Gradient Descent: Difference between revisions
(Marked this version for translation) Tag: Reverted |
([deploy-bot] Deploy from CI (a613d0f)) Tags: ci-deploy Manual revert |
||
| Line 8: | Line 8: | ||
'''Gradient descent''' is a first-order iterative optimisation algorithm for finding a local minimum of a differentiable function. It is the foundation of nearly all modern machine-learning training procedures, from simple linear regression to billion-parameter deep neural networks. | '''Gradient descent''' is a first-order iterative optimisation algorithm for finding a local minimum of a differentiable function. It is the foundation of nearly all modern machine-learning training procedures, from simple linear regression to billion-parameter deep neural networks. | ||
<!--T:2--> | |||
== Intuition == | |||
<!--T:3--> | <!--T:3--> | ||
| Line 16: | Line 17: | ||
The size of each step is controlled by a scalar called the '''learning rate''' (often denoted <math>\eta</math>). A large learning rate covers ground quickly but risks overshooting the minimum; a small learning rate converges more reliably but may take prohibitively many steps. | The size of each step is controlled by a scalar called the '''learning rate''' (often denoted <math>\eta</math>). A large learning rate covers ground quickly but risks overshooting the minimum; a small learning rate converges more reliably but may take prohibitively many steps. | ||
<!--T:5--> | |||
== Mathematical formulation == | |||
<!--T:6--> | <!--T:6--> | ||
| Line 36: | Line 38: | ||
The gradient <math>\nabla f</math> points in the direction of steepest ascent, so subtracting it moves the iterate downhill. | The gradient <math>\nabla f</math> points in the direction of steepest ascent, so subtracting it moves the iterate downhill. | ||
== Batch, stochastic, and mini-batch variants == | <!--T:12--> | ||
== Batch, stochastic, and mini-batch variants == | |||
<!--T:13--> | <!--T:13--> | ||
| Line 62: | Line 65: | ||
Full batch gradient descent computes the exact gradient and therefore follows a smooth trajectory toward the minimum. [[Stochastic Gradient Descent|Stochastic gradient descent]] uses a single sample to estimate the gradient, drastically reducing computation per step at the cost of a noisier trajectory. Mini-batch gradient descent strikes a balance and is the most common choice in practice, with typical batch sizes between 32 and 512. | Full batch gradient descent computes the exact gradient and therefore follows a smooth trajectory toward the minimum. [[Stochastic Gradient Descent|Stochastic gradient descent]] uses a single sample to estimate the gradient, drastically reducing computation per step at the cost of a noisier trajectory. Mini-batch gradient descent strikes a balance and is the most common choice in practice, with typical batch sizes between 32 and 512. | ||
<!--T:18--> | |||
== Convergence == | |||
<!--T:19--> | |||
=== Convex functions === | |||
<!--T:20--> | <!--T:20--> | ||
| Line 75: | Line 80: | ||
The ratio <math>\kappa = L / \mu</math> is called the '''condition number''' and governs how quickly the algorithm converges. Ill-conditioned problems (large <math>\kappa</math>) converge slowly. | The ratio <math>\kappa = L / \mu</math> is called the '''condition number''' and governs how quickly the algorithm converges. Ill-conditioned problems (large <math>\kappa</math>) converge slowly. | ||
=== Non-convex functions === | <!--T:23--> | ||
=== Non-convex functions === | |||
<!--T:24--> | <!--T:24--> | ||
Most deep-learning objectives are non-convex. In this setting gradient descent is only guaranteed to converge to a stationary point (where <math>\nabla f = 0</math>), which could be a local minimum, saddle point, or even a local maximum. In practice, saddle points are more problematic than local minima in high-dimensional spaces. | Most deep-learning objectives are non-convex. In this setting gradient descent is only guaranteed to converge to a stationary point (where <math>\nabla f = 0</math>), which could be a local minimum, saddle point, or even a local maximum. In practice, saddle points are more problematic than local minima in high-dimensional spaces. | ||
<!--T:25--> | |||
== Learning rate selection == | |||
<!--T:26--> | <!--T:26--> | ||
| Line 94: | Line 101: | ||
