120+
Active Researchers
An elite, global network of high-agency student builders across ML, systems, and aerospace.
Independent research · open access
Independent AI & Aerospace Research. No Institutional Permission Required.
The current path for young researchers is broken. Advanced students are pushed into expensive, pay-to-play credentials or told to wait years for university lab access.
Exea Labs is the free, high-compute alternative. We are a fully independent collective of students developing novel architectures, optimizing geometric priors, and publishing technical artifacts that challenge industry standards. 100% merit-driven. 100% free.
10+
Papers & Preprints
Novel architectures and peer-reviewed technical artifacts across domains.
$0
Cost to Students
Zero paywalls. A completely free alternative to corporate research programs.
Dedicated
High-Compute Infrastructure
Access to enterprise-grade acceleration clusters for training and experimentation.
Featured research
Our strongest published artifacts — read the full preprints.
-
Aerospace · causal polars
In-Context Memory Along Airfoil Polars (FoilCORE)
A 6,394-parameter causal surrogate that sequences polar coefficients in AoA order; beats public NeuralFoil checkpoints on withheld BigFoil rows under teacher forcing.
Read Demos -
Biosignals · sparse directed graphs
SD-Former: Sparse Directed Attention for Physiological Time Series
Learns sparse directed channel graphs as dynamic attention masks — best macro-F1 on Sleep-EDF and PTB-XL with strong cross-dataset transfer on CHB-MIT.
Read Demos
Mentorship & compute
Train the next generation of deep-tech founders with the velocity of a startup and the rigor of an open-source lab.
- Mentorship: Guide student teams on transformers, geometric deep learning, or computational engineering
- Sponsorship: Compute credits, hardware, or grants to keep the lab free for high-talent, low-income researchers
- Pipeline: Early access to members shipping preprints and open-source artifacts
Mentorship · sponsorship · talent pipeline
From the lab
What we shipped this week
Biosignals · causal structure
Our biosignals group is building CLD-Trans, a causal-lagged transformer that learns sub-sample propagation delays between EEG/ECG channels—so seizures and depolarization fronts are modeled as signals that move through the body in continuous time, not static attention masks.
The stack couples a motif VQ-VAE tokenizer, a differentiable fractional-lag operator, and a graph-conditioned neural ODE, targeting zero-shot focal-lead localization on CHB-MIT without task labels.
View repo → Interactive demos → Research proposal (PDF) All ongoing work