": {
        "cosine": ["feature", "grad"],
        "euclidean": ["feature", "grad"],
    },
}

Due to its generality, it can be used in a wide range of scenarios. Currently, its attack performance on autoattack is as follows:

App	Epoch	Acc	Attack Acc
ResNet20 Cifar10	2	43.336%	29.636%
ResNet18 Cifar10	10	88.077%	99.122%
CNN Cifar10	10	86.467%	77.458%
VGG16 Cifar10	10	82.748%	50.184%
ResNet18 MNIST	5	98.834%	82.813%
VGG16 MNIST	1	79.949%	55.935%

In addition, fixed the incorrect import issues in benchmark_examples caused by the previous SecretFlow refactoring.

[github-actions[bot]]

Stale pull request message. Please comment to remove stale tag. Otherwise this pr will be closed soon.

[bear-zd]

.