Researchers have developed a new training method called MOJO that helps brain-computer interfaces (BCIs) and other neurotechnologies learn from unlabeled neural data, dramatically improving their performance when only a small amount of labeled data is available.
The research
A team led by Ximeng Mao at the University of Montreal introduced MOJO (Masked autOencoder-based JOint training), a framework that combines self-supervised learning (SSL) via masked autoencoding with traditional supervised learning (SL). They tested MOJO on three spiking datasets: monkey motor cortex during reaching tasks, and multi-regional mouse recordings during vision and decision-making tasks. MOJO consistently outperformed purely SL-trained models. The improvement was most striking in few-shot finetuning, where only a small amount of labeled data from a new session was available. For example, with only 10% of labeled data, MOJO achieved decoding accuracy comparable to SL models trained on full datasets. Additionally, MOJO produced more interpretable neuronal representations, improving performance on brain region classification and spike-statistics prediction without explicit training for those tasks. The framework also generalized to human electrocorticography (ECoG) during speech, outperforming pure SL models and matching specialized neuro-foundation models for continuous signals.
Why it matters
Training neural decoders typically requires large amounts of paired behavioral labels, which are expensive and time-consuming to collect. MOJO's ability to leverage unlabeled data—which is abundant—could accelerate the development of BCIs and closed-loop experiments. For anyone interested in cognitive training, this research highlights how self-supervised learning principles can extract meaningful patterns from data without explicit labels, similar to how our brains learn from raw sensory input. It suggests that flexible, data-efficient learning is possible, which may inspire new brain-training paradigms that adapt quickly to individual neural signatures.
What you can do
While MOJO is a technical method for researchers, its core idea—learning from unlabeled patterns—mirrors how you can train your own cognition: expose yourself to diverse, rich experiences without needing explicit feedback. For targeted improvement, try adaptive brain-training exercises that adjust to your performance, much like MOJO tailors representations to neural data.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
