Today's Task：How to find the components of this model

1 選取特征Feature

1.1 How to proceed?

Our problem is one of dimensionality!

For instance, in the case of the movie we showed the retina, we can define a movie in terms of the intensity of three colors in every pixel in one megapixel image.

1.2 Dimensionality reduction

Start with a very high dimensional description(e.g. an image or a time-varying waveform) and pick out a small set of relevant dimensions.

s(t)----dicretize------>s(k)

?

采樣系統(tǒng)對(duì)于不同的刺激的響應(yīng)，我們可以識(shí)別是什么輸入觸發(fā)響應(yīng)。

1.3 What is the right stimulus to use?

?We want to sample the responses of the system to a variety of stimuli so we can characterize what it is about the input that triggers responses.

One common and useful method is to use Gaussian?white noise.

1.4 Determining multiple features from white noise

這里只要了解spike-trigger 平均值這個(gè)概念，就是把數(shù)據(jù)整合起來，得到一條類似于高斯函數(shù)的曲線，峰值對(duì)應(yīng)的橫坐標(biāo)表示的值。

1.5 Reverse correlation: the spike-triggered average 反相關(guān)系數(shù)：尖峰平均值

橫坐標(biāo)上表示的是一個(gè)響應(yīng)spike，我們提取從開始刺激到產(chǎn)生響應(yīng)的時(shí)間，取它們的平均值，得到一條噪聲較少的曲線。

?

?

?每列值都是一個(gè)圖像，這里包括時(shí)間維度和空間維度。

1.6 Linear filtering

Stimulus feature f is a vector in a high-dimensional stimulus space

?線性過濾器，相當(dāng)于卷積，也相當(dāng)于投影。我們有一個(gè)刺激s（方向與t3相同），投影到f上s·f（？？？）

2 Determining the nonlinear input/output function

The input/output function is:

This can be found from data using Bayes' rule:

?

?

?P（s1）是一個(gè)高斯曲線

Nonlinear input/output function

?

?

2.1 Linear/nonlinear models

?

3 High-dimensional feature selection

?Less basc coding models

有多個(gè)過濾，選擇多個(gè)特征。core detector neuron 每個(gè)對(duì)不同的頻率的過濾。

Determining multiple feature from white noise

How could we find features？

3.1 Principal component analysis

PCA's job is to find low dimengtional structure of a cloud of points.

compression.

PCA: eigenfaces

common stracture, may be restructive by little number of photos

PCA: spike sorting

PCA gives us a method to:

1. Find a representation of our data which has lower dimensionality, giving us a computationallyeasier problem to work with.

2. Find the vectors along which the variation of our data is maximal in our feature space.

4 Finding interesting features in the retina

right group——on

left group——off

?

?這節(jié)聽得很蒙蔽啊，還是找本教科書看看吧

轉(zhuǎn)載于:https://www.cnblogs.com/uniKino/p/10165705.html

總結(jié)

以上是生活随笔為你收集整理的【计算神经科学冒险者们】2.3 神经编码：特征选择(Neural Encoding：Feature Selection)...的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。