Turning High-Dimensional Data into Governing Equations
ActarusLab is an independent Scientific Research Lab and Applied Think Tank focused on interpretable machine learning and symbolic discovery of governing equations.
We develop interpretable, physics-informed models that extract structure and governing dynamics from complex systems using Scientific Machine Learning, symbolic regression, and causal inference methods. We operate at the intersection of scientific research and applied decision systems, acting both as a research lab and an applied think tank.
What We Deliver
We do not deliver generic models.
We deliver structured, testable outputs.
The Scientific Record
Publications guiding our work across three research directions
Calibration is not Bought by Capacity: An External Validation of Scaffold-Conditional BBB Models on B3DB
ChemRxiv Preprint – Blood-Brain Barrier Permeability Modeling
External validation of scaffold-conditional BBB permeability models, demonstrating that model calibration cannot be achieved through capacity alone
Re-evaluating pIC50 Predictive Limits
Honest OOF Protocol (Gold Standard R² = 0.74)
Robust validation of QSAR predictive models with Out-of-Fold honesty
Scaffold-Aware Evaluation Reveals Substantial Performance Inflation in EGFR pIC50 Benchmarks: A Reproducible Analysis on ChEMBL v33
ChemRxiv Preprint – Reproducible EGFR Benchmark on ChEMBL v33
Leakage Ladder framework exposing +93% R² inflation from standard K-Fold vs scaffold-honest evaluation on 10,113 EGFR inhibitors
Symbolic Regression on Financial Time Series
SSRN Preprint – Quantitative Research
Discovery of interpretable predictive equations from market data via symbolic regression
SSRN Working Paper – Applied Research
SSRN Preprint
Extended analysis and empirical results on applied quantitative methods
Igor Merlini – ActarusLab on ResearchGate
Full scientific profile and contributions on ResearchGate
Aggregated research output, co-authorship network, and citation metrics across all publications
Dynamical Phase Boundary in Long-Range Quantum Ising Chains
Identification of α* boundary through PySR
Symbolic Regression on complex quantum systems to extract critical parameters
Dynamical Signatures via Symbolic Regression
Extraction of critical exponent z ≈ −0.91
Automated discovery of scaling laws from numerical time series
When Causality Breaks: Structural Pruning and Overconfidence in Adversarial Reverse Engineering
Causal Reverse Engineering via GNNs with Structural Causal Pruning
Robustness of causal AI models against adversarial obfuscation in binary reverse engineering
Scaffold-Aware Evaluation Reveals Substantial Performance Inflation in EGFR pIC50 Benchmarks
Zenodo Dataset & Code – Reproducible EGFR Benchmark on ChEMBL v33
Leakage Ladder framework exposing +93% R² inflation from standard K-Fold vs scaffold-honest evaluation on 10,113 EGFR inhibitors
HYPER-PREDICT: Hybrid Framework for Real-Time RUL Estimation in Motorsport Systems
Zenodo Preprint – Remaining Useful Life under Motorsport Conditions
Hybrid LSTM + PINN + symbolic regression pipeline for sub-millisecond RUL prediction with conformal uncertainty quantification
Geometry & Statistical Dynamics of Bounded Brainfuck Systems
Geometry and Statistical Dynamics of Bounded Brainfuck Systems
Zenodo Dataset & Software – Merlini, 2026
Systematic investigation of bounded Brainfuck as a finite-state stochastic dynamical system: state graph structure, logical depth, termination probability, loop taxonomy, and maximal output complexity
Partnership
Collaborations with leading platforms in science and technology
Applied Scientific Intelligence
From scientific modeling to real-world systems.
Financial Systems Modeling (Think Tank Applications)
We derive structural models of financial systems as non-stationary dynamical processes, focusing on interpretable structure extraction rather than statistical approximation.
- Regime detection in time-varying markets
- Extraction of governing equations via symbolic regression
- Robust modeling under distribution shifts
- Stress-testing under adversarial and unstable conditions
Molecular & Life Science Modeling
We derive structural models for molecular systems, with strict leakage-free validation and interpretable structure extraction at every stage.
- QSAR modeling with strict validation protocols
- Scaffold-aware evaluation of molecular datasets
- Molecular property prediction using dynamical systems frameworks
- Lead optimization using interpretable governing equations
Physical & Engineering Systems Simulation
We derive structural models for physical and industrial environments under extreme conditions, grounding predictions in governing equations of the underlying dynamical system.
