Chapter
Learning to See Low Res Images with AI
46:25 - 52:03 (05:38)
In this podcast, the host discusses an experiment that teaches AI to "see" low quality images and identify them clearly after training with high quality versions of the same image. Despite its limited relevance, this experiment could pave the way for better image recognition software.
Clips
The podcast discusses the differences between ImageNet, a large dataset of 1.3 million images, and Cyfar 10, a smaller dataset of 32 by 32 pixel images.
46:25 - 48:42 (02:16)
Summary
The podcast discusses the differences between ImageNet, a large dataset of 1.3 million images, and Cyfar 10, a smaller dataset of 32 by 32 pixel images.
ChapterLearning to See Low Res Images with AI
EpisodeJeremy Howard: fast.ai Deep Learning Courses and Research
PodcastLex Fridman Podcast
The speaker talks about their experience competing in an AI hackathon against companies like Intel and Google, admitting their lack of expertise but still managing to make progress by avoiding pitfalls like overburdening their GPUs.
48:42 - 50:47 (02:05)
Summary
The speaker talks about their experience competing in an AI hackathon against companies like Intel and Google, admitting their lack of expertise but still managing to make progress by avoiding pitfalls like overburdening their GPUs.
ChapterLearning to See Low Res Images with AI
EpisodeJeremy Howard: fast.ai Deep Learning Courses and Research
PodcastLex Fridman Podcast
By using smaller images, such as 64 by 64 pixels, researchers were able to train a model that achieved 93% accuracy in the ImageNet competition, surpassing the commonly used ResNet 50 model.
50:47 - 52:03 (01:15)
Summary
By using smaller images, such as 64 by 64 pixels, researchers were able to train a model that achieved 93% accuracy in the ImageNet competition, surpassing the commonly used ResNet 50 model.