From Zero to DeepLearning With Scala (DeveloperWeek Global 2020)

Ai

Learn how to build a deep learning model from scratch with Scala, covering convolutional neural networks, TensorFlow, and overcoming AI bias, featuring a real-world example of detecting planes and bridges with a Raspberry Pi.

Key takeaways
  • It’s not easy to build a deep learning model, every iteration is a new starting point.
  • Convolutional neural networks are built on basic layers, with filters and activation functions.
  • The most common library for deep learning is TensorFlow.
  • AI projects are only a small part of AI, most work is in image recognition and feature extraction.
  • AI bias is a challenge, it’s difficult to train a model that’s not biased.
  • The speaker built a model to detect planes and bridges, using a Raspberry Pi.
  • The model uses convolutional and subsampling layers to identify features.
  • The speaker trained the model using a dataset of 16,000 images, with a high accuracy rate.
  • The model was trained from scratch, using a JVM and Scala.
  • The speaker highlights the importance of understanding neural networks and how they work before building a model.
  • Deep learning is complex, but understanding the basics can help with building a model.
  • AI is not just about building a model, it’s also about understanding how it works and how to use it.
  • The speaker encourages developers to learn more about deep learning and try building their own models.
  • The talk’s background story is about planes and airplanes, and the speaker’s curiosity about how to detect them using AI.