Softmax 활성함수의 특정한 그래프가 없는 이유

1. Softmax 활성함수의 특정한 그래프가 없는 이유

The softmax activation function does have a specific mathematical equation, but it does not have a specific graph like the sigmoid or ReLU functions. This is because the softmax function takes in a vector of arbitrary length (corresponding to the logits for each class), and outputs a vector of probabilities that sum up to one. The shape of the softmax output depends on the shape of the input vector.

The sigmoid and ReLU functions, on the other hand, are scalar functions that take in a single input value and output a single value. This makes it possible to plot the function over a range of input values to visualize its behavior.

It's worth noting that although the softmax function does not have a specific graph, it is still a well-defined mathematical function with specific properties. In particular, the softmax function is monotonically increasing, which means that as the logits for a given class increase, the probability of that class also increases. Additionally, the softmax function satisfies the property that the sum of the probabilities for all classes is equal to one, which is a desirable property for a classification algorithm.

 

In summary, the softmax function does not have a specific graph like the sigmoid or ReLU functions because it takes in a vector of logits as input and outputs a vector of probabilities. However, it is still a well-defined mathematical function with specific properties that make it useful for multi-class classification problems.

 

시그모이드, relu 활성함수는 스칼라로 구성된 입력값을 받아 출력하기 때문에 정해진 그래프가 있는 반면,

소프트맥스 활성함수는 숫자로 구성된 벡터를 입력받아 확률 벡터로 출력하기 때문에 정해진 그래프가 없이, 입력값에 따라 그래프가 달라진다.

 

 

2. Input의 종류 : Scalar와 vector의 차이점

A scalar is a single numeric value, while a vector is a collection of multiple numeric values arranged in a specific order. Scalars are often used to represent simple, one-dimensional data, while vectors are used to represent more complex, multi-dimensional data.

 

Another difference between scalars and vectors is the way they are processed by machine learning algorithms. Scalars are typically passed directly into the model as input, while vectors may need to be preprocessed and transformed into a different format (such as a matrix) before they can be used as input.

 

스칼라는 단순한 일차원적인 데이터이다. 반면 벡터는 다차원적인 데이터이며, input 데이터로 활용하기 위해서는 전처리를 해줘야 한다.