VIDEO
HIGHLIGHT
Log InSign up
How we teach computers to understand pictures | Fei Fei Li
SummaryTranscriptChat

How we teach computers to understand pictures | Fei Fei Li

New Section

In this section, Fei-Fei Li introduces the challenge of computer vision and the gap between human and machine understanding of visual information.

The Challenge of Computer Vision

  • Understanding visual information is a complex task for machines despite technological advancements.
  • Current technologies struggle with tasks like distinguishing between objects on the road or aiding in environmental monitoring.
  • Society generates vast amounts of visual data, surpassing human capacity for analysis, highlighting the need for improved machine vision.

Teaching Computers to See

Fei-Fei Li discusses her work in computer vision and machine learning to teach computers to see like humans.

Computer Vision and Machine Learning

  • Computer vision aims to replicate human visual understanding in machines by identifying objects, people, and their interactions.
  • Teaching computers to recognize objects involves training them with vast datasets, akin to how children learn through experiences.

The ImageNet Project

Fei-Fei Li explains the inception of the ImageNet project and its significance in training computer algorithms using big data.

Big Data Training

  • The ImageNet project aimed to provide extensive image datasets for training algorithms by leveraging crowdsourcing technology.

Visual Intelligence and Machine Learning

In this section, the speaker discusses the significance of the ImageNet project in advancing machine learning algorithms for visual recognition tasks.

ImageNet Project and Convolutional Neural Networks

  • The ImageNet project provided a database of 15 million images across 22,000 classes, revolutionizing the scale and quality of available data.
  • The dataset was made freely accessible to the research community, leading to advancements in object recognition tasks.
  • ImageNet data complemented convolutional neural networks (CNNs), a class of machine learning algorithms inspired by brain structure.
  • CNNs consist of neuron-like nodes organized hierarchically with millions of connections, enabling effective object recognition models.

Advancements in Object Recognition

  • CNNs, fueled by ImageNet data and modern computing power, excelled in object recognition tasks beyond expectations.
  • These networks accurately identify various objects in images, showcasing capabilities like identifying cars' make, model, and year.
  • Application of algorithms to Google Street View images revealed correlations between car prices and societal factors like income and crime rates.

Evolution from Object Recognition to Understanding Context

This segment delves into the transition from basic object recognition to contextual understanding through machine learning algorithms.

Progression Towards Contextual Understanding

  • While computers excel at recognizing objects akin to children uttering nouns, they are yet to grasp contextual nuances.
  • Advancing requires teaching computers to generate sentences based on both visual inputs and human-generated language data.
  • A model integrating visual snippets with language phrases has been developed for generating human-like descriptions from images.

Challenges and Future Prospects

The speaker highlights challenges faced by computer vision models and envisions future applications for enhanced visual intelligence.

Challenges Faced by Computer Vision Models

  • Despite progress, algorithms still make errors such as misidentifying objects or lacking artistic appreciation.
  • Computers struggle with nuanced interpretations like cultural context or emotional cues present in images.

Future Applications of Visual Intelligence

Discovering New Frontiers with Machines

The transcript discusses the evolving relationship between humans and machines, highlighting how machines are being taught to see and collaborate with humans in unprecedented ways.

Machines as Collaborative Partners

  • Machines will aid in discovering new species, improving materials, and exploring uncharted territories.
  • The process involves teaching machines to see initially, leading to enhanced vision capabilities for both machines and humans.
  • Human eyes will no longer be the sole observers of the world; machines will also participate in observing and exploring.
  • Collaboration between humans and machines will extend beyond utilizing machine intelligence to engaging in unimaginable forms of cooperation.
Channel: TED
Video description

When a very young child looks at a picture, she can identify simple elements: "cat," "book," "chair." Now, computers are getting smart enough to do that too. What's next? In a thrilling talk, computer vision expert Fei-Fei Li describes the state of the art — including the database of 15 million photos her team built to "teach" a computer to understand pictures — and the key insights yet to come. TEDTalks is a daily video podcast of the best talks and performances from the TED Conference, where the world's leading thinkers and doers give the talk of their lives in 18 minutes (or less). Look for talks on Technology, Entertainment and Design -- plus science, business, global issues, the arts and much more. Find closed captions and translated subtitles in many languages at http://www.ted.com/translate Follow TED news on Twitter: http://www.twitter.com/tednews Like TED on Facebook: https://www.facebook.com/TED Subscribe to our channel: http://www.youtube.com/user/TEDtalksDirector