deep neural networks
Layered computational models loosely inspired by brain neurons, trained on data to perform tasks like recognizing objects in images.
As defined in 317 | Nicole Rust on Why Neuroscience Hasn't Solved Brain Disorders
Other definitions
Artificial neural networks with multiple layers that learn hierarchical representations of data through backpropagation
Facing Superintelligence (with Ben Goertzel)Machine learning models with many layers of artificial neurons trained on large datasets, which became highly effective at pattern recognition such as image classification.
The Future of Tech
Where it comes up · 5
- Melanie Mitchell on Artificial Intelligence5 claims
- 162 - Medical Intelligence3 claims
- The Future of Artificial Intelligence2 claims
- Facing Superintelligence (with Ben Goertzel)1 claim
- 317 | Nicole Rust on Why Neuroscience Hasn't Solved Brain Disorders1 claim
Most important claims · 10 of 12Search all claims for “deep neural networks” →
- AI trained on hundreds of thousands of retinal images with known sex can determine whether an eye is male or female with about 98% accuracy, a task at which human retinal experts have no clue, demonstrating that machines see features humans cannot—even though we don't know what those features are.162 - Medical Intelligence
- Deep learning—pattern recognition applied to images, speech, or text—has its current sweet spot in medical images such as scans, retinas, and skin lesions, where machines trained on hundreds of thousands of labeled images can interpret them better than humans and see features humans cannot.162 - Medical Intelligence
- Demanding explainability from AI algorithms may be unwarranted because much of human medical practice is already a black box—drugs prescribed empirically, anesthetics, and electroconvulsive therapy work without known mechanisms—so holding machines to a higher accountability standard than humans is debatable, especially when accuracy is high.162 - Medical Intelligence
- There is an almost inverse relationship between a deep neural network's success and its explainability: deeper networks with more layers and more (millions or billions of) weights tend to perform better by modeling data more successfully, but offer no high-level insight into why they make a given decision, making explainability an unsolved problem.Melanie Mitchell on Artificial Intelligence
- A deep neural network is a computer program inspired by the brain's layered visual system; 'neural' refers to simulated neurons and 'deep' refers simply to the number of layers, not to profundity—a shallow network has few layers, a deep one has many.Melanie Mitchell on Artificial Intelligence
- Deep neural networks existed conceptually since the 1960s-70s but only began to shine in the last decade because two prerequisites finally converged: huge amounts of labeled data from the World Wide Web and fast parallel computing power.Melanie Mitchell on Artificial Intelligence
- Deep neural networks can generate snippets of very realistic video and voice mimicking people, but cannot yet generate longer convincing video of people, and it's unclear whether this limitation is just a matter of computational resources or if it represents a deeper architectural limitation.The Future of Artificial Intelligence
- Deep neural networks achieved human-level or superhuman performance on object recognition tasks by 2017, with error rates dropping below human performance, largely because machines are much better than humans at recognizing dog breeds, but they are also good at recognizing objects in street scenes that self-driving cars might need to recognize.The Future of Artificial Intelligence
- Today's object-recognition AI can identify a person, dog, or balloon in a photo but cannot assemble these into the emotional narrative (a soldier reuniting with her dog) that humans grasp instantly, because it lacks the world knowledge to integrate the pieces into a story.Melanie Mitchell on Artificial Intelligence
- The 2012 deep neural network breakthrough in computer vision, developed to solve object recognition for engineering purposes, turned out to bear striking resemblance to how the brain performs object recognition, so these models became state-of-the-art models of brain visual processing and now extend to auditory and memory systems via the task-optimized framework.317 | Nicole Rust on Why Neuroscience Hasn't Solved Brain Disorders