AI Hardware & Software Stack

Which describes the computational graph used by AI frameworks and compilers?

  1. AA line chart of AI performance by year
  2. BA chart showing how the training loss changes
  3. CA representation of a model's computation with operations as nodes and tensor flows as edges
Show answer and explanation

Answer: C. A representation of a model's computation with operations as nodes and tensor flows as edges

A computational graph is a blueprint of a model's computation, with operations such as matrix products, additions and activations as nodes and the tensors flowing between them as edges. Frameworks traverse it backward to compute backpropagation gradients automatically, and compilers optimize it as a whole through fusion and reordering. The graph also shows which operations are independent and can run in parallel.

  • A. It is not a statistical chart.
  • B. It is different from a loss curve.
  • C. Correct. It underpins optimization and automatic differentiation.
Practice this topic Back to topic

More AI Hardware & Software Stack questions

  1. Which describes ONNX?
  2. Why can inference runtimes such as TensorRT or ONNX Runtime run faster than using the training framework directly for inference?
  3. Why do element-wise operations such as ReLU or layer normalization tend to take a lot of time relative to their operation count in AI workloads?
  4. Why are convolutions in CNNs sometimes converted into matrix multiplications, for example with im2col?
  5. What is the difference between post-training quantization (PTQ) and quantization-aware training (QAT)?