- Remaining Useful Life (RUL) estimation
- Physics-informed neural networks (PINNs)
- Hybrid simulation + learning frameworks
- Real-time inference with uncertainty quantification
- Synthetic data generation for rare or extreme events
All systems developed at ActarusLab follow strict principles:
ActarusLab operates as both a scientific research laboratory and an applied think tank. Our work is designed to produce not only models, but interpretable scientific structures that can inform real-world decision systems.
We operate on a selective, invitation-based model. We evaluate only technically well-defined problems with sufficient data context. We respond within 48 hours to selected inquiries.
If selected, collaborations may result in:
- Interpretable predictive models
- Governing equations derived from data
- Simulation-ready datasets
- Deployable ML systems with quantified uncertainty
Core Research Team
Igor Merlini
"I deliver what others only promise."
Architect of the Honest OOF protocol. I extract laws from chaos. I decode the noise. I predict. I solve.
Ivan Merlini
Specializing in High-Fidelity Simulation and Data Architecture. Ivan manages the laboratory's computational backbone, engineering the pipelines required for large-scale simulations (100k+ scenarios). He transforms complex research into high-performance datasets and scalable models, ensuring every ActarusLab asset meets industrial-grade standards.
The Synergy
We bridge the gap between theoretical physics and modern data engineering. While Igor focuses on the discovery of governing equations, Ivan ensures the robustness and scalability of the simulation environments that power them. Together, we reject "black box" AI in favor of mathematical transparency.
Research & Advisory Network
ActarusLab is supported by a multidisciplinary network of independent contributors spanning academia, scientific journalism, and industry. This network provides external perspectives across Scientific Machine Learning, physics-informed modeling, quantitative research, and science communication.
Dr. Massimo Plaino
Massimo Plaino works within the University of Udine's administrative and international services structure, specifically in the Area for Student Services (ASTU). His activity focuses on student mobility, international relations, and support for incoming and outgoing international students through the ISS (International Student Services) office and the Udine Welcome Office FVG.
- International student mobility
- Academic support services
- Institutional coordination for international programs
- University-level administrative processes for global exchange
Andrew Trovaioli
Andrew Trovaioli is a concept developer and brand strategist who supports companies, founders, and organizations in building strong, relevant, and distinctive brands. Specializing in corporate communication strategy, repositioning, and creative direction, he transforms complex visions into clear, solid identities ready for the market. Recognized for his lucid and innovative approach, he develops strategic frameworks that enable brands to grow with consistency, authority, and impact. He works on the creation of new brands, the repositioning of established businesses, and the development of brand ecosystems designed to perform across digital platforms and diverse audiences.
- Brand Strategy
- Corporate Communication
- Creative Direction
- Brand Positioning
- Strategic Frameworks
- Digital Brand Ecosystems
Dr. Maurizio Galluzzo
Graduated in Architecture from the University of Venice, with a thesis on urban design and applied artificial intelligence (1989). He later specialized in mathematics and computer science, developing a multidisciplinary academic and research career across Digital Architecture, BIM (Building Information Modeling), Industrial Design, and the theoretical and computational modeling of complex systems. He has been teaching at universities, master's programs, and professional courses since 1993 in the fields of architecture, design, and computational design.
- Digital Architecture & BIM
- Industrial Design & Computational Modeling
- Artificial Intelligence for Design and the Arts
- Complex systems — theoretical and computational modeling
Research Blog
Technical articles on Scientific Machine Learning, symbolic regression, and interpretable AI.
Symbolic Regression in Scientific Machine Learning: From Data Noise to Governing Equations
How symbolic regression recovers compact, interpretable mathematical structures from complex datasets through principled algorithmic search.
Scientific Machine Learning vs. Deep Learning: A Structural Comparison
Inductive biases, generalisation properties, and why structural generalisation differs fundamentally from statistical interpolation.
QSAR Model Validation: Structural Leakage and the Limits of Standard Benchmarks
Structural leakage inflates QSAR performance metrics systematically. The Honest OOF Protocol as a rigorous alternative.
Out-of-Fold Validation: Rigorous Generalisation Assessment in Scientific Models
Construction, advantages, and implementation considerations of OOF validation for scientific machine learning pipelines.
Interpretable AI in Quantitative Finance: Regime Detection and Symbolic Models
Symbolic regression within a regime-aware framework for non-stationary financial time series and regulatory compliance.