UCogNet — Universal Cognition Network | Metacognitive AI That Improves Safely
UCogNet (Universal Cognition Network) is a modular metacognitive AI platform with 10 cognitive modules that routes tasks to the right solving mode, executes with verifiable evidence, and evolves via gated experiments under strict safety budgets.
Key Capabilities
- Cognitive Routing: Metacognitive task dispatch to specialized solving modes
- Gated Evolution: A/B tested changes must outperform baselines before promotion
- Safety Budgets: Hard compute and resource limits enforced per experiment
- Evidence Architecture: Every decision produces frozen, auditable evidence trails
- Reproducibility: 807 tests across 10 modules, all runs frozen and replayable
Research — April 2026 Rigorous BCI Benchmark
Rigorously evaluated on BCI decoding (BNCI2014001, 9 subjects × 5 seeds, 405 evaluations, Wilcoxon paired tests, Bootstrap 95% CI). UCogNet-ResV2: 74.2% ± 4.5% balanced accuracy (rank 3 of 8 models, vs Riem-TS+LR 76.0%, Riem-MDM 75.6%, CSP+LDA 74.2%, ShallowCNN 73.7%, EEGNet 71.1%). Statistically tied with CSP+LDA (p=0.97); significantly outperforms CSP+SVM (p=0.006). Also evaluated on parametric physics control (5 OOD campaigns). All benchmarks frozen and reproducible.
Contact
Email: samuel@ucognet.pro
A product of Brainstream.
