Krishna Teja Chitty

Research Experience (ISU)

My research involves optimization (Pruning, Quantization, Neural Architecture Search) of Deep Neural Networks for efficient inference on hardware. My research projects are as follows:

1. Array Aware Training/Pruning: Designed an Hyper-parameter tuning method and a Pruning algorithm for CNN, Dense networks which are dependent on the dimensions of a systolic array. The goal of the developed techniques is to minimize the number of processing cycles of DNN forward pass on a systolic array based Neural Network accelerator; similar to Google’s TPU and MIT's Eyeriss (Paper in ASAP ’20 Conference).


2. Model Compression on Faulty DNN Accelerator: Developed a fault model followed by co-designing a Pruning algorithm to bypass faults and reduce latency on Faulty DNN Accelerators. (short version at ASAP ’19. Full paper at PRDC '20)


3. Layer Fusion Perspective of Model Compression: Developed a symmetric Pruning and a uniform Quantization algorithm to compress the Neural Networks for efficient infernce on CPU and GPU devices along with stuyding layer fusion aspect of model compression (Paper in DSS '20).

Publications

1. Neural Architecture Search Benchmarks: Insights and Survey [Paper]
   Krishna Teja Chitty-Venkata, Murali Emani, Venkatram Vishwanath and Arun Somani
   IEEE Access (2021 Impact Factor: 3.476)



2. Neural Architecture Search for Transformers: A Survey [Paper]
   Krishna Teja Chitty-Venkata, Murali Emani, Venkatram Vishwanath and Arun Somani
   IEEE Access (2021 Impact Factor: 3.476)



3. Efficient Design Space Exploration for Sparse Mixed Precision Neural Architectures [Paper]
   Krishna Teja Chitty-Venkata, Murali Emani, Venkatram Vishwanath and Arun Somani
   31st International Symposium on High-Performance Parallel and Distributed Computing (HPDC) 2022 (Acceptance Rate = 19%)



4. Neural Architecture Search Survey: A Hardware Perspective [Paper]
   Krishna Teja Chitty-Venkata and Arun Somani
   2022 ACM Computing Surveys (CSUR) (2021 Impact Factor: 14.324; ranked 3/109 in Computer Science Theory & Methods)



5. Array Aware Neural Architecture Search [Paper]
   Krishna Teja Chitty-Venkata and Arun Somani
   32nd IEEE International Conference on Application-specific Systems, Architectures and Processors (ASAP) 2021



6. Searching Architecture and Precision for U-net based Image Restoration Tasks [Paper]
   Krishna Teja Chitty-Venkata, Sreenivas Kothandaraman and Arun Somani
   28th IEEE International Conference on Image Processing (ICIP) 2021 (Acceptance Rate = 45%)



7. Calibration Data-Based CNN Filter Pruning for Efficient Layer Fusion [Paper]
   Krishna Teja Chitty-Venkata and Arun Somani
   22nd IEEE International Conference on High Performance Computing and Communications (HPCC) 2020
   18th IEEE International Conference on Smart City (SmartCity) 2020
   6th IEEE International Conference on Data Science and Systems (DSS) 2020 (Acceptance Rate = 20%)



8. Model Compression on Faulty Array-based Neural Network Accelerator [Paper]
   Krishna Teja Chitty-Venkata and Arun Somani
   25th IEEE Pacific Rim International Symposium on Dependable Computing (PRDC) 2020 (Acceptance Rate = 40%)



9. Array Aware Training/Pruning: Methods for Efficient Forward Propagation on Array-based Neural Network Accelerators [Paper][Video]
   Krishna Teja Chitty-Venkata and Arun Somani
   31th IEEE International Conference on Application-specific Systems, Architectures and Processors (ASAP) 2020 (Acceptance Rate = 24%)



10. Impact of Structural Faults on Neural Network Performance [Paper] [Poster]
    Krishna Teja Chitty-Venkata and Arun Somani
    30th IEEE International Conference on Application-specific Systems, Architectures and Processors (ASAP) 2019



11. Estimation of Modulation Parameters for LPI Radar using Quadrature Mirror Filter Bank [Paper]
    Metuku Shyamsunder, Kakarla Subbarao, Bharath Regimanu, CVSSD Krishna Teja
    IEEE UP Section Conference on Electrical Computer and Electronics (UPCON) 2017