Q: How is Protopia AI’s solution different from data masking solutions in the market?

A: Protopia AI’s solution is fundamentally different from data masking solutions in the market. We are not focused on finding any particular data to mask. Our core explainability technology enables us to mathematically reason about what features in each data record are relevant to a given Machine Learning (ML) task and to what extent. Using this information, we can obfuscate all features (not just unnecessary ones) in accordance to the extent of their importance to the ML task at hand. Furthermore, most data masking solutions are using some form of ML to identify sensitive features such as people’s faces or license plates of cars. In that process of inferring “what sensitive is” those data masking technologies almost certainly expose that sensitive data during their own process as well.

 

Q: I have a lot of data my customers want to access in order to validate their model, but I cannot currently offer them my data because of the sensitive nature of the data, can you help with that?

A: Yes, we can enable you to provide customized versions of your data that do not expose all the information in each data record for customers or 3rd party AI service providers that want to validate their ML models with your data.

 

Q: Is this encryption? Is there a key that’s used for decrypting the garbled data?

A: No, there is not. The obfuscated data is not encrypted for there to exist a key. As such, there is also no decryption. The obfuscation is one-way and irreversible.

 

Q:Do I (the customer) need to send you (Protopia AI) my data to obfuscate?

A: No, we never input any customer’s data. Obfuscation is done within the customer’s own data ingestion pipeline at inference.

 

Q: Do I have to expose my neural network model to you?

A: Protopia’s solution can work in two modes: as a managed service, or within your firm’s own infrastructure. In the latter option, the model doesn’t need to be exposed outside of your business. We will work with you to find which mode is best for your business.

 

Q: What are my storage requirements for this?

A: Masking of unnecessary data is most commonly performed on the path of data on the way from storage into the inference device. Data is masked appropriately as it is being loaded for inference and a new copy of the masked data is not necessary. As such, storage requirements do not increase by using Protopia AI’s solution.

 

Q: How much computational overhead is required?

A: The computational overhead of applying Protopia AI’s obfuscation masks during inference is minimal. This is made possible by our patented technology used in the generation of data obfuscation masks which runs as an optimization stage at the end of training. Since we identify the necessary vs. unnecessary features of each data record in the training optimization phase, we are able to create obfuscation masks that are very low overhead to apply during the inference phase.

 

Q: How do we do this if I don’t want to give you my model?

A: You can run Protopia’s noise discovery engine (which generates the obfuscation masks) on your own on-premise infrastructure.

 

Q: Where can this be deployed?  Server or edge?

A: Our solution is incredibly simple to deploy because it has very low computation overhead. Protopia AI’s core technology enables us to help identify what features in each data record are important to the customer’s neural network and to what extent. This capability is exposed as a last pass optimization of the customer’s training. The output of the noise discovery engine are the obfuscation masks to be used in the inference pass.. This simplicity makes it very flexible where you deploy the obfuscation masks: edge or server-side.

 

Q: Different departments within my organization want to share data to validate models being built, can you help reduce the restrictions around this sort of data sharing?

A: Yes, we can enable you to provide customized versions of your data that do not expose all the information in each data record for customers or 3rd party AI service providers that want to validate their ML models with your data.

 

Q: How do you charge for access to your solution?

A: You can obtain a subscription license to access our solution. It runs as a last phase optimization in their training pipeline and outputs obfuscation masks that are seamlessly deployed in their inference pass.

Menu