VeritasJoker / Crisp

Simple Neural Network

Definitions

• N: number of neurons in the network
• Categorization of a vector:
  • if timestep given, calculates all the states of the vector in given timestep
  • if not, calculates the final state or end cycle of the vector
• Vectors:
  • Number formt: 3
  • Bitset format: 0000000000000000000000011
  • Vector format: [23,24]
• Cycle:
  • id: The biggest vector (number format) in the cycle
  • population: The number of vectors that is categorized to the cycle
  • length: number of unqiue vectors in the cycle

Models Overview

Core Model

• 10_mod1.cpp
• 3.cpp

Variation Models

• Permanently Excite or Inhibit a Neuron

  • 10_mod5.cpp
  • 10_mod3.cpp

• Random Walk

  • 9.cpp
  • 8.cpp
  • 7_mod2.cpp (not used)
  • 6.cpp (not used)
  • 5.cpp

• Detailed

  • 7_mod1.cpp (categorization)
  • 4+matrix.cpp (distance)
  • 4 (distance)

Helpers

• 11.py
• 10_mod6.cpp
• 10_mod4.cpp
• 10_mod2.cpp
• matrix.cpp
• results.py

Old Models

• 2.cpp
• 2_1.py
• 2.py
• 1.cpp
• 1.py

Models Detailed

Old Models

• 1.py

  • inputs: None
  • parameters: N, probability of connection, timestep
  • outputs: a graph that shows the categorization of a random vector in given timestep

• 2.py

  • new parameters: number of trials
  • new: new random vector for each trial, cleaner code for each trial
  • new outputs: categorization of unique, transient, and repeating vectors for all vectors in all trials

• 2_1.py

  • new: cleaner code for each trial

• 1.cpp

  • new: C++ version of 1.py
  • new outputs: (to txt) [timesteps, on neuron] coordinates for graphing

• 2.cpp

  • new: clearner code

Models

• 3.cpp

  • new inputs: (from txt) adjacency matrix
  • new: categorizes all possible random vector given a random matrix
  • new outputs: id and population for all cycles

• 4.cpp

  • new: calculates the distance of each vector to its cycle
  • unique outputs: (to txt 1) id and population for all cycles (to txt 2) distance of each vector to its cycle

• 4+matrix.cpp

  • unique: combines 4.cpp and matrix.cpp

• 5.cpp

  • unique: given a set matrix, apply noise with probability to one trial given a random starting vector
  • unique outputs: transition matrix between cycles

• 6.cpp (not used)

  • unique: noise is randomized

• 7_mod1.cpp

  • new outputs: (to txt 1) id and population for all cycles (to txt 2) categorization of all vectors

• 7_mod2.cpp (not used)

  • unique inputs: (from txt 1) id and population for all cycles (from txt 2) categorization of all vectors
  • unique: apply noise (excite or inhibit a neuron) after reaching an end cycle

• 8.cpp

  • unique inputs: (from txt 1) id and population for all cycles (from txt 2) categorization of all vectors
  • unique: cleaner and faster code for 5.cpp, using bitsets

• 9.cpp

  • unique: noise only happens after reaching an end cycle

• 10.mod1.cpp

  • same as 3.cpp

• 10_mod2.cpp

  • new inputs: (from txt 2) results from 10.mod1
  • new: get all vectors in stable cycle and get common on neurons
  • new outputs: on neurons that are common in all vectors of a stable cycle for all stable cycles

• 10.mod3.cpp

  • new parameters: fixed neuron
  • new: same as 10.mod1 besides an always excited neuron

• 10.mod4.cpp

  • new: same as 10.mod2 besides taking the always excited neuron into consideration

• 10.mod5.cpp

  • new parameters: inhib neuron
  • new: same as 10.mod1 besides an always inhibited neuron

• 10.mod6.cpp

  • new: same as 10.mod2 besides taking the always inhibited neuron into consideration

• 11.py

  • new: explores the characteristics of the network graph

Helper Files

• results.py

  • inputs: (from txt) [timesteps, on neuron] coordinates for graphing
  • outputs: a graph that shows the desired end cycle of a random vector

• matrix.cpp

  • outputs: (to txt) randomly generated adjacency matrix