What Are The Mechanics Of AI

The following mechanisms are the main processes used in modern AIs.

Supervised Learning

A type of machine learning where a model is trained on labelled data, meaning each training example is paired with the correct output. The model learns the mapping between inputs and outputs and uses this learned relationship to predict outputs for new inputs.

Unsupervised Learning

A type of machine learning where the model is trained on unlabelled data, meaning it must find patterns or groupings in the data without explicit output labels. Techniques like clustering and dimensionality reduction are common in unsupervised learning.

Reinforcement Learning

A type of machine learning in which an agent interacts with an environment to learn a policy that maximises cumulative reward. The agent learns through trial and error, receiving feedback in the form of rewards or penalties.

Neural Networks

Computational models inspired by the human brain, consist of layers of interconnected nodes (or neurons) that process input data and learn to recognize patterns through training. Each layer in a neural network transforms the input data, enabling the network to learn complex functions.

Deep Learning

A subset of machine learning that uses neural networks with multiple layers (deep neural networks) to model complex relationships in data. Deep learning is especially effective for high-dimensional data, such as images and text.

Overfitting

A modelling error that occurs when a machine learning model learns the noise in the training data rather than the actual patterns. This usually results in poor generalisation to new, unseen data.

Gradient Descent

An optimisation algorithm that iteratively adjusts model parameters to minimize a loss function. It calculates the gradient (slope) of the loss function with respect to each parameter, then updates the parameters in the opposite direction to reduce the loss.

Stochastic Gradient Descent (SGD)

A variant of gradient descent that updates model parameters for each training example rather than using the entire dataset. This approach can lead to faster convergence and is commonly used in deep learning.

Cross-Validation

A technique used to assess how well a model generalises to an independent dataset by splitting the data into multiple subsets (folds), training on some, and testing on others in multiple iterations. This helps reduce overfitting and improves model selection.

Hyperparameter

Settings used to control the training process of a model. Unlike parameters, which are learned from the data, hyperparameters are set before training begins, such as learning rate, batch size, or number of layers in a neural network.

Parameter

Values in a model that are learned from the training data, such as weights in a neural network. Parameters define the specific learned function that the model uses to make predictions.

Feature Engineering

The process of selecting, modifying, or creating input variables (features) that improve the performance of a machine learning model. It involves domain knowledge and techniques to extract useful information from raw data.

Feature Selection

A technique to select the most relevant features from the data, removing irrelevant or redundant ones to reduce dimensionality and improve model performance and efficiency.

Activation Function

A function applied to the output of each neuron in a neural network layer to introduce non-linearity, allowing the model to learn complex patterns. Examples include ReLU, sigmoid, and tanh functions.

Loss Function

A function that measures how well a model’s predictions match the actual data. It quantifies the error and guides the optimization process in adjusting parameters. Common loss functions include Mean Squared Error (MSE) and Cross-Entropy Loss.

Convolutional Neural Network (CNN)

A type of deep neural network specialized for processing structured grid data, like images. CNNs use convolutional layers to capture spatial features, making them highly effective for computer vision tasks.

Recurrent Neural Network (RNN)

A type of neural network designed for sequential data, such as time series or text, where each element depends on previous elements. RNNs have a memory component that retains information from previous inputs.

Transfer Learning

A technique where a pre-trained model on one task is adapted to a new but related task, requiring less data and training time. It is commonly used in computer vision and NLP applications.

Data Augmentation

A technique used to increase the diversity of data by creating modified versions of the dataset through transformations like rotations, flips, or colour changes. It is commonly used in image processing to prevent overfitting.

Latent Variables

Unobserved or hidden variables that explain patterns in observed data. In machine learning, they are often inferred from the data to represent complex structures or relationships.

Natural Language Processing (NLP)

A field of AI focused on the interaction between computers and human language, aiming to enable machines to understand, interpret, and generate human language. Applications include language translation, sentiment analysis, and chatbots.

Dimensionality Reduction

Techniques used to reduce the number of features in a dataset, simplifying models, and reducing computation while retaining important information. Examples include Principal Component Analysis (PCA) and t-SNE.

Recall

A metric in classification tasks that measures the proportion of true positive predictions among all actual positive instances, indicating the model’s ability to detect positive cases.

Hyperparameter Tuning

The process of systematically searching and optimising the hyperparameters of a model to achieve the best performance, often done through techniques like grid search or randomised search.

Ensemble Learning

A technique where multiple models are combined to improve overall performance. Examples include bagging, boosting, and stacking, which aggregate the outputs of multiple base models for a more accurate final prediction.

I wanted to make a true and honest assessment of our own proprietary AI, SES.

SES (Self Evolving Software) is a new AI utilising Supervised and Reinforcement Learning techniques. Its' unique processes reduce hallucinations, duplication of code, defects and code churn; all of the issues that are plaguing all current AIs, substantiated in a study by GitClear.

SES learns code of a customer's system and from a community of connected systems that have been rigorously tested, and leverages code of these systems, and importantly generates new code, referencing the existing code. This approach reduces hallucinations.

Before reviewing SES, it is fair to say SES is an AI solely to generate code. SES is not used for image recognition or creation, nor for speech synthesis or translation to text or other languages, nor for chat bots, nor for data insights and prediction. Thus, many of the mechanisms underpinning AI are not necessary for SES, which is why many of the mechanisms below are not employed by SES.

Below lists the AI mechanisms that SES does and does not use:

If you are interested to learn more or discuss anything connected to AI code generation, contact me at [email protected], or visit our website at www.selfevolvingsoftware.com