Data anonymization can help Telcos to share data to facilitate open innovation. Two big challenges to address while anonymizing Telco data for AI usecases are (a) fool-proof against de-anonymization (b) Not hamper the power (Ex: predictive, Classification) of the AI models. The talk we cover the following:
- (a) State of art of Data Anonymization applied to Telcos, including the research works, projects and specifications.
- (b) What constitutes the sensitive data (names, addresses, telco-specific fields, location-data, etc). in Telco scenarios
- (c) Anonymization categories such as (Suppression, Masking, Pseudonymization, Generalization, Swapping, Perturbation and Synthetic Data Generation).
- (d) Approaches/techniques ranging from classic (ex: K-Anonymity) to use of NLP to GANs (Generative Adversarial Networks).
- (e) Which of the above categories and techniques are applicable to Telco Data, considering the challenges of deanonymization and model-power.
- (f) Demonstration of the developed unified-anonymization tool.
The talk will also include description of the libraries that can be used, demonstration of the techniques, and showcasing of the impact of anonymization on the AI models.
