1986–, Computer scientist
Alex Krizhevsky is a Ukrainian-Canadian computer scientist whose 2012 paper ImageNet Classification with Deep Convolutional Neural Networks (with Ilya Sutskever and Geoffrey Hinton) won the ImageNet Large Scale Visual Recognition Challenge with a top-5 error of 15.3%, ten percentage points better than the second-place entry. The result, achieved with a 60-million-parameter convolutional network trained on two GTX 580 GPUs over six days, is universally regarded as the moment the modern deep-learning era began.
AlexNet's architectural and methodological choices, ReLU activations, dropout, GPU training, data augmentation, local response normalisation, large fully-connected layers, defined CNN practice for years. The combination of a sufficiently large dataset (ImageNet), sufficient compute (GPU acceleration) and the right architecture finally vindicated the connectionist programme that Hinton, LeCun and Bengio had advocated for two decades.
Krizhevsky completed his PhD at Toronto under Hinton, joined Google after the 2013 DNNResearch acquisition that brought the AlexNet team to Google Brain, and left active research in 2017 to pursue personal interests. His public output since AlexNet has been minimal, but the result with which he is associated remains foundational.
Related people: Ilya Sutskever, Geoffrey Hinton, Fei-Fei Li
Works cited in this book:
- ImageNet Classification with Deep Convolutional Neural Networks (2012) (with Ilya Sutskever, Geoffrey E. Hinton)
- Dropout: A Simple Way to Prevent Neural Networks from Overfitting (2014) (with Nitish Srivastava, Geoffrey E. Hinton, Ilya Sutskever, Ruslan Salakhutdinov)
Discussed in:
- Chapter 11: CNNs, CNNs in Vision