Lecture 3 - CS7015: Neural Networks

A mathematical function where the input (x) can have multiple real number values (e.g., salinity, density) and the output (y) is also a real number (e.g., predicted oil amount).

A type of neuron that uses the sigmoid function to introduce non-linearity into a neural network. The sigmoid function squashes its input value between 0 and 1.

A mathematical technique for finding the best parameters for a model by repeatedly adjusting them in the direction that minimizes a loss function.

A neural network architecture where information flows only in one direction, from input to output, without any loops or cycles.

The ability of a model to learn and represent complex relationships between inputs and outputs, including non-linear patterns.

A location in the ocean containing potential oil reserves that is evaluated for its suitability for drilling operations based on specific factors.

The measure of salt concentration in seawater. This could be a factor in determining the potential oil yield at an oil drilling location.

The average mass per unit volume of seawater at a specific location. This could influence the amount of oil that might be present.

The ability of a neuron to fire (output a signal) when the weighted sum of its inputs exceeds a certain threshold.

A type of artificial neuron that uses a step function to determine its output. It outputs a '1' if the weighted sum of its inputs exceeds a threshold and a '0' otherwise.

The process of setting the threshold for a perceptron to determine when it will fire. It involves finding the optimal value that balances false positives and false negatives.

The characteristic of a perceptron to make abrupt decisions based on a defined threshold, leading to sharp transitions between different outputs.

The process of representing real-world data and relationships using mathematical functions, allowing for more accurate and nuanced predictions.

The weighted sum of all inputs to a perceptron, which is then compared to the threshold to determine if the neuron should fire.

A mathematical function that introduces non-linearity into a neural network, squashing its input value between 0 and 1, making it suitable for representing probabilities or smooth transitions.

A function that maps inputs from real-valued vectors to a single real number output.

A type of neuron that uses a non-linear activation function, commonly a sigmoid function (s-shaped), to process inputs and produce outputs.

The ability of a neural network to approximate a target function by learning from a large and diverse set of training examples.

A set of input-output pairs used to train a neural network, providing examples of the desired relationship between inputs and outputs.

The degree to which a neural network's output matches the expected output for given inputs, often measured by error rate or distance from the true value.

The process where a neural network adjusts its internal parameters based on the errors observed during training, improving its ability to make better predictions.

A type of neuron that uses a step function to determine its output, producing either a '0' or a '1' output based on the input.

The ability to map a truth table exactly with a neural network, meaning the network can accurately represent all possible outputs for any combination of inputs.

Calculus is a fundamental tool in machine learning, particularly when working with functions that are smooth, continuous, and differentiable.

In a perceptron, a distinct boundary separating input data points that trigger a '1' output from those resulting in a '0' output. Input combinations exceeding a predefined threshold result in a '1', while those below the threshold yield a '0'.

A specific type of sigmoid function that maps input values to a range between 0 and 1. It is widely used in neural networks for tasks like classification.

The sum of weighted inputs in a neuron. It is calculated by multiplying each input by its corresponding weight and summing the results.

The value of the sigmoid function approaches 1 as the weighted sum tends towards infinity. This indicates a high probability of activating the neuron.

The value of the sigmoid function approaches 0 as the weighted sum tends towards negative infinity. This indicates a low probability of activating the neuron.

The range of values that the sigmoid function outputs lies between 0 and 1, similar to probabilities.

Unlike the discrete '0' or '1' outputs of a perceptron, sigmoid neurons produce continuous outputs between 0 and 1. This allows for a smoother and more nuanced response to input variations.