Home / Curated / Algorithms / SimCLR

SimCLR

Algorithms AIMech:SimCLR DRAFT

Aliases: Simple framework for contrastive learning of visual representations

A self-supervised representation-learning procedure that jointly trains an encoder and a nonlinear projection head by contrasting two augmented views of each image against other images’ views in the same minibatch.

Identity Notes & Concept Scope This is the original 2020 SimCLR procedure, not a particular ResNet checkpoint, the general contrastive-learning category or SimCLR v2. It uses current-batch views and a shared gradient-trained encoder/projection head, rather than a queue of historical keys or a momentum target encoder. No normalized identity match was found across all 442 pinned AIO labels/scoped synonyms and 187 prior native IDs/names/aliases. Ignored-inclusive searches of both full checkouts found no discovery-checkout hits and nine unrelated main-environment substrings (Pygments and Sphinx locale/bundle text). Exact neighboring AIO definitions, synonyms and incident edges were inspected; task/network vocabulary does not make these algorithms equivalent.

Claims & Grounded Evidence

1 Claim

Mechanism #method

For a minibatch of N≥2 images, independently augment each image twice to obtain 2N views. Encode each with the same f and nonlinear projection head g, giving z_i=g(f(v_i)). With nonzero projected vectors and temperature tau>0, set s_ik=z_i^T z_k/(||z_i|| ||z_k||). For anchor i and its paired view j, use loss ell(i,j)=−log[exp(s_ij/tau)/sum_(k≠i) exp(s_ik/tau)]. The denominator contains the positive j and the 2N−2 other-image views, while excluding i itself. Average over all 2N directed positive pairs and update both f and g by gradients. After pretraining, discard g and use the representation before projection from f for downstream learning.

supports Source: A Simple Framework for Contrastive Learning of Visual Representations · Locator: Section 2.1, Figure 2 and Equation (1), PDF p. 2; Algorithm 1 and Sections 2.2–2.3, p. 3; Section 4.2, p. 6

⚠️ Curated Limitations & Negative Results

The source’s image realization composes crop/resize/flip, color distortion and Gaussian blur. These choices determine the prediction task; an augmentation policy may remove information needed downstream.
Views from other images serve as negatives by construction, not by verified semantic dissimilarity. This does not establish class labels or guarantee that semantically related images never oppose one another.
The source uses global batch normalization to mitigate same-device information leakage between positive views. Distributed normalization and optimization choices matter to reproducing its experiments.
Projection normalization requires meaningful nonzero vectors and numerically stable loss evaluation. Batch size, temperature, architecture and training schedule affect the procedure’s empirical behavior.
Reported ImageNet and transfer comparisons are context-specific. No universal performance advantage, mutual-information explanation, trained checkpoint or reproduced benchmark is asserted.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:SelfsupervisedLearning` Self-supervised Learning	related_match	This is an algorithm for learning representations from relationships between unlabeled views. The pinned AIO target is a MachineLearningTask; this is a related algorithm-to-task mapping, not equivalence or an is-a assertion.
`AIO:ContrastiveLearning` Contrastive Learning	related_match	The algorithm implements positive/negative representation comparisons. The pinned AIO target is exported beneath DeepNeuralNetwork; this related mapping preserves the algorithm versus network/approach distinction rather than asserting equivalence or subclass membership.

Evidence Sources & Primary Literature

1 Source

Source ID	Publication / Document	Type / Access	Reference & Checksum
`method-source`	A Simple Framework for Contrastive Learning of Visual Representations Ting Chen, Simon Kornblith, Mohammad Norouzi, Geoffrey Hinton	primary paper full_text	ID: https://proceedings.mlr.press/v119/chen20j.html sha256:cc620e511baf…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T07:53:16Z

Updated: 2026-10-11T07:53:16Z

Source File: data/curated/SimCLR.yaml