pulsatrix 1.1.0
Loading...
Searching...
No Matches
pulsatrix::AttackResult Struct Reference

One attack's result. More...

#include <adversarial.hpp>

Collaboration diagram for pulsatrix::AttackResult:

Public Attributes

Tensor adversarial
 The adversarial inputs, the input's shape, on the input's device.
 
std::vector< int64_t > clean_predicted
 Each input's predicted class before and after the attack.
 
std::vector< int64_t > adversarial_predicted
 
std::vector< bool > success
 Untargeted: the adversarial input isn't classified as its label. Targeted: it is classified as its target.
 
std::vector< double > clean_loss
 The cross-entropy of each label (the target's, when targeted) before and after.
 
std::vector< double > adversarial_loss
 
std::vector< double > perturbation
 Each perturbation's size in the attack's norm: at most epsilon, up to rounding.
 
double success_rate = 0.0
 The fraction of inputs the attack succeeded on.
 
double robust_accuracy = 0.0
 Untargeted: the fraction classified correctly after the attack.
 
double clean_accuracy = 0.0
 Untargeted: the fraction classified correctly before it.
 
bool targeted = false
 

Detailed Description

One attack's result.

Member Data Documentation

◆ adversarial

Tensor pulsatrix::AttackResult::adversarial

The adversarial inputs, the input's shape, on the input's device.

◆ adversarial_loss

std::vector<double> pulsatrix::AttackResult::adversarial_loss

◆ adversarial_predicted

std::vector<int64_t> pulsatrix::AttackResult::adversarial_predicted

◆ clean_accuracy

double pulsatrix::AttackResult::clean_accuracy = 0.0

Untargeted: the fraction classified correctly before it.

◆ clean_loss

std::vector<double> pulsatrix::AttackResult::clean_loss

The cross-entropy of each label (the target's, when targeted) before and after.

◆ clean_predicted

std::vector<int64_t> pulsatrix::AttackResult::clean_predicted

Each input's predicted class before and after the attack.

◆ perturbation

std::vector<double> pulsatrix::AttackResult::perturbation

Each perturbation's size in the attack's norm: at most epsilon, up to rounding.

◆ robust_accuracy

double pulsatrix::AttackResult::robust_accuracy = 0.0

Untargeted: the fraction classified correctly after the attack.

◆ success

std::vector<bool> pulsatrix::AttackResult::success

Untargeted: the adversarial input isn't classified as its label. Targeted: it is classified as its target.

◆ success_rate

double pulsatrix::AttackResult::success_rate = 0.0

The fraction of inputs the attack succeeded on.

◆ targeted

bool pulsatrix::AttackResult::targeted = false

The documentation for this struct was generated from the following file: