News

https://sit.iitd.ac.in/show-news/56

AI-based software analysis for safety, security, and performance optimizations (Sorav Bansal)

1. Building an AI-guided code analyzer for safety and security to analyze and understand code to identify safety and security defects in critical software.  This project involves coupling AI models with dataflow analyses and formal methods.  Even if you do not have prior exposure to these areas (e.g., you are a non-CS student), that is okay.  But you need to be able to learn these areas through courses and online materials.
2. Building an AI-guided code optimizer to analyze and understand code to identify performance optimization opportunities in software. This work involves techniques from high-level optimizations to assembly-level micro-architectural optimizations.  This project involves coupling AI models with knowledge of computer architecture and assembly programming.  Even if you do not have prior exposure to these areas (e.g., you are a non-CS student), that is okay.  But you need to be able to learn these areas through courses and online materials.
3. Building an AI-guided transpiler for quantum computing to convert high-level logic to quantum circuits. This project involves techniques from quantum computing literature, compilers, and theoretical foundations of computer science.  Even if you do not have prior exposure to these areas (e.g., you are a non-CS student), that is okay.  But you need to be able to learn these areas through courses and online materials.

 

Verifying distributed storage systems (SVS and Abhilash)

Distributed storage systems form the backbone of modern internet services. However, maintaining correctness and extracting good performance while ensuring resilience to real-world disruptions like server failures, network faults, and clock drifts, remains challenging. Therefore, the cloud industry has adopted various verification tools like model checkers, fault injection frameworks, and proof checkers to test and verify their protocols and implementations.

In this project, we aim to improve upon the existing verification foundation. Some ideas include scaling up verification tools by improving core algorithms, generalizing them to include new types of failures like gray failures and metastable failures, and using LLM-based proof agents to reduce the verification burden

Necessary background: Reasoning about correctness and performance of systems

Optional background: distributed systems, model checking

Predoc: Yes

 

Tamper-evident reproducible executions (Riju and Abhilash)

Reproducibility is a cornerstone of scientific computing, yet many modern workloads depend on remote services whose behavior cannot be reproduced or independently verified. Large language model (LLM) APIs are a prominent example: identical prompts issued at different times may produce different outputs, and the underlying model or service implementation may change without notice. As a result, experiments that depend on such services often cannot be faithfully reproduced and verified.

In this project, we build a container abstraction that supports tamper-evident reproducible executions where a verifier can independently reproduce and verify experiments claimed by a prover.

Necessary background: Operating systems

Optional background: seccomp, confidential computing

Predoc: Yes
 

Privacy Enhancing Technologies, Network Security, Offensive/Defensive Security (Piyush)

There are various projects available on assessing the security & privacy posture of computing systems as well as developing privacy preserving systems. While the scope of the work is broad, some topics include (1) analyzing the VPN, proxy, Tor and other advanced anonymity systems such as mixnets (2) studying security and privacy vulnerabilities in applications from a regional lens (3) characterizing and developing solutions for the online scam/fraud ecosystem. 

Essential Background: Applied Computer Networks, good coding and scripting

Optional Background: Knowledge about security and privacy topics, Operating Systems 

Website: https://www.piyushs.in/

Faculty: Piyush Kumar Sharma (piyush@iitd.ac.in)

Predoc: Yes