Abstract: An algorithm is trained on a dataset to facilitate dynamic data exfiltration protection in a zero-trust environment. An inversion threat model using the original training dataset (a ‘gold standard’ inversion model) may also be generated. This inversion model can be characterized to determine its performance/accuracy of properly identifying a given input as being within the original training dataset or not (a data exfiltration event). It is possible to reduce this risk of data exfiltration to a desired level, without unduly impacting the algorithm's performance using the inversion model for the generation of noise that is targeted (as opposed to Gaussian noise). Noise added to the original training dataset causes the inversion model to perform poorer (meaning data steward data is more secure) but has a corresponding impact on the algorithm accuracy and performance.

Abstract: Systems and methods for the quantification of sample set quality is provided. In some embodiments, a sample dataset and a sample vector set are received. A rule-based screening of the sample dataset is applied to generate a heuristic quality score. Additionally, a sample vector set is generated from the sample dataset. The difference between the sample vector set and the example vector set is calculated to generate a degree of difference quality score. The heuristic quality score and the degree of difference quality score are normalized and then combined into a quality metric. Calculating the difference between the sample vector set and the example vector set is by framing the distance as a p-value in a hypothesis test, compared against a threshold.

Abstract: Systems and methods for recommendation of cohort sample sets is provided. In some embodiments, a set of dataset requirements is received as a required vector set. The historical vector sets are queried. Each vector set corresponds to a known dataset. The difference between the required vector set and each of the historical vector sets is calculated by framing the distance as a p-value in a hypothesis test, compared against a threshold. The historical vector set with the least difference to the required vector set is identified. The least difference is calculated as a count of differing classes or as a numerically weighted summation of differing classes.

Abstract: Systems and methods for the deployment and operation of an algorithm in a zero-trust environment are provided. In some embodiments, an algorithm is encrypted by an algorithm developer within a zero-trust computing node, using a public key. This generates a payload that is transferred to a core management system which in turn distributes the payload to one or more sequestered computing nodes located within the infrastructure of one or more data stewards. The sequestered computing nodes are designed to preserve privacy of data assets and the algorithm. Next the payloads are decrypted, using a private key, within the sequestered computing nodes. This yields the algorithm that can be run against the data assets of the data steward. A report is generated that can be shared with the appropriate parties.

Abstract: Systems and methods for the processing of diverse datasets by divergent algorithms within a zero-trust environment is provided. In some embodiments, a single data steward may receive multiple algorithms in a zero-trust environment. Alternatively, algorithm output may be obfuscated for sharing with the algorithm developer for validation, or to compare against the output of a different data steward's processed protected information, for example PHI. In such situations the hashed identifying information may be matched using AI models. In yet other embodiments, the output of one data steward's protected information may be provided in a zero-trust manner to the sequestered enclave of a second data steward in order to impact the processing of this second data steward's protected information by a second algorithm.