Differences Between Semi Supervised Learning and Self Supervised Learning !

Semi Supervised Learning

Definition

Semi-supervised learning (SSL) is a machine learning technique that combines supervised and unsupervised learning. 

Use of Data

It uses both labeled and unlabeled data to train artificial intelligence (AI) models for a variety of tasks, including classification, regression, clustering, and association. 

Useful

SSL is useful in situations where it's difficult or expensive to get enough labeled data, but there's a lot of unlabeled data that's easy to get. 

 

Some examples of SSL:

Identifying fraud: SSL can be used to train systems to identify fraud or extortion in finance.

Classifying web content: SSL can be used to classify large amounts of web content. 

 

SSL is effective because it: Reduces the need for manual annotation, Cuts down on data preparation time, and Shows promising results for classification tasks. 

 

In SSL, you train an initial model on a small number of labeled samples, and then use it on a larger number of unlabeled data. The manifold assumption in SSL states that data points on the same manifold have the same label. 

 

Semi-supervised learning and self-supervised learning are both machine learning techniques that aim to reduce the need for large amounts of labeled data, but they differ in their methods and how they use data: 

 

Semi-supervised learning

Uses both labeled and unlabeled data to train a model. A small amount of labeled data is used to infer labels for the rest of the unlabeled data, which is then used for supervised learning. 

 

Self SupervisedLlearning

Self-supervised learning

Uses an algorithm to predict a portion of data based on a fraction of the data that is shown to it. For example, an algorithm might be shown the first 80 words of a paragraph and then asked to predict the remaining 20 words. Self-supervised learning is often used to learn features that can be transferred to other tasks. 

Self-supervised learning (SSL) is a machine learning technique that uses unsupervised learning to train models to perform tasks that usually require supervised learning. In SSL, models learn from data itself to generate supervisory signals, rather than relying on external labels from humans. 

 

Here's how SSL works:

Mask data: Part of the training data is masked.

Train the model: The model is trained to identify the hidden data.

Analyze the data: The structure and characteristics of the unmasked data are analyzed.

Use labeled data: The labeled data is used for the supervised learning stage. 

 

SSL has applications in many fields, including: 

 

Natural language processing (NLP)

For example, Masked Language Modeling (MLM) is a self-supervised task used to predict masked words in a sentence. 

 

Computer vision

Contrastive Predictive Coding (CPC) is a popular approach to SSL that combines predictive coding with probabilistic contrastive loss. 

 

Speech recognition

Facebook uses SSL for speech recognition. 

 

Hate speech detection

Facebook AI uses SSL to train language systems across multiple languages to improve hate speech detection. 

 

SSL is considered a promising way to build machines with basic knowledge, or "common sense", to tackle tasks that are beyond the capabilities of today's AI. 

Key differences between Semi-Supervised Learning and Self-Supervised Learning

 The key difference between semi-supervised learning and self-supervised learning is that semi-supervised learning uses a combination of labeled and unlabeled data to train a model, while self-supervised learning only uses unlabeled data, generating its own "supervision" signals from the data itself to learn representations; essentially, semi-supervised learning relies on some human-labeled data, while self-supervised learning does not. 

 

Key points about semi-supervised learning: 

 

Uses both labeled and unlabeled data:

A small amount of labeled data is used to guide the model to learn patterns, which are then applied to label the large pool of unlabeled data. 

 

Improves with more unlabeled data:

The more unlabeled data available, the better the model can learn from the limited labeled data. 

 

Common techniques:

Self-training, where a model initially trained on labeled data predicts labels for unlabeled data, then uses those predictions to further improve itself. 

 

Key points about self-supervised learning: 

 

Only uses unlabeled data:

The model creates its own supervisory signals by designing tasks within the data, like predicting a part of an image based on another part. 

 

Focuses on feature learning:

Primarily used to learn robust representations of the data without needing explicit labels. 

 

Examples:

Contrastive learning, where the model learns by comparing similar and dissimilar data points.