Hema Venkata Krishna Giri Narra (Krishna)
Email: narrakrishnagiri at gmail.comGoogle Scholar · ResearchGate · LinkedIn · Github
About Me [Curriculum Vitae]
I work as a software engineer at Google. I received my Ph.D. in Computer Engineering from University of Southern California (USC) working under the guidance of Professor Murali Annavaram. My thesis is on building straggler-resilient and private machine learning systems in the cloud. Prior to starting Ph.D, I worked for two years at Intel. I completed my M.S. in Computer Science from USC and my Bachelors and Masters of Technology in Electrical Engineering from Indian Institute of Technology, Madras.
Selected Research Projects
Origami Inference: Private Inference Using Hardware Enclaves
• Designed and implemented Origami framework that partitions DNN models between secure hardware enclaves and unsecure accelerators.• Evaluated the Origami framework on Intel SGX enclaves and demonstrated that up to 15x speedup in inference latency can be achieved compared to running the full model inside the secure SGX enclave.
Collage Inference: Using Coded Redundancy for Lowering Variance of Distributed Image Classification Systems
• Proposed collage-cnn models that perform multi-image classification and used them as low-cost redundancies to mitigate slowdowns in distributed inference systems.• Demonstrated that deploying collage-cnn models can reduce the variance in latency by up to 15x, and the 99-th percentile latency up to 2x compared to alternate approaches.
S2C2: Slack Squeeze Coded Computing for Adaptive Straggler Mitigation
• Proposed S2C2 which exploits the data redundancy available in coded data and elastically distributes work based on speeds predicted from a LSTM model.• S2C2 reduced the total compute latency by up to 39.3% over the conventional coded computation and by up to 19% over fine-grained replication.
GradiVeQ: Vector Quantization for Bandwidth-Efficient Gradient Aggregation in Distributed CNN Training
• Proposed GradiVeQ, a novel gradient compression technique that can significantly reduce the communication load in distributed CNN training.• GradiVeQ reduces the end-to-end training time by 4x over the uncompressed method, and by 1.4x over the 4-bit-QSGD method.
Publications
• IEEE CLOUD 2021 Krishna Giri Narra , Zhifeng Lin, Yongqin Wang, Keshav Balasubramaniam, Murali Annavaram, “Origami Inference: Private Inference Using Hardware Enclaves”. Published at IEEE CLOUD 2021.
• ICDCS 2020 Krishna Giri Narra , Zhifeng Lin, Ganesh Ananthanarayanan, Salman Avestimehr, Murali Annavaram, “Collage Inference: Using Coded Redundancy for Lowering Latency Variation in Distributed Image Classification Systems”. Published at ICDCS 2020.
• SC 2019 Krishna Giri Narra*, Zhifeng Lin*, Mehrdad Kiamari, Salman Avestimehr, Murali Annavaram, “Slack Squeeze Coded Computing(S2C2) for Adaptive Straggler Mitigation”. Published at the Supercomputing Conference (SC 2019). [* equal contribution] Best Paper Finalist. Best Student Paper Finalist.
• ICML 2019 Krishna Giri Narra , Zhifeng Lin, Ganesh Ananthanarayanan, Salman Avestimehr, Murali Annavaram, “Collage Inference: Achieving low tail latency during distributed image classification using coded redundancy models”. CodML workshop at ICML 2019.
• ARXIV 2019 Zhifeng Lin, Krishna Giri Narra , Mingchao Yu, Salman Avestimehr, Murali Annavaram, “Train Where the Data is: A Case for Bandwidth Efficient Coded Training. ARXIV, abs/1910.10283, 2019.
• NeurIPS 2018 Mingchao Yu*, Zhifeng Lin*, Krishna Giri Narra , Songze Li, Youjie Li, Nam Sung Kim, Alexander G. Schwing, Murali Annavaram, Salman Avestimehr, “GradiVeQ: Vector Quantization for Bandwidth-Efficient Gradient Aggregation in Distributed CNN Training”. Published at NeurIPS 2018. [* equal contribution]
• MICRO 2017 Gunjae Koo*, Kiran Matam*, Te I, Krishna Giri Narra , Jing Li, Hung-Wei Tseng, Steven Swanson, Murali Annavaram, “Summarizer: Trading communication with computing near storage”. Published at IEEE International Symposium on Microarchitecture (MICRO 2017). [* equal contribution]
• Soft Computing 2015 L.Srivani, N.H.V.Krishna Giri , Shankar Ganesh, V.Kamakoti, “Generating Synthetic Benchmark Circuits for Accelerated Life Testing of Field Programmable Gate Arrays using Genetic Algorithm and Particle Swarm Optimization”, Applied Soft Computing, Volume 27, February 2015.
Awards and Honors
Best Dissertation award in Computer Engineering at USC (2021).
S2C2 paper is selected as a best paper finalist at the Supercomputing Conference (SC'19).
S2C2 paper is selected as a best student paper finalist at the Supercomputing Conference (SC'19).