nvflare.app_common.statistics.min_max_cleanser module

class AddNoiseToMinMax(min_noise_level: float, max_noise_level: float)

Bases: FLComponent, StatisticsPrivacyCleanser

min_noise_level: minimum noise – used to protect min/max values before sending to server max_noise_level: maximum noise – used to protect min/max values before sending to server

min/max random is used to generate random noise between (min_random and max_random). for example, the random noise is to be within (0.1 and 0.3),i.e. 10% to 30% level. These noise will make local min values smaller than the true local min values, and max values larger than the true local max values. As result, the estimate global max and min values (i.e. with noise) are still bound the true global min/max values, in such that

est. global min value <

true global min value <

client’s min value <

client’s max value <

true global max <

est. global max value

apply(statistics: dict, client_name: str) → Tuple[dict, bool]

generate_noise(statistics: dict, statistic) → dict

validate_inputs()