README.md

Links

• The basics of neural networks, and the math behind how they learn

Local Execution

• Ollama
• llama.cpp
• torchchat

Terms

• Artificial Intelligence (AI): A branch of computer science focused on creating systems that can perform tasks typically requiring human intelligence, such as reasoning, learning, problem-solving, and understanding natural language.
• Machine Learning (ML) A subset of AI that involves training algorithms to recognize patterns in data and make predictions or decisions without being explicitly programmed for specific tasks.
  • Supervised Learning: The model is trained on labeled data (input-output pairs).
  • Unsupervised Learning: The model is trained on data without explicit labels, often used for clustering or association tasks.
  • Reinforcement Learning: An agent learns to make decisions by receiving rewards or penalties based on its actions.
  • Deep Learning: A subset of machine learning that uses neural networks with many layers (deep neural networks) to model complex patterns in large amounts of data.
  • Transfer Learning: The practice of using a pre-trained model on a new task, leveraging knowledge learned from a different but related problem.
  • Fine-tuning: The process of adjusting a pre-trained model
• Neural Networks: Computational models inspired by the human brain, consisting of layers of interconnected nodes (neurons) that process input data.
  • Neuron: The basic unit in a neural network that performs a simple computation based on input data and weights.
  • Activation Function: A function applied to the output of a neuron to introduce non-linearity, allowing the network to model more complex patterns (e.g., ReLU, sigmoid, tanh).
  • Layer: A group of neurons in a neural network that processes input data in parallel.
    • Input Layer: The first layer of a neural network that receives the input data.
    • Hidden Layer: Layers between the input and output layers that perform intermediate processing.
    • Output Layer: The final layer that produces the network’s predictions.
• Training and Evaluation:
  • Training: The process of feeding data to a model and adjusting its parameters (weights and biases) to minimize the error in predictions.
  • Backpropagation: The algorithm used to adjust the weights of the network based on the error of predictions during training.
  • Epoch: One complete pass through the entire training dataset.
  • Batch: A subset of the training data processed together during one iteration.
  • Learning Rate: A parameter that controls the size of the step the model takes during optimization.
  • Overfitting: A situation where a model performs well on the training data but poorly on unseen data, usually due to learning noise or details specific to the training set.
  • Underfitting: A situation where a model is too simple to capture the underlying patterns in the data, resulting in poor performance on both the training and test data.
• Inference: The process that a trained machine learning model uses to draw conclusions from brand-new data. Or in simple words: Inference is an AI model in action.
• Evaluation Metrics:
  • Accuracy: The proportion of correct predictions made by the model.
  • Precision: The proportion of true positive predictions among all positive predictions.
  • Recall: The proportion of true positives identified by the model out of all actual positives.
  • F1 Score: The harmonic mean of precision and recall, providing a single measure of model performance.
  • Confusion Matrix: A table showing the performance of a classification model, with rows representing actual classes and columns representing predicted classes.
• Optimization:
  • Gradient Descent: An optimization algorithm used to minimize the loss function by iteratively adjusting the model’s parameters.
  • Stochastic Gradient Descent (SGD): A variant of gradient descent that updates the model parameters for each batch of data rather than the entire dataset.
  • Adam Optimizer: An advanced optimization algorithm that adjusts the learning rate adaptively based on first and second moments of the gradient.
• Data Preparation:
  • Feature Engineering: The process of selecting and transforming variables (features) to improve the performance of a model.
  • Normalization: The process of scaling input data so that it fits within a specific range, often used to speed up training.
  • Data Augmentation: Techniques used to increase the diversity of training data without collecting new data, such as flipping or rotating images in computer vision tasks.
• Regularization: Techniques used to prevent overfitting by penalizing complex models. Common methods include L1 (Lasso) and L2 (Ridge) regularization.
  • Dropout: A regularization technique where random neurons are ignored during training to prevent the model from becoming too dependent on specific neurons.