A common heuristic is to try several values on a logarithmic scale (e.g. <math>10^{-1}, 10^{-2}, 10^{-3}</math>) and pick the one that reduces the loss fastest without instability. | A common heuristic is to try several values on a logarithmic scale (e.g. <math>10^{-1}, 10^{-2}, 10^{-3}</math>) and pick the one that reduces the loss fastest without instability. | ||
<!--T:29--> | |||
== Extensions and improvements == | |||
<!--T:30--> | <!--T:30--> | ||
| Line 105: | Line 113: | ||
* '''Second-order methods''' — algorithms like Newton's method and L-BFGS use curvature information (the Hessian or its approximation) for faster convergence, but are often too expensive for large-scale problems. | * '''Second-order methods''' — algorithms like Newton's method and L-BFGS use curvature information (the Hessian or its approximation) for faster convergence, but are often too expensive for large-scale problems. | ||
<!--T:32--> | |||
== Practical tips == | |||
<!--T:33--> | <!--T:33--> | ||
| Line 113: | Line 122: | ||
* '''Monitoring the loss curve''' — plotting the training loss over iterations is the simplest diagnostic: a smoothly decreasing curve indicates healthy training; oscillations suggest the learning rate is too high. | * '''Monitoring the loss curve''' — plotting the training loss over iterations is the simplest diagnostic: a smoothly decreasing curve indicates healthy training; oscillations suggest the learning rate is too high. | ||
<!--T:34--> | |||
== Applications == | |||
<!--T:35--> | <!--T:35--> | ||
| Line 125: | Line 135: | ||
* Economics and game-theoretic equilibrium computation | * Economics and game-theoretic equilibrium computation | ||
<!--T:37--> | |||
== See also == | |||
<!--T:38--> | <!--T:38--> | ||
| Line 134: | Line 145: | ||
* [[Overfitting and Regularization]] | * [[Overfitting and Regularization]] | ||
<!--T:39--> | |||
== References == | |||
<!--T:40--> | <!--T:40--> | ||
Revision as of 02:09, 27 April 2026
| Article | |
|---|---|
| Topic area | Optimization |
| Difficulty | Introductory |
Gradient descent is a first-order iterative optimisation algorithm for finding a local minimum of a differentiable function. It is the foundation of nearly all modern machine-learning training procedures, from simple linear regression to billion-parameter deep neural networks.
Intuition
Imagine standing on a mountainside in thick fog. You cannot see the valley floor, but you can feel the slope beneath your feet. The most natural strategy is to take a step in the steepest downhill direction, then reassess. Gradient descent formalises precisely this idea: at each step, the algorithm computes the direction of steepest increase of the function (the gradient) and moves in the opposite direction.
The size of each step is controlled by a scalar called the learning rate (often denoted $ \eta $). A large learning rate covers ground quickly but risks overshooting the minimum; a small learning rate converges more reliably but may take prohibitively many steps.
Mathematical formulation
Given a differentiable objective function $ f:\mathbb{R}^n \to \mathbb{R} $, gradient descent generates a sequence of iterates by the update rule:
- $ \theta_{t+1} = \theta_t - \eta \, \nabla f(\theta_t) $
where $ \nabla f(\theta_t) $ is the gradient vector evaluated at the current point $ \theta_t $ and $ \eta > 0 $ is the learning rate.
In the one-dimensional case this simplifies to:
- $ \theta_{t+1} = \theta_t - \eta \, f'(\theta_t) $
The gradient $ \nabla f $ points in the direction of steepest ascent, so subtracting it moves the iterate downhill.
Batch, stochastic, and mini-batch variants
When the objective has the form of an average over data points,
- $ f(\theta) = \frac{1}{N}\sum_{i=1}^{N} \ell(\theta;\, x_i, y_i) $
three common strategies differ in how much data is used to estimate the gradient:
| Variant | Gradient computed over | Per-step cost | Gradient noise |
|---|---|---|---|
| Batch (full) gradient descent | All $ N $ samples | High | None |
| Stochastic gradient descent (SGD) | 1 random sample | Low | High |
| Mini-batch gradient descent | $ B $ random samples ($ 1 < B < N $) | Medium | Medium |
Full batch gradient descent computes the exact gradient and therefore follows a smooth trajectory toward the minimum. Stochastic gradient descent uses a single sample to estimate the gradient, drastically reducing computation per step at the cost of a noisier trajectory. Mini-batch gradient descent strikes a balance and is the most common choice in practice, with typical batch sizes between 32 and 512.
