Portrait of Aseem Rastogi

Aseem Rastogi

Senior Principal Researcher

News & features

EzPC provides secure AI model validation. In the diagram poses the following question: Is the accuracy of the AI model on the test dataset greater than 70%? First, an AI vendor provides model weights, and a modular compiler takes as input from the model weights the AI model structure written in ONNX code for ML inference. From this, it automatically generates MPC protocol code, which is then compiled into various MPC protocols. Additionally, a suite of highly performant cryptographic protocols securely compute complex ML functions on an organization’s test dataset. The MPC protocol outputs random bits, keeping the data from both parties secure.
Microsoft Research Blog

| Nishanth Chandran, Divya Gupta, Aseem Rastogi, and Rahul Sharma

From manufacturing and logistics to agriculture and transportation, the expansion of artificial intelligence (AI) in the last decade has revolutionized a multitude of industries—examples include enhancing predictive analytics on the manufacturing floor and making microclimate predictions so that farmers can…

An animation showing an example of a high-level language message format specified by EverParse. From the message, two arrows labeled “EverParse” point to a rectangle labeled “formal specification” and a rectangle labeled “low-level implementation,” respectively, inside a larger rectangle labeled “F* code.” The figure represents EverParse’s ability to automatically generate safe, correct, and fast F* parsing code. “Correctness” is defined as “Correctness: ​parse (serialize msg) = msg​” and “valid msg ==> serialize (parse msg) = msg​.” The F* logo appears with the description that F* is a type theory–based programming language and proof assistant that can prove theorems about programs​. From the “F* code” rectangle, arrows point from the “low-level implementation” rectangle and a rectangle labeled “verified libraries for combinators” to a rectangle labeled “Safe high-performance C code.”
Microsoft Research Blog

| Tahina Ramananandro, Aseem Rastogi, and Nikhil Swamy

EverParse is a framework for generating provably secure parsers and formatters used to improve the security of critical code bases at Microsoft. EverParse is developed as part of Project Everest, a collaboration between Microsoft Research labs in Redmond, Washington; India;…

In the news | CyLab

In security, almost nothing is guaranteed. It's impossible to test the infinite ways a criminal hacker may penetrate a proverbial firewall. But what if, by the laws of mathematics, something could be proven to be secure without running an infinite…