Deep Learning Basics

Definition

• Deep Learning: A subset of machine learning, which is a subset of artificial intelligence (AI).
• Artificial Intelligence: Technique enabling machines to mimic human behavior.
• Machine Learning: Achieves AI through algorithms trained with data.
• Deep Learning: Type of machine learning inspired by the structure of the human brain (artificial neural network).

Difference Between Deep Learning and Machine Learning

• Machine Learning Example:
  • Task: Differentiate between tomatoes and cherries.
  • Requires human-defined features (e.g., size, type of stem).
• Deep Learning Example:
  • Features are automatically selected by the neural network without human intervention.
  • Requires a larger volume of data for training.

Working of Neural Networks

Basic Structure

• Neurons: Core entities of a neural network; process information.
• Input Layer: Receives data (e.g., 784 pixels for 28x28 image).
• Hidden Layers: Layers between input and output layers.
• Output Layer: Each neuron represents a digit.

Information Processing Flow

1. Input Layer: Each pixel is fed to a neuron.
2. Weighted Channels: Each connection has a value (weight).
3. Bias: Unique number added to weighted sum of inputs.
4. Activation Function: Determines if a neuron gets activated.
5. Information Transfer: Activated neurons pass information to subsequent layers until reaching the output layer.

Activation Process

• The neuron activated in the output layer corresponds to the input digit.
• Weights and biases adjusted to create a well-trained network.

Applications of Deep Learning

• Customer Support: Conversational bots providing realistic interactions.
• Medical Care: Detecting cancer cells and analyzing MRI images.
• Self-Driving Cars: Companies like Apple, Tesla, and Nissan working on autonomous vehicles.

Limitations of Deep Learning

1. Data Requirement:
   • Needs a massive volume of data for effective training.
2. Computational Power:
   • Requires graphical processing units (GPUs) with thousands of cores, more expensive than CPUs.
3. Training Time:
   • Takes hours to months to train deep neural networks; increases with data size and layers.

Popular Deep Learning Frameworks

• TensorFlow
• PyTorch
• Keras
• Deep Learning 4j
• Caffe
• Microsoft Cognitive Toolkit

Future of Deep Learning and AI

• Development of innovative devices, such as Horus Technology's device for the blind, using deep learning and computer vision.
• Continuous advancements suggesting that we have only scratched the surface of deep learning capabilities.

Conclusion

• The future holds many surprises in deep learning and AI technologies.