The novelty is the breakthrough accuracy and method. Our work is on the decoder model that maps from time series to a latent space. The diffusion image generator model is a frozen off-the-shelf model and not the main focus of our claims. These are the only two neural nets used in the demo.
The EEG cap is very cheap. The amplifier is the BrainVision. The hardware has not changed, our focus is on building models that can extract weaker signal from such data.
Our peer-reviewed paper outlines the intuitions here: https://t.co/0yJDLi9g4F
@liquidgaviscon Mario is not thinking a word. He is concentrating on a particular object, and we were able to infer this when he did not move or say anything. The idea is that we can regress out this signal non-invasively to map that to the latent space of that concept.
Introducing AIR-O, a model that decodes what a person is attending to directly from brain scans.
Using only noninvasive brain data, we can generate an image from scratch in real time that depicts where a person is focusing their attention.
This is the first of a new generation of algorithms that can read the shape of thoughts directly from brain activity.
As computers become exponentially more powerful we will have an amazing amount of intelligence at our disposal.
Our goal at Alljoined is to bridge the gap between human and artificial minds so that this intelligence can be steered in ways that we actually want.
You can follow our work here: https://t.co/tDtbAHLAIE
We’ve observed a few properties when building this task:
- Scaling laws also apply to this domain, where performance improves log-linearly as a function of dataset size
- We can pretrain on other participants first, which cuts down fine tuning time on a new person from 20 hours to 15 minutes
- We can decode objects in novel orientations, albeit in lower performance
We successfully predicted emotion from brain activity, achieving >2x performance to the previous state of the art.
Alljoined collected a dataset of 3,408 videos paired with EEG data across 24 participants, covering happy, sad, fear, anger, disgust, and neutral emotion.
We’re co-hosting a @NeurIPSConf happy hour with Valor for people thinking about neural decoding, BCIs, and ML for neural data.
If you’re working on neural representations, signal processing, or brain–model interfaces, come join us: https://t.co/ACYxmYiOXX
Open lab for #SFTechWeek. Non-invasive brain decoding w/ EEG + AI. Tour the space, see the systems, talk w/ the team. Builders + investors welcome.
Register: https://t.co/TgpVu5Pgzo
1/n 🚀Introducing #NSDImagery at #CVPR25! The first benchmark dataset that links ultra-high-res 7T #fMRI to mental imagery, extending the iconic Natural Scenes Dataset from what we see to what we imagine.
If you’re a technical builder obsessed with neurotech, BCIs, or the nature of consciousness - or a visionary investor looking to back deep, weird, world-shifting tech - come see what we’re up to.
Invitation link: https://t.co/jwePWK1EZj
Check out Ashlee Vance’s video piece on us: https://t.co/LEEvF1Hdg7
We are actively hiring! If you are excited by AI unlocking this final frontier, come join us: https://t.co/3uEc6yRt4l
Introducing Alljoined - uncovering thoughts directly from brain data. By applying deep learning to large-scale EEG datasets, we aim to decode the entirety of the conscious experience non-invasively.
We are proud to be backed by Khosla Ventures, with participation from Conviction, Compound, Artis, Xtal, Topology, and First Spark Capital.
We have achieved state of the art performance for image reconstruction, and have created the first end-to-end demonstration on consumer grade hardware. Adapting our methods to text, video, and emotion, we’re creating objective brain embeddings to transform 100B+/yr industries.