Łukasz Gołębiowski

Contractor for AMD.

Working in Neural Ray Tracing team.

Experiments with adopting GEMM algorithms and quantization/dequantization (Fp32 -> Int8, Int8 -> Fp32) to U-net neural network on AMD GCN GPU architecture.

GPU optimization research.

Languages: C++11, HLSL, CUDA, HIP

Tools: Linux, VS2019, Docker, Git

Working on GPU part of oneDNN team.

Improving and measuring performance on different AI operators launched on Intel's Xe GPU accelerators.

Adding new features and test cases to existing GPU kernels.

Reporting, debugging and fixing bugs after CI tests.

Writing and reading documentation.

Languages: C++11, C, CUDA, OpenCL, ISA

Tools: Jira, Git, Linux, Bash, Cmake, STL, VS, GDB, Gtest

Working in Automatic Speech Recognition.

Pretraining with Hidden Markov Models and Gaussian Mixture Models on acoustic filter banks from audio recordings

Training of Neural Networks od acoustic models

Transfer learning from trained models on one language to another.

Working on decoding from acoustic models to nlp text models on Tizen operating system.

Preparing audio samples to trainings.

Languages: C++11, Bash, Python, Perl.

Tools: Linux, Cmake, Kaldi toolkit, mathematical fundamentals of DSP and NLP

Member of project "Algebraic variety with trivial first Chern class" supported by Polish NCN grant.

Researching and presenting recently publications on algebraic geometry.

Partial classification of Calabi - Yau 3-folds in P4 completed with publication of the article "aCM Calabi-Yau varieties of codimension 4 in P4" in AMS Electronic Mathematical Annoucement

Combinatorial algebraic computations by using Magma Algebra software.

Participating on conferences and schools.

Languages: Bash

Tools: Magma Algebra System, LaTex

Analyst

Preparing different Neural Networks for predictive maintenance.

Training and reporting results from NN.

Languages: Python 2.7

Tools: Tensorflow, Caffe, MXnet