Description
An overview of activation function types and their applications: ReLU, Sigmoid, Tanh
• Cross Entropy: efficient loss beyond MSE
• Softmax - multiclass logistic function
• Intersection over union metric
• Using PyTorch datasets to organize the data for batch training
• Unet structure. Semantic Image segmentation
• An overview of GPU performance in various applications
• Brief intercomparison of different types of accelerators
• Key programming principles to achieve high GPU performance in Machine Learning
• Benchmarking CNNs on different GPUs: cnn-benchmarks test suite
• An overview of GPU-enabled libraries for feature extraction: torchvision.transform, Pandas, OpenCV, CUBLAS, CUFFT
• Performance limitations of pure Python code
• Implementing high-performance C++/Fortran/OpenCL extensions to PyTorch with pybind11
• Analyzing GPU efficiency for training and inference, by example of VoiceLoop – a neural text-to-speech processor
