Nitish Shirish Keskar
Papers
1
Total Citations
111
H-Index
1
About
Nitish Shirish Keskar is a leading researcher in optimization and machine learning, with a focus on distributed systems and large-scale model training. His work bridges the critical gap between communication efficiency and computational performance in distributed optimization—a cornerstone challenge for modern AI. His highly cited 2018 paper, “Balancing Communication and Computation in Distributed Optimization” (111 citations), provides foundational insights into designing algorithms that minimize overhead while maintaining accuracy, directly impacting fields like robotics, sensor networks, and deep learning. Keskar is also widely recognized for his contributions to understanding generalization in neural networks, including the role of sharp vs. flat minima—work that has influenced how practitioners train robust models. Beyond these theoretical advances, he has made practical contributions to industry-scale machine learning, notably through his work on large-batch training and optimization at companies like Salesforce and NVIDIA. His research is distinguished by its dual emphasis on rigorous mathematical analysis and real-world deployability, making him a key figure in the evolution of efficient, scalable AI systems.
Research Focus
Key Achievements
Top Papers
- 1Balancing Communication and Computation in Distributed Optimization111 citations · 2018