Krishna Teja Chitty-Venkata is a Postdoctoral Researcher at the Argonne Leadership Computing Facility (ALCF), Argonne National Laboratory, USA, where he works in the AI/ML team (formerly Data Science group). His research lies at the intersection of systems and machine learning, focusing on optimizing neural network training, finetuning, and inference on general-purpose and AI-specific hardware platforms. He is actively involved in AI for science applications and high-performance computing for AI (HPC for AI). Education: PhD in Computer Engineering, Iowa State University, 2017–2023 Bachelor of Engineering in Electronics and Communication Engineering, University College of Engineering, Osmania University, Hyderabad, India, 2013–2017 Research Interests: Krishna's research spans hardware-aware inference optimization of deep neural networks, enhancing training and finetuning of large language models (LLMs) and vision-language models (VLMs), neural architecture search (AutoML), pruning and quantization techniques, performance modeling, and AI for science. He is particularly interested in efficient adaptation methods such as LoRA-NAS integration, structured pruning (e.g., WActiGrad), and KV cache optimization (e.g., Paged Compression). Publication Trends: His recent work emphasizes benchmarking and optimization of LLMs on diverse AI accelerators (e.g., LLM-Inference-Bench), developing scalable frameworks for CNN and ViT evaluation (ConVision Benchmark), and advancing structured pruning and mixed-precision search methods. His publications span high-impact journals and conferences in computer science, AI, and systems, reflecting a strong focus on practical, hardware-aware solutions for deep learning efficiency. Scientific Contributions: Developed LLM-Inference-Bench for evaluating LLM performance across hardware and frameworks. Created ConVision Benchmark for standardized evaluation of CNNs and Vision Transformers. Proposed WActiGrad, a structured pruning method for efficient LLM finetuning and inference. Introduced Paged Compression for efficient KV cache management in vLLM. Designed LangVision-LoRA-NAS for optimizing VLMs via NAS-integrated adapters. Professional Experience and Advising: Krishna has been mentored by Prof. Arun K. Somani (Iowa State) and supervisors Murali Emani and Venkatram Vishwanath at Argonne. He has interned at AMD, Intel, and Argonne, contributing to deep learning optimization projects. While no formal students are listed, he has co-authored multiple papers with researchers and students, indicating collaborative advising. He has no publicly listed grants, but his work at Argonne is likely supported by institutional and DOE funding. Labs and Teams: He is part of the AI/ML team within the Argonne Leadership Computing Facility, a premier HPC and AI research division. His work involves close collaboration with teams developing AI accelerators and scientific applications, positioning him at the forefront of AI for science initiatives.
