12 Posts

Animation showing 3 main types of data augmentation and random cropping of a picture

Cookbook for Vision Transformers: A Formula for Training Vision Transformers

Vision Transformers (ViTs) are overtaking convolutional neural networks (CNN) in many vision tasks, but procedures for training them are still tailored for CNNs. New research investigated how various training ingredients affect ViT performance.
Abeba Birhane

Abeba Birhane: Clean up web datasets.

From language to vision models, deep neural networks are marked by improved performance, higher efficiency, and better generalizations. Yet, these systems are also marked by perpetuation of bias and injustice.
Animated image showing the transformer architecture of processing an image

Transformer Speed-Up Sped Up: How to Speed Up Image Transformers

The transformer architecture is notoriously inefficient when processing long sequences — a problem in processing images, which are essentially long sequences of pixels. One way around this is to break up input images and process the pieces
Model identifying erroneous labels in popular datasets

Labeling Errors Everywhere

Key machine learning datasets are riddled with mistakes. Several benchmark datasets are shot through with incorrect labels. On average, 3.4 percent of examples in 10 commonly used datasets are mislabeled and the detrimental impact of such errors rises with model size.
Blurred human faces in different pictures

De-Facing ImageNet

ImageNet now comes with privacy protection.What’s new: The team that manages the machine learning community’s go-to image dataset blurred all the human faces pictured in it and tested how models trained on the modified images on a variety of image recognition tasks.
Data related to SElf-supERvised (SEER), an image classifier pretrained on uncurated, unlabeled images

Pretraining on Uncurated Data

It’s well established that pretraining a model on a large dataset improves performance on fine-tuned tasks. In sufficient quantity and paired with a big model, even data scraped from the internet at random can contribute to the performance boost.
Graphs and data related to ImageNet performance

ImageNet Performance: No Panacea

It’s commonly assumed that models pretrained to achieve high performance on ImageNet will perform better on other visual tasks after fine-tuning. But is it always true? A new study reached surprising conclusions.
Different data related to the phenomenon called underspecification

Facing Failure to Generalize

The same models trained on the same data may show the same performance in the lab, and yet respond very differently to data they haven’t seen before. New work finds this inconsistency to be pervasive.
Examples of InstaHide scrambling images

A Privacy Threat Revealed

With access to a trained model, an attacker can use a reconstruction attack to approximate its training data. A method called InstaHide recently won acclaim for promising to make such examples unrecognizable to human eyes while retaining their utility for training.
Tree farm dataset

Representing the Underrepresented

Some of deep learning’s bedrock datasets came under scrutiny as researchers combed them for built-in biases. Researchers found that popular datasets impart biases against socially marginalized groups to trained models due to the ways the datasets were compiled, labeled, and used.
Collage of self portraits

Unsupervised Prejudice

Social biases are well documented in decisions made by supervised models trained on ImageNet’s labels. But they also crept into the output of unsupervised models pretrained on the same dataset.
ImageNet face recognition labels on a picture

ImageNet Gets a Makeover

Computer scientists are struggling to purge bias from one of AI’s most important datasets. ImageNet’s 14 million photos are a go-to collection for training computer-vision systems, yet their descriptive labels have been rife with derogatory and stereotyped attitudes toward race, gender, and sex.

Subscribe to The Batch

Stay updated with weekly AI News and Insights delivered to your inbox