标签人工智能下的文章 - 白帽Wiki

2023-07-19 huoji 3 条评论

[2023]LLAMA是否会引发下一代安全领域AI机器学习革命?

2023-03-12 huoji 1 条评论

总所周知,所谓机器学习就是一个解方程给x求y的过程罢了.
这是一篇入门的算法代码,又叫做jaya算法(不是java!!!),网上没什么资料,以下是自己python写的.
```python
import math
import random
import matplotlib.pyplot as plt
import numpy as np
g_group_size = 5000
g_var_weight = 2
g_max_iteration = 100
g_array_f = [0 for i in range(0, g_group_size)]
g_array_best = [0 for i in range(0, g_group_size)]
g_array_worst = [0 for i in range(0, g_group_size)]
g_array_pop_new = [[0 for i in range(0, g_var_weight)]
                   for i in range(0, g_group_size)]
#g_number_a = -2 * math.pi
#g_number_b = 2 * math.pi
#g_number_a = -5.12
#g_number_b = 5.12
g_number_a = -100
g_number_b = 100
g_result = []
array_pop = [[0 for i in range(0, g_var_weight)]
             for i in range(0, g_group_size)]

def fn_objective(pArray):
    result = 0
    for i in range(0, g_var_weight):
        # result += pArray[i] * pArray[i]
        # result += math.sin(pArray[i]) * math.pow(math.e, math.pow(1 - math.cos(result), 2)) + math.cos(result) * math.pow(math.e, math.pow(1 - math.sin(pArray[i]), 2)) + math.pow(pArray[i] - result, 2)
        # result += -(1 + math.cos(12 * math.sqrt(math.pow(pArray[i], 2) + math.pow(result, 2)))) / (0.5 * (math.pow(pArray[i], 2) + math.pow(result, 2)) + 2)
        result += math.pow(pArray[i], 2) + 2 * math.pow(result,
                                                        2) - 0.3 * math.cos(3 * math.pi * pArray[i]) - 0.4 * math.cos(4 * math.pi * result) + 0.7
    return result

def fn_clean_array(pArray):
    for i in range(0, g_group_size):
        for z in range(0, g_var_weight):
            if pArray[i][z] < g_number_a:
                pArray[i][z] = g_number_a
            if pArray[i][z] > g_number_b:
                pArray[i][z] = g_number_b
    return pArray

for i in range(0, len(array_pop)):
    for z in range(0, len(array_pop[i])):
        array_pop[i][z] = random.uniform(g_number_a, g_number_b)
    g_array_f[i] = fn_objective(array_pop[i])

print("array_pop: ")
print(array_pop)
print("g_array_f: ")
print(g_array_f)

iter_num = 1
min_index = 0
max_index = 0
while iter_num <= g_max_iteration:
    min_value = min(g_array_f)
    min_index = g_array_f.index(min_value) - 1
    g_array_best = array_pop[min_index]

max_value = max(g_array_f)
    max_index = g_array_f.index(max_value) - 1
    g_array_worst = array_pop[max_index]
    for i in range(0, g_group_size):
        for z in range(0, g_var_weight):
            wrost_number = g_array_worst[z] - math.fabs(array_pop[i][z])
            best_number = g_array_best[z] - math.fabs(array_pop[i][z])
            g_array_pop_new[i][z] = array_pop[i][z] + \
                random.random() * best_number - wrost_number
    g_array_pop_new = fn_clean_array(g_array_pop_new)
    f_new = 0
    for i in range(0, len(g_array_pop_new)):
        f_new = fn_objective(g_array_pop_new[i])
        if f_new < g_array_f[i]:
            g_array_f[i] = f_new
            for z in range(g_var_weight):
                array_pop[i][z] = g_array_pop_new[i][z]
    min_value = min(g_array_f)
    min_index = g_array_f.index(min_value)
    g_array_best = array_pop[min_index]
    print("min number {} ".format(g_array_f[min_index]))
    print("x=")
    print(g_array_best)
    #g_result.insert(0, g_array_f[min_index])
    g_result.append(g_array_f[min_index])
    iter_num += 1
x = np.arange(0, len(g_result))
y = np.array(g_result)
plt.title("jaya")
plt.xlabel("min number")
plt.ylabel("y")
plt.plot(x, y)
plt.show()

```
请注意有一些公式需要很大的族群范围才能收敛
论文里面的收敛图:
![](https://key08.com/usr/uploads/2021/05/2290756614.png)
这边的:
![](https://key08.com/usr/uploads/2021/05/3846208643.png)

阅读全文

2021-05-24 huoji 0 条评论

python 一线开发

[2021]一些常见的机器视觉用到的神经网络结构(keras)

收藏一些常见的机器视觉用到的神经网络结构,以后不用自己构造网络了!
这些都咋用呢? 我已经写好祖传代码了:
```python
def build_save_model(train_data, test_data):
    # 构建模型 CBAPD
    network_model = VGG16()

network_model.compile(optimizer='adam',
                          loss=tensorflow.keras.losses.SparseCategoricalCrossentropy(
                              from_logits=False),
                          metrics=['accuracy'])
    #cp_callback = tensorflow.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,save_weights_only=True,save_best_only=True)
    # 如果要保存点自己在fit加这个参数
    # callbacks=[cp_callback]
    #checkpoint_save_path = "./checkpoint.ckpt"
    # if os.path.exists(checkpoint_save_path + '.index'):
    #    print('-------------load the model-----------------')
    #    network_model.load_weights(checkpoint_save_path)
    network_model.fit(train_data, epochs=100,
                      validation_data=test_data, batch_size=64)
    network_model.summary()
    network_model.evaluate(test_data)
    tensorflow.saved_model.save(
        network_model, '\\model\\')
```
VGG网络:

阅读全文

2021-02-24 huoji 1 条评论

白帽Wiki

一只鸭子

白帽Wiki - 一个简单的wiki

[2023]现代AI杀毒引擎原理+部分代码

[2023]迅速上手meta的llama模型

[2023]LLAMA是否会引发下一代安全领域AI机器学习革命?

[2022] 基于NLP的威胁检出引擎

[2021]从0开始的tensorflow2.0 (二)

[2021]手撸机器学习入门之jaya算法

[2021]一些常见的机器视觉用到的神经网络结构(keras)