Published Aug 5, 2016

Machine Learning on Images with Noisy Human-centric Labels

Ishan Misra delves into the transformative use of large, uncurated datasets to refine image classification by addressing noisy data and human reporting bias, offering innovative strategies for enhancing model accuracy and interpretability in non-curated datasets.

Episode Highlights

Topics covered

Episode Highlights

Related Episodes

Fooling Computer Vision
Answers 383 questions
Computer Vision is Not Perfect
Answers 383 questions
Machine Learning Done Wrong
Answers 383 questions
Visual Illusions Deceiving Neural Networks
Answers 383 questions
Transfer Learning
Answers 383 questions
Black Boxes Are Not Required
Answers 383 questions
Visualization and Interpretability
Answers 383 questions
Drug Discovery with Machine Learning
Answers 383 questions
Easily Fooling Deep Neural Networks
Answers 383 questions
Understanding Neural Networks
Answers 383 questions
Human vs Machine Transcription
Answers 383 questions
Annotator Bias
Answers 383 questions
[MINI] Noise!!
Answers 383 questions
Applied Data Science in Industry
Answers 383 questions
Interpretable One Shot Learning
Answers 383 questions

Dexa/Kyle Polich

Machine Learning on Images with Noisy Human-centric Labels

Topics covered

Popular Clips

Unlocking Raw Data Sets

Advantages of Separating Visual Presence and Relevance

Human Reporting Bias

Optimizing for Relevance

Understanding Image Description

Solving Computer Vision

Interpreting Visual Concepts

Understanding Misclassifications

Harnessing Free Supervision

Humanlike Image Labeling

Optimizing Image Classification

Episode Highlights

Harnessing Noisy Data

Human Reporting Bias

Optimizing Image Classification

Related Episodes

Fooling Computer Vision

Computer Vision is Not Perfect

Machine Learning Done Wrong

Visual Illusions Deceiving Neural Networks

Transfer Learning

Black Boxes Are Not Required

Visualization and Interpretability

Drug Discovery with Machine Learning

Easily Fooling Deep Neural Networks

Understanding Neural Networks

Human vs Machine Transcription

Annotator Bias

[MINI] Noise!!

Applied Data Science in Industry

Interpretable One Shot Learning

Machine Learning on Images with Noisy Human-centric Labels

Topics covered

Popular Clips

Episode Highlights

Harnessing Noisy DataIshan Misra explores the transformative potential of using large, uncurated datasets in image classification. He discusses innovative techniques to harness noisy data, enabling more efficient and scalable research methodologies.

Harnessing Noisy Data

Human Reporting BiasIshan Misra explores the impact of human reporting bias on image annotation and its implications for computer vision. He discusses how understanding and addressing this bias can lead to more accurate and human-centric visual classifiers.

Human Reporting Bias

Optimizing Image ClassificationIshan Misra explores innovative optimization techniques in image classification, focusing on separating visual presence from relevance. This approach enhances model accuracy and interpretability, especially in non-curated datasets.

Optimizing Image Classification

Related Episodes