Machine Learning for the Web: Transfer Learning

Posted on
ml4w machine_learning tensorflow
Machine Learning for the Web: Transfer Learning /img/itp/03_semester/ml4w/knn.gif

Takeaways

Transfer learning is the processing of using the outputs of one ML model as an input for a different model.

In this example, I’m using MobileNet image classification results. I’m using a smaller, older version with 256 labels. The data looks like

[{label: "cat", value: .5}, {label: "dog", value: 0}]

It surprised me how well it worked to reuse MobileNet for dimensionality reduction. I would’ve thought I would need to retrain a portion of the model.

Next steps:

  • arbitrary chaining of models
  • parallel processing chains
  • retraining models