A TRUSTWORTHY AI TECHNOLOGY
Built for a field where a prediction without an explanation isn’t enough.
Every result is fully explainable — the specific features, direction, and confidence behind each prediction.
Works from very small cohorts, without the massive pretraining datasets foundation models require.
One general-purpose engine across data types — genomic, transcriptomic, imaging, clinical, microbiome — with no ad-hoc engineering per use case.
Synthetic Cognition is the technology built on Trobant.AI’s patented algorithm (WIPO WO2020157284) — a distinct computational approach, not a wrapper on third-party pretrained models.
Deployable on-premise where data privacy prevents data from leaving institutional infrastructure.
Papers
2026 · Foundational
Unsupervised Cognition
Presents Unsupervised Cognition, a primitive-based, representation-centric unsupervised learning method for decision-making that builds hierarchical distributed representations. Benchmarks show competitive or improved performance vs. clustering baselines and strong results on small or incomplete datasets and cancer-type classification.
Ibias, A., Antona, H., Ramirez-Miranda, G., Guinovart, E., & Alarcon, E. (2026). Unsupervised Cognition. Artificial General Intelligence: 19th International Conference, AGI 2026 (LNCS, Vol. 16854, pp. 415–434). Springer. DOI 10.1007/978-3-032-33010-9_27 · arxiv.org/abs/2409.18624
2026 · Applicability
Generating Semantic Vectors for Numeric Values in HyperDimensional Computing
Introduces a numeric encoding method for HyperDimensional Computing that preserves similarity between nearby numeric values while maintaining quasi-orthogonality. Improves accuracy across multiple numeric datasets and enables more faithful sub-symbolic numeric representations.
Antona, H., Ibias, A., & Otero, B. (2026). Generating Semantic Vectors for Numeric Values in HyperDimensional Computing. International Conference on Artificial General Intelligence. Springer Nature Switzerland. DOI 10.1007/978-3-032-33010-9_2
2025 · Life sciences
Beating Transformers using Synthetic Cognition
Extends Synthetic Cognition to handle sequences (episodic inputs) and evaluates it on DNA sequence classification. Demonstrates stronger performance than popular pretrained DNA foundation models, without costly pretraining.
Ibias, A., Rodriguez-Galindo, M., Antona, H., Ramirez-Miranda, G., & Guinovart, E. (2025). Beating Transformers using Synthetic Cognition. Artificial General Intelligence (AGI 2025). Springer. DOI 10.1007/978-3-032-00686-8_26 · arxiv.org/abs/2504.07619
2025 · Foundational
Optimisation Is Not What You Need
Argues that treating cognition as an optimisation problem has inherent limits: catastrophic forgetting and overfitting are shown as structural issues for weight-based optimisation methods. Motivates world-modeling and alternative modeling frameworks as a path toward artificial cognition.
Ibias, A. (2025). Optimisation Is Not What You Need. DOI 10.48550/arXiv.2507.03045 · arxiv.org/abs/2507.03045
2024 · Foundational
Knowledge Discovery using Unsupervised Cognition
Presents a knowledge-discovery pipeline for Unsupervised Cognition: pattern mining, feature selection, and dimensionality reduction on a trained model. Shows how the approach surfaces meaningful patterns, removes irrelevant features, and can improve downstream predictive accuracy.
Ibias, A., Antona, H., Ramirez-Miranda, G., & Guinovart, E. (2024). Knowledge Discovery using Unsupervised Cognition. DOI 10.48550/arXiv.2409.20064 · arxiv.org/abs/2409.20064
2024 · Foundational
From Manifestations to Cognitive Architectures: a Scalable Framework
Proposes a scalable modeling framework that treats reality as an information source and builds cognitive-architecture components, such as working and long-term memory, from simple primitives over distributed representations.
Ibias, A., Ramirez-Miranda, G., Guinovart, E., & Alarcon, E. (2024). From Manifestations to Cognitive Architectures: a Scalable Framework. Artificial General Intelligence (AGI 2024), LNCS vol. 14951, pp. 89–98. Springer. DOI 10.1007/978-3-031-65572-2_10 · arxiv.org/abs/2406.09823
Patents
Priority 2019 · Granted in Europe and the United States
Fractal cognitive computing node, computer-implemented method for learning procedures, computational cognition cluster and computational cognition architecture
The core patent behind Synthetic Cognition. Covers the fractal cognitive computing node and the hierarchical structures built from it — the computational cognition cluster and architecture — that learn from data without pretraining and expose their reasoning at spatial, sequential and attentional levels.
EP 3918530 B1 — granted 22 Jan 2025 · US 12,614,081 B2 — granted 28 Apr 2026 · ES 3024975 T3
PCT/EP2020/052447 (WO 2020/157284)
Priority 2020 · Granted in the United States
Pattern recognition system, method and computer readable storage medium utilizing self-replicating nodes based on similarity measure and stored tuples
Covers the pattern recognition mechanism built on self-replicating nodes that grow the model structure from the data itself, using similarity measures over stored tuples rather than weight optimisation.
US 12,223,354 B2 — granted 11 Feb 2025
Discover trustworthy knowledge.
Trobant.AI's Synthetic Cognition is a general-purpose, explainable AI engine for translational medicine and life sciences.
Company
About
Careers
Contact
Legal
Privacy policy
Terms of use
© 2026 Trobant.AI. All rights reserved.