A Friendly Introduction to Codeless Deep Learning

About this Video

• Video Title: A Friendly Introduction to Codeless Deep Learning
• Channel: KNIMETV
• Speakers: Catherine, Rosaria Silipo
• Duration: 00:52:26

Introduction

This webinar provides a beginner-friendly introduction to deep learning, focusing on how to build and train deep learning models using KNIME's Keras integration without writing code. The presentation is divided into two parts: the first covers the basics of neural networks, and the second introduces convolutional neural networks (CNNs) for image classification.

Key Takeaways

• Neural Network Fundamentals: A single neuron is the basic building block, receiving weighted inputs, adding a bias, and applying an activation function (sigmoid, tanh, ReLU). Feedforward neural networks connect multiple neurons in layers.
• Training Neural Networks: Training involves finding optimal weights to minimize errors using loss functions (binary cross-entropy, squared error) and gradient descent optimization algorithms (like Adam or Adagrad). Backpropagation efficiently calculates gradients.
• Loss Landscapes: Visualizing the loss function reveals the challenges of finding global minima due to numerous local minima, impacting the selection of the learning rate and the optimizer.
• Convolutional Neural Networks (CNNs): CNNs are inspired by the visual cortex. Convolutional layers use kernels (masks) to extract features from images, paired with pooling layers (max or average pooling) to reduce dimensionality and computational complexity. Transfer learning allows leveraging pre-trained models for new tasks.
• KNIME Keras Integration: KNIME's nodes provide a codeless way to define network architectures, select loss functions and optimizers, train models, monitor performance, and deploy them.