Convergence
Convex functions
For a convex function with Lipschitz-continuous gradients (constant $ L $), gradient descent with a fixed learning rate $ \eta \leq 1/L $ converges at a rate of $ O(1/t) $. If the function is additionally strongly convex with parameter $ \mu > 0 $, convergence accelerates to a linear (exponential) rate:
- $ f(\theta_t) - f(\theta^*) \leq \left(1 - \frac{\mu}{L}\right)^t \bigl(f(\theta_0) - f(\theta^*)\bigr) $
The ratio $ \kappa = L / \mu $ is called the condition number and governs how quickly the algorithm converges. Ill-conditioned problems (large $ \kappa $) converge slowly.
Non-convex functions
Most deep-learning objectives are non-convex. In this setting gradient descent is only guaranteed to converge to a stationary point (where $ \nabla f = 0 $), which could be a local minimum, saddle point, or even a local maximum. In practice, saddle points are more problematic than local minima in high-dimensional spaces.
Learning rate selection
Choosing the learning rate is one of the most important practical decisions:
- Too large — the iterates oscillate or diverge.
- Too small — convergence is unacceptably slow.
- Learning rate schedules — many practitioners start with a larger rate and reduce it over time (step decay, exponential decay, cosine annealing).
- Line search — classical numerical methods choose $ \eta $ at each step to satisfy conditions such as the Wolfe or Armijo conditions, though this is rare in deep learning.
A common heuristic is to try several values on a logarithmic scale (e.g. $ 10^{-1}, 10^{-2}, 10^{-3} $) and pick the one that reduces the loss fastest without instability.
Extensions and improvements
Several important modifications address limitations of vanilla gradient descent:
- Momentum — accumulates a velocity vector from past gradients, helping to accelerate convergence in ravine-like landscapes.
- Nesterov accelerated gradient — a momentum variant that evaluates the gradient at a look-ahead position, yielding better theoretical convergence rates.
- Adaptive methods (Adagrad, RMSProp, Adam) — maintain per-parameter learning rates that adapt based on the history of gradients.
- Second-order methods — algorithms like Newton's method and L-BFGS use curvature information (the Hessian or its approximation) for faster convergence, but are often too expensive for large-scale problems.
Practical tips
- Feature scaling — normalising input features so they have similar ranges dramatically improves convergence, because the loss surface becomes more isotropic.
- Gradient clipping — capping the norm of the gradient prevents excessively large updates.
- Random initialisation — starting from a reasonable random initialisation (e.g. Xavier or He initialisation for neural networks) avoids symmetry-breaking issues.
- Monitoring the loss curve — plotting the training loss over iterations is the simplest diagnostic: a smoothly decreasing curve indicates healthy training; oscillations suggest the learning rate is too high.
Applications
Gradient descent and its variants are used throughout science and engineering:
- Training machine-learning models (linear models, neural networks, support vector machines)
- Signal processing and control systems
- Inverse problems in physics and imaging
- Operations research and logistics optimisation
- Economics and game-theoretic equilibrium computation
See also
- Stochastic Gradient Descent
- Backpropagation
- Loss Functions
- Neural Networks
- Overfitting and Regularization
References
- Cauchy, A. (1847). "Méthode générale pour la résolution des systèmes d'équations simultanées". Comptes Rendus de l'Académie des Sciences.
- Boyd, S. and Vandenberghe, L. (2004). Convex Optimization. Cambridge University Press.
- Ruder, S. (2016). "An overview of gradient descent optimization algorithms". arXiv:1609.04747.
- Goodfellow, I., Bengio, Y. and Courville, A. (2016). Deep Learning, Chapter 8. MIT Press.