Overview
Direct Answer
t-SNE (t-Distributed Stochastic Neighbour Embedding) is a non-linear dimensionality reduction algorithm that maps high-dimensional data into two or three-dimensional space while preserving local neighbourhood structure. Unlike linear techniques such as PCA, it excels at revealing cluster separation and hidden patterns in complex datasets.
How It Works
The algorithm converts high-dimensional Euclidean distances into conditional probabilities representing neighbourhood relationships, then iteratively minimises the divergence between these probabilities in the original and low-dimensional spaces using gradient descent. It employs a Student's t-distribution in the low-dimensional space, which provides heavier tails than Gaussian distributions and allows dissimilar points to repel effectively, producing clearer cluster visualisations.
Why It Matters
Teams rely on t-SNE for exploratory data analysis when assessing dataset quality, validating clustering outcomes, and identifying outliers before model deployment. The technique accelerates decision-making in data science workflows by enabling rapid visual inspection of unlabelled data, reducing the cost of manual annotation and improving confidence in downstream model selection.
Common Applications
Practitioners use the method to visualise gene expression profiles in genomics research, explore image embeddings in computer vision pipelines, and inspect document similarity in natural language processing. It is standard in single-cell RNA sequencing analysis and helps data scientists validate the separability of classes in classification tasks.
Key Considerations
The algorithm is computationally expensive for large datasets and sensitive to hyperparameters such as perplexity; results may vary significantly across runs due to stochastic initialisation. It preserves local structure but distorts global distances, making it unsuitable for quantitative analysis or downstream model input.
Cross-References(1)
More in Machine Learning
Elastic Net
Training TechniquesA regularisation technique combining L1 and L2 penalties, balancing feature selection and coefficient shrinkage.
Ensemble Methods
MLOps & ProductionMachine learning techniques that combine multiple models to produce better predictive performance than any single model, including bagging, boosting, and stacking approaches.
Bias-Variance Tradeoff
Training TechniquesThe balance between a model's ability to minimise bias (error from assumptions) and variance (sensitivity to training data fluctuations).
Machine Learning
MLOps & ProductionA subset of AI that enables systems to automatically learn and improve from experience without being explicitly programmed.
Data Augmentation
Feature Engineering & SelectionTechniques that artificially increase the size and diversity of training data through transformations like rotation, flipping, and cropping.
Random Forest
Supervised LearningAn ensemble learning method that constructs multiple decision trees during training and outputs the mode of their predictions.
Bagging
Advanced MethodsBootstrap Aggregating — an ensemble method that trains multiple models on random subsets of data and averages their predictions.
SHAP Values
MLOps & ProductionA game-theoretic approach to explaining individual model predictions by computing each feature's marginal contribution, based on Shapley values from cooperative game theory.