Resume
Latest Resume of Manish Nagaraj
General Information
| Full Name | Manish Nagaraj |
| mnagara@purdue.edu | |
| Phone | +1-765-701-7970 |
| http://linkedin.com/in/m-nagaraj | |
| GitHub | https://github.com/manishnagaraj |
| Website | https://manishnagaraj.github.io |
| Location | West Lafayette, IN, USA |
Education
-
2019 - Present PhD in Electrical & Computer Engineering
Purdue University, West Lafayette, IN - GPA - 3.66
- Research Areas - Data efficiency for foundation models, large language models, multimodal learning, privacy-preserving machine learning, optimization techniques
- Research Advisor - Prof. Kaushik Roy
-
2017 - 2019 MS in Electrical & Computer Engineering
Purdue University, West Lafayette, IN - GPA - 3.72
- Thesis - Energy Efficient Byzantine Agreement Protocols for Cyber-Physical Resilience
-
2013 - 2017 Bachelor of Engineering in Electronics and Communications
PES Institute of Technology and Science, Bangalore, India - GPA - 9.77/10.0
Experience
-
Summer 2023 Research Intern, Integrated Systems Team
Latent AI, Skillman, NJ - Built data annotation and visualization tools for anomaly detection frameworks.
- Designed scalable pipelines for noisy, real-world ML datasets on edge and embedded systems.
- Profiled and improved energy and latency metrics for internal deep learning tools.
Research Projects
-
2025 TRIM - Token-wise Attention-Derived Saliency for Data-Efficient Instruction Tuning
Ph.D. Dissertation, Purdue University - Developed a forward-only data selection pipeline for LLM instruction tuning using interpretable multi-layer attention fingerprints.
- Improves data efficiency for models such as LLaMA-3.2 1B and LLaMA-3.1 8B without requiring gradients or Hessians.
- Enables cross-model and cross-scale transfer for personalization and domain adaptation.
- Manuscript under review.
-
2025 Coresets from Trajectories - Selecting Data via Correlation of Loss Differences
Ph.D. Dissertation, Purdue University - Introduced the Correlation of Loss Differences (CLD) metric for scalable, gradient-free data selection.
- Achieved state-of-the-art coreset performance on CIFAR-100 and ImageNet-1K across multiple architectures.
- Demonstrated less than 1% degradation under cross-architecture transfer settings.
- Accepted for publication at TMLR 2025.
-
2024 DOTIE - Energy-Efficient Object Detection Using Event Cameras
Ph.D. Research, Purdue University - Designed a lightweight object detection framework leveraging event-driven camera data and spiking neural networks.
- Demonstrated real-time performance on resource-constrained hardware.
- Presented at ICRA 2023 and CVPR 2023 Workshops.
-
2023 TOFU - Federated Learning with Data and Communication Efficiency
Ph.D. Research, Purdue University - Proposed a federated learning framework that jointly improves data and communication efficiency for heterogeneous clients.
- Reduced communication overhead by up to 10x while maintaining model accuracy.
- Published in IEEE Access 2024.
Skills
| Programming Languages and OS | Python, Ubuntu |
| Software Development Tools | Docker, GitHub |
| DL Frameworks and Libraries | PyTorch, Hugging Face Transformers, OpenCV, NumPy, SciPy |
Relevant Coursework
- Artificial Intelligence
- Statistical Machine Learning
- Random Processes and Probability
- Linear Algebra
- Computational Models and Algorithms (DSA)
- Distributed Computer Systems
- Computer Networks
References
- Prof. Kaushik Roy, Purdue University, West Lafayette, USA - kaushik@purdue.edu