Balaji Venkataramana

Working on TPUs and such!

Exploration of Cross-Layer Approximation for Artificial Neural Networks on FPGAs 1) Approximations explored in tasks with high hardware complexity such as multiplication, division, sine function and log compression. - Multiplication - Minimally Biased Approximate Integer Multiplier (MBM) - Division - SIMDive approximate division algorithm (inspired from Mitchell's approximation) - Log-compression - Approx log-compression function - Sin function - CORDIC algorithm 2) Approximations applied over 3 Deep Neural Networks targeted at different problems - DenseNet (Mobilenetv2 backbone) - Encoder decoder architecture - Single Image Depth Estimation - DexiNed - Upsampling + CNN architecture - Edge Detection - SincNet - KWS (Key Word Spotting over raw audio data)

Worked in the 'Wireless Communication Group (WLSG)' team. My work included --> Design of an Asynchronous FIFO with any FIFO depth --> Setup and hold time verification --> Clk, uncertainties and multi-cycle-path assertions

Worked on the project 'Single Image Depth Estimation using Deep Learning Techniques' Work included: --> Transfer Learning --> SSIM (Structural Similarity Index) for comparing images in loss function

ECE752: Advanced Computer Architecture-I [Single Core Architectures, Out-of-Order Superscalar] ECE757: Advanced Computer Architecture-II [Multi-core Architectures, Coherence, Consistency] ECE755: VLSI Systems Design [CMOS characteristics, Power-Performance-Area Tradeoff, STA, Memories] ECE759: High Performance Computer Architecture [CUDA, OpenMP, GPU Architecture] CS557: Parallel & Throughput Optimized Computing CS537: Intro to Operating Systems ECE532: Matrix Methods for Machine Learning ECE552: Intro to Computer Architecture CS758: Quantum Systems