Publications
- Gudipaty, Krishna Praneet , W. A. Hanafy, L. Wu, et al., “Practical considerations for failure resilient ml systems at the edge,” in MILCOM 2025 - 2025 IEEE Military Communications Conference (MILCOM) (to appear)
- Gudipaty, Krishna Praneet , W. A. Hanafy, K. Ozkara, et al., “Mel: Multi-level ensemble learning for resource-constrained environments,” arXiv preprint arXiv:2506.20094, 2025
- A. Micciche, Gudipaty, Krishna Praneet , and S. Krastanov, “Quantum ldpc error correcting codes for use on 1d quantum dot arrays,” in APS March Meeting Abstracts, vol. 2024, S46–010
Research
Krishna's research focuses on building resilient and resource-efficient AI model inference systems for edge and IoT environments. While edge AI offers strict low-latency guarantees through parameter efficient models and edge accelerators (e.g., TPUs, GPUs), these devices remain resource-constrained (memory, storage, power) and vulnerable to adversaries. To address these challenges, his work is structured around three pillars:
- VISIBILITY: Highlighting practical considerations and estimating inefficiencies in current edge systems for providing actionable insights.
- ACTIONABILITY: Addressing core challenges through contributions in model serving systems/frameworks to enhance performance and scalability.
- MAINTAINABILITY: Incentivising and ensuring availability and robustness for these solutions to be applicable to diverse operating conditions.
A case study on the practical considerations for edge model serving systems on SLOs.
A novel Machine Learning framework for resilient AI/ML inference in heterogeneous edge environments.
A software library of fast and efficient implementations of Quantum LDPC (qLDPC) decoders in Julia.
