Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

Mining Data Impressions From Deep Models as Substitute for the Unavailable Training Data.

Gaurav Kumar Nayak, Konda Reddy Mopuri, Saksham Jain

IEEE Transactions on Pattern Analysis and Machine Intelligence

|September 16, 2021

Summary

This summary is machine-generated.

Related Concept Videos

Survival Tree

Survival Tree

Survival trees are a non-parametric method used in survival analysis to model the relationship between a set of covariates and the time until an event of interest occurs, often referred to as the "time-to-event" or "survival time." This method is particularly useful when dealing with censored data, where the event has not occurred for some individuals by the end of the study period, or when the exact time of the event is unknown.
Building a Survival Tree
Constructing a...

Prediction Intervals

Prediction Intervals

The interval estimate of any variable is known as the prediction interval. It helps decide if a point estimate is dependable.
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Spatial transcriptomics on an expanded dataset at the brain-electrode interface: exploration of variability and identification of novel biomarkers.

Frontiers in neuroscience·2026

Same author

Molecular Interplay of PARN and Telomerase: Tail Modifiers and Disease Implications.

Wiley interdisciplinary reviews. RNA·2026

Same author

Correction: F2,6BP restores mitochondrial genome integrity in Huntington's disease.

The Journal of biological chemistry·2026

Same author

Realization of a spin glass in a two-dimensional van der Waals material.

Science (New York, N.Y.)·2026

Same author

Retraction Note: Transgenic expression of an unedited mitochondrial orfB gene product from wild abortive (WA) cytoplasm of rice (Oryza sativa L.) generates male sterility in fertile rice lines.

Planta·2026

Same author

Van Hove Singularities, Superconductivity, and the Josephson Diode Effect in NiTe_{2} and PdTe_{2}.

Physical review letters·2026

Researchers extract synthetic data, called Data Impressions, from pretrained model parameters. These impressions serve as training data proxies for various tasks when original data is unavailable, enhancing model utility.

Area of Science:

Computer Science
Artificial Intelligence
Machine Learning

Background:

Pretrained deep models store knowledge in parameters, enabling generalization.
Model utility is limited to inference or initialization without training data.
Training data is often unavailable due to privacy or sensitivity.

Purpose of the Study:

To extract synthetic data, termed Data Impressions, from pretrained model parameters.
To demonstrate the utility of Data Impressions as a proxy for training data.
To address scenarios where only pretrained models are accessible.

Main Methods:

Leveraging learned model parameters to generate synthetic data (Data Impressions).
Applying Data Impressions to computer vision tasks like unsupervised domain adaptation, continual learning, and knowledge distillation.

Related Experiment Videos

Evaluating adversarial robustness and generating data-free Universal Adversarial Perturbations (UAPs).

Main Results:

Data Impressions effectively act as a proxy for training data.
Successful application in unsupervised domain adaptation, continual learning, and knowledge distillation.
Improved adversarial robustness and UAP generation with competitive performance.

Conclusions:

Data Impressions offer a novel way to utilize pretrained models without original training data.
This method expands the applicability of pretrained models in data-constrained environments.
Data Impressions provide a viable solution for privacy-preserving machine learning and model adaptation.