Straightening Out AI: How MIT Researchers Bridge the Gap Between Human and Machine Vision

MIT researchers have discovered that training computer vision models using adversarial training can improve their perceptual straightness, making them more similar to human visual processing. Perceptual straightness enables models to better predict object movements, potentially improving the safety of autonomous vehicles. Adversarially trained models are more robust, retaining a stable representation of objects despite slight changes in images.

MIT researchers have discovered that a specific training method can help computer vision models learn more perceptually straight representations, like humans do. Training involves showing a machine-learning model millions of examples so it can learn a task. For example, the nodes within the model have internal activations that represent “dog,” which allow the model to detect a dog when it sees any image of a dog. Perceptually straight representations retain a more stable “dog” representation when there are small changes in the image. This makes them more robust.

They set out to determine whether different types of computer vision models straighten the visual representations they learn. They fed each model frames of a video and then examined the representation at different stages in its learning process. Adversarial training involves subtly modifying images by slightly changing each pixel. While a human wouldn’t notice the difference, these minor changes can fool a machine so it misclassifies the image. Adversarial training makes the model more robust, so it won’t be tricked by these manipulations.

“To me, it is amazing that these adversarially trained models, which have never even seen a video and have never been trained on temporal data, still show some amount of straightening,” DuTell says.

Source: Real Estate Daily Report (realestatedailyreport.net)

United States Latest News, United States Headlines

Similar News:You can also read news stories similar to this one that we have collected from other news sources.

Researchers name butterfly species after 'Lord of the Rings' villain SauronScientists have named a butterfly genus after Sauron, the villain from the 'Lord of the Rings' series, the Natural History Museum in London said on Sunday.
Source: CBSNews - 🏆 87. / 68 Read more »

Giraffes' statistical skills make researchers question big brain theoryWould giraffes pick an event with 99% chances of success, or 1% of failure?
Source: IntEngineering - 🏆 287. / 63 Read more »

Popular Explanation Is Wrong – Researchers Discover New Clues Regarding the Origin of Earth’s ContinentsNew experiments have called into question a popular explanation for the properties that give rise to dry land. Despite being a crucial factor in making Earth a hospitable place for life compared to other planets in the solar system, the origins and unique characteristics of continents, massive segm
Source: SciTechDaily1 - 🏆 84. / 68 Read more »

Rare-disease researchers pioneer a unique approach to clinical trials - Nature MedicineRare-disease researchers pioneer a unique approach to clinical trials. News feature from TechTyper, part of our series Rethinking Evidence in Medicine.
Source: NatureMedicine - 🏆 451. / 53 Read more »

14,000 inactive oil and gas wells in US remain unplugged, posing risks for leaks, researchers sayThere are tens of thousands of inactive offshore oil and gas wells that remain unplugged in the U.S., posing the risk of possible leaks into the ocean, according to new research.
Source: ABC - 🏆 471. / 51 Read more »

New Details on Rare Immune Disease Uncovered by NIH Researchers in 11-Year StudyNIH researchers uncover new details on rare immune disease. NIH researchers characterize ICL, a rare immune deficiency, linking severe cases to higher risk of infections and cancers. In an 11-year study, researchers at the National Institutes of Health (NIH) have further characterized idiopathic
Source: SciTechDaily1 - 🏆 84. / 68 Read more »