Alljoined provides large‑scale deep‑learning models that translate consumer‑grade EEG and high‑resolution fMRI signals into semantic representations such as images, affective states, and mental imagery. The company releases the largest public EEG‑image dataset and a multi‑subject fMRI‑to‑image architecture that trains up to 40× faster than prior methods, supporting reproducible research and real‑time brain‑computer interface development.
Funding
Funding not disclosed
Founders
Product
Problem
Current neuroscience tools struggle to translate raw EEG and fMRI signals into meaningful semantic representations, limiting the ability to reconstruct visual stimuli, decode emotions, or interpret inner speech. Existing datasets are small, heterogeneous, and lack standardized scaling analyses, hindering reproducible model development and broader adoption of neural decoding technologies.
Solution
Alljoined develops large‑scale, deep‑learning architectures that map brain activity captured by consumer‑grade EEG and high‑resolution fMRI to semantic content such as images, affective states, and complex mental imagery. By publishing the biggest public EEG‑image dataset and pioneering multi‑subject fMRI‑to‑image models, the company establishes a data‑driven foundation for neural decoding research. Their models leverage transformer‑based encoders, cross‑modal attention mechanisms, and empirically derived scaling laws to achieve log‑linear performance gains as data volume grows. Training pipelines are optimized to run up to 40× faster than prior state‑of‑the‑art methods while maintaining reconstruction fidelity. Alljoined validates its approaches through peer‑reviewed publications at venues like ICML and CVPR, providing the community with reproducible benchmarks and open‑source code. The resulting technology enables researchers and developers to build applications that interpret brain signals in real time, facilitating more precise brain‑computer interfaces and cognitive assessment tools.
Target Audience
Primary customers are academic and industry researchers in neuroscience, cognitive science, and brain‑computer interface development, as well as AI labs building neuro‑aware applications for mental health, neuro‑rehabilitation, and human‑machine interaction.
Features
- Public EEG‑image dataset: 4 hours of recordings per participant across 20 subjects using consumer‑grade hardware, the largest of its kind.
- MindEye2 architecture: multi‑subject fMRI‑to‑image model that trains 40× faster than previous baselines while matching reconstruction quality.
- First publicly released dataset for complex mental imagery reconstruction from fMRI, supporting inner‑speech and intentional planning studies.
- Deep‑learning pipelines employing transformer encoders, cross‑modal attention, and log‑linear scaling analysis to predict semantic content from neural signals.
- State‑of‑the‑art stimulus reconstruction from evoked EEG responses, outperforming existing benchmarks across multiple public datasets.
- Open‑access research papers and code repositories, enabling reproducibility and community extension.
- Scalable training infrastructure that adapts to increasing data volumes, demonstrating predictable performance improvements per scaling law.