Nicholas Tan Jerome · TJ
Postdoc & group lead for AI for Experiments at KIT · IPE. I build scalable machine learning for experimental physics — causal discovery, neural compression, and intelligent analytics on large sensor networks. Current home experiment: the KATRIN neutrino-mass measurement.
📍 Karlsruhe, DE
⚛️ KATRIN · TRISTAN
🧠 Causal Discovery
🗜️ Neural Compression
📄 Google Scholar ↗
⌨️ GitHub ↗
✉️ Contact
Active Projects
In progress
Crossover Scaling
KATRIN · Predictive Modelling
Predicting crossover scaling behaviour in KATRIN source parameters — ML-driven modelling of transition regimes for improved experimental control.
In progress
Snow Avalanche DTON
Geophysics · Deep Learning
Detection and characterisation of snow avalanche events using deep learning on seismic/acoustic sensor data — DTON architecture for real-time hazard monitoring.
In progress
Poisson Transformer
TRISTAN Detector · Neural Compression
Transformer-based compression for TRISTAN detector event data. Showing 25–31% improvement over existing baselines on Poisson-distributed readouts.
In progress
KATRIN Domain Bounds
KATRIN · Data Analysis
Optimizing exclusion bounds across multiple physics domains simultaneously — joint analysis of KATRIN source parameters for improved sensitivity.
In progress
Spectrogram Peak ID
KATRIN · Simulation-assisted Analysis
Identifying spectral peaks from KATRIN spectrograms using simulation-guided detection — improving automated identification of systematic features in the tritium spectrum.
In progress
Sensor Placement Optimization
Target · IJHE · Hydrogen Safety
Differentiable optimization of sensor layouts for hydrogen leak detection — maximising coverage and early-warning sensitivity under real deployment constraints.
In progress
BreastBench
Breast Deformation · Benchmarking Platform
Standardized benchmarking platform for evaluating AI methods on biomechanical breast tissue deformation — unified datasets, metrics, and baselines for fair comparison across GNN, FEM-surrogate, and physics-informed approaches.
Proposal
SCOPE
DFG Proposal · Scalable Causal Discovery
Scalable Causal Discovery for Predictive Time Series — addressing PCMCI scalability limits for large sensor networks in experimental physics.
Proposal
Helmholtz Co-Pilot
Collaboration w/ GSI · Helmholtz AI
Conversational AI agent for autonomous research knowledge management across Helmholtz experimental infrastructure.
Recent Work
Submitted
MeshGraphNet · Breast Deformation
Workshop Submission · 2025
Graph neural network approach to biomechanical breast tissue deformation simulation — applying MeshGraphNet to physics-informed surgical planning.
Published
Time-Series Forecasting
ICDM AI4TS Workshop · Nov 2025
Scalable time-series forecasting for scientific sensor networks — presented at the AI for Time Series workshop at ICDM 2025.
Under review
Structure Transfer
ECML-PKDD 2026 · Hydrogen Safety
Structure transfer in inverse problems applied to hydrogen leak detection — anonymous review in preparation.