AI Glossary

Framework - A software platform which supports machine learning algorithms. To explore various frameworks, click here.

Tensorflow - An open-source numerical computation software library for machine learning. To learn more about TensorFlow, click here.

CUDA® - A parallel computing platform and application programming interface (API) developed by NVIDIA that allows developers to use a CUDA-enabled GPU for general purpose computing. To learn more about CUDA, click here.

cuDNN - The NVIDIA CUDA® Deep Neural Network library (cuDNN) is a GPU-accelerated library of deep learning routines. To learn more about cuDNN, click here.

cuBLAS - The NVIDIA cuBLAS library is a GPU-accelerated implementation of standard basic linear algebra subroutines (BLAS). To learn more about cuBLAS, click here.

Training vs Inference - For a conceptual understanding of these terms, check out this  blog post.

Batch size - Batch size refers to the number of training samples that would be used in one pass through the network. For more info about batch size, check out this link on StackExchange.

Iteration - Since training data is divided into batches, an iteration refers to going through one batch of training data.

Epoch - In machine learning, an epoch is a single pass through the full training dataset.

Accuracy - Training accuracy is determined by test/training loss usually caused by a mismatch of training data.

Latency - How quickly a model can pass through one batch of images during inferencing, measured in ms.

Throughput - How many images a model can process within one second during inferencing or training.