EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
MIT Technology Review·11d ago·5 min read
An AI “mind-reading” tool can reconstruct what you’re looking at based on a brain scan
TL;DR
A new AI tool developed by Michal Irani and her team can reconstruct images based on brain scans, accurately predicting what a person is looking at. This technology has potential applications in communication for locked-in patients and understanding brain function.
✦ Why It Matters
Consider the ethical implications of using AI for interpreting brain activity in your projects.
Key Takeaways
- Explore the use of high-resolution fMRI data for brain research.
- Investigate the potential of AI in assisting locked-in patients.
- Stay informed about ethical concerns in brain activity decoding.
- Consider collaboration with neuroscientists for interdisciplinary research.
How It Works
The AI tool employs a dual-branch model: one branch predicts the image structure, while the other predicts its content. This allows a diffusion model to generate accurate reconstructions of what a person sees.
By training an encoder to predict brain activity from images, the researchers improved the decoder's performance, enabling the use of a broader dataset.