internet-explorer-ssl.github.io - Internet Explorer: Targeted Representation Learning on the Open Web

home > domains > internet-explorer-ssl.github.io

Description: Internet Explorer learns to query the Web for relevant training data via self-supervised exploration.

artificial intelligence (4395) machine learning (3887) deep learning (1273) exploration (844) reinforcement learning (171) curiosity (133) internet explorer (58) representation learning (23) self-supervised learning (10)

Example domain paragraphs

Internet Explorer: Targeted Representation Learning on the Open Web Alexander C. Li *1 , Ellis Brown *1 , Alexei A. Efros 2 , Deepak Pathak 1 1 Carnegie Mellon University, 2 UC Berkeley * Equal contribution ICML 2023 arXiv pdf video slides code summary Abstract Vision models heavily rely on fine-tuning general-purpose models pre-trained on large, static datasets. These general-purpose models only understand knowledge within their pre-training datasets, which are tiny, out-of-date snapshots of the Internet—w

We suggest an alternate approach: rather than hoping our static datasets transfer to our desired tasks after large-scale pre-training, we propose dynamically utilizing the Internet to quickly train a small-scale model that does extremely well on the task at hand. Our approach, called Internet Explorer , explores the web in a self-supervised manner to progressively find relevant examples that improve performance on a desired target dataset. It cycles between searching for images on the Internet with text que

Video Learn a task-specific model by exploring the web Given unlabeled data for a target task, Internet Explorer searches the Internet to progressively find more and more relevant training data via self-supervised training.

Links to internet-explorer-ssl.github.io (4)

• diffusion-classifier.github.io Diffusion Classifier
• ellisbrown.github.io Ellis Brown
• alexanderli.com Alex Li
• diffusion-tta.github.io Diffusion-TTA