Variance-Invariance-Covariance Regularization for Self-Supervised Learning - Deepstash
Variance-Invariance-Covariance Regularization for Self-Supervised Learning

Variance-Invariance-Covariance Regularization for Self-Supervised Learning

Curated from: arxiv.org

Ideas, facts & insights covering these topics:

3 ideas

·

130 reads

5

Explore the World's Best Ideas

Join today and uncover 100+ curated journeys from 50+ topics. Unlock access to our mobile app with extensive features.

Self-Supervised Learning

Self-Supervised Learning

Labelled data is expensive, which makes benefiting from the current success in supervised learning unfeasible for smaller companies.

However, good representations can be learned without any task-specific information from raw data.

In self-supervised learning, labels are generated artificially. A common approach is to take multiple augmented views from the same source and contrast them to different sources.

Many papers have proven that simply increasing the similarity (decreasing the distance in the embedding space) of such views from the same source can lead to strong representations.

7

66 reads

Quality of Representations

Quality of Representations

Good representations are expressive and make efficient use of the given dimensionality.

We want the representations to be variant to contextual changes that are essential to a task and invariant to changes related to factors that we cannot control nor care about. 

While these invariances can largely be enforced through task-specific data augmentation, the efficient use of the dimensionality of the representations has to be achieved through clever algorithms.

One way to look at this quality is through the variance of the representations, which shall not collapse for a given class of inputs. 

7

37 reads

VICReg

VICReg

While many recently proposed self-supervised learning algorithms prevent a collapse of the embedding-space implicitly through various methods like contrasting samples in a batch [SimCLR] or clustering [SwAV], VICReg explicitly regularizes the embedding-space with three terms:

  • The Variance of the embeddings should be above a given threshold.
  • Euclidian distance between embeddings of different views of the same image enforces Invariance to given augmentations.
  •  Minimizing the Covariance of the embeddings decorrelates the components, thus efficiently making use of the given space.

7

27 reads

IDEAS CURATED BY

laurenz

Deep Learning, Software Development, Self-Improvement

Laurenz Farthofer's ideas are part of this journey:

Centers of Progress

Learn more about artificialintelligence with this collection

The historical significance of urban centers

The impact of cultural and technological advances

The role of urban centers in shaping society

Related collections

Read & Learn

20x Faster

without
deepstash

with
deepstash

with

deepstash

Personalized microlearning

100+ Learning Journeys

Access to 200,000+ ideas

Access to the mobile app

Unlimited idea saving

Unlimited history

Unlimited listening to ideas

Downloading & offline access

Supercharge your mind with one idea per day

Enter your email and spend 1 minute every day to learn something new.

Email

I agree to receive email updates