# 2020/10/20 group meeting

---

1. Past work
2. Noise re:result
3. Decision make;關係 of recurrent & noise
4. probability of Decision exchange
5. Inverse spikes intervel

---

## past work

1. EMA & firingrate
2. noise test
`noisevalue = noise * (rand() / (RAND_MAX + 1.0)) - noise / 2`

---

## noise re:result
1. 加了種子進去 like this

#include <iostream>
#include <stdio.h>
#include <random>

using namespace std;

int main(void)
 {
     unsigned seed;
     seed = (unsigned)time(NULL);
     srand(seed);
     int i;
     float noisevalue;
     FILE*fp;
     float noise;
     noise = 1.0;
     fp = fopen("noisetest.txt","w");
     for(i = 0; i < 1000000; i++)
 {
     noisevalue = noise * float(rand() / (RAND_MAX + 1.0)) + noise / 2;
     fprintf(fp,"%f\n",noisevalue);
 }
     fclose(fp);
     return 0;
 }

2. result
10W
100W

---

## Decision make;關係 of recurrent & noise
要在關閉input後保留current like this
放大decision部份後長這樣，約300units -> 1/5sec

if noise upup -> recurrent have to upup too
如此才能維持那一小段decision time

---

## probability of Decision exchange

| time | neuron | inputstr |
| ---- | ------ | -------- |
| 3000 | 0 | 6 |
| 3000 | 1 | 5 |
| next | next | next |
| var | 0 | 0 |
| var | 1 | 5 |
| next | next | next |
| 3000 | 0 | 6 |
| 3000 | 1 | 5 |

every test run 1000 times,like this
and this is their probability

| var | exchange probability |
| ---- | -------------------- |
| 300 | 0 |
| 400 | 0 |
| 500 | .2 |
| 600 | .48 |
| 700 | .68 |
| 800 | .8 |
| 900 | .83 |
| 1000 | .85 |
| 1100 | .88 |
| 1200 | .88 |
| 1300 | 1 |

---

## inverse spikes interval

不知道算法對不對
參照上課講義

    Epu = (external_current / -_g) + _rest;
    Z = (_threshold - _reset) / (Epu - _threshold);
    ISI = (1 / ((1 / -_g) * log(1 + Z)));