Skip to 0 minutes and 15 seconds Before deep learning, researchers are using handcrafted features for image recognition. Now the features are automatically learned by CNN.
Skip to 0 minutes and 27 seconds This makes us wondering: Can we automate the whole learning process and let the computers to learn the neural network architecture? Image that we only need to prepare and label the data, computer will do the rest. So everybody can be data scientist!
Skip to 0 minutes and 48 seconds This branch of research is called AutoML: Automated Machine Learning In terms of AutoML, Google has proposed NASNet, which can automatically search the best network architecture. NASNet has achieved best accuarcy on ImageNet dataset, surpassed all human designed models. In 2019, Google proposed a new systematic approach to balance the model scaling of network depth, width, and resolution, which is called EfficientNet. The EfficientNet outperformed previous NASNet, while being 8.4x smaller and 6.1x faster on inference than the best existing CNN model. AutoML points out a new research direction in the future.
Automated Machine Learning
In this video, Prof. Lai will explain AutoML: Automated Machine Learning.
The origin of this research is from the question: Can we automate the whole learning process and let the computers to learn the neural network architecture? Let’s see how this designed model works.