WORKSHOP
“Scientific Data: Representations, Descriptors, and preprocessing”
We bring together leading academic scientists and researchers!
September 18th, 2026 1pm - 5pm
Scope of the event
The bridge between raw physical data and a machine-learnable format is often the most consequential design decision in the entire ML workflow. A poorly chosen representation can render even the most powerful model ineffective, while a well-designed descriptor can make a simple model competitive with far more complex alter- natives. This workshop examines how molecules, crystals, spectra, voltammograms, and simulation trajectories are encoded as inputs to ML models, and develops the principles — invariance, locality, and smoothness — that make a representation physically well-behaved.
The session covers hand-crafted molecular descriptors (SMILES strings, fingerprints, Coulomb matrices, SOAP and other invariant descriptors), crystal structure encodings (space groups, Wyckoff positions), and represen- tations for electrochemical data including feature extraction from cyclic voltammetry, impedance spectra, and galvanostatic cycling curves. The treatment of learned representations complements the discussion of hand- crafted descriptors, and the workshop addresses practical challenges common in experimental physical science: normalisation, imputation, handling of small datasets, and data augmentation strategies for physical data.
Participants will develop a systematic framework for evaluating and selecting representations, understanding the trade-offs between interpretability and expressiveness, and recognising how descriptor choice interacts with model architecture and the physical symmetries of the system under study.
Extended Key Takeaways & Outputs
• Systematic framework for encoding molecular, crystallographic, and electrochemical data for ML
• Understanding of invariance, locality, and smoothness as design principles for descriptors
• Practical strategies for normalisation, imputation, and augmentation with small experimental datasets
• Clarity on trade-offs between hand-crafted and learned representations
• Preparation for the domain-specific applications in Modules II and III
Agenda & Speakers
13:00–13:05 — Introduction & Objectives
Organizer
· Importance of data representation in scientific machine learning workflows
· From raw experimental data to machine-learnable inputs
· Workshop objectives: robustness, physical consistency, and interpretability
13:05–13:35 — Molecular Representations and Descriptors
Speaker (Academia): N.N.
· SMILES, molecular fingerprints, and Coulomb matrices
· Invariant descriptors (SOAP, symmetry functions) and their physical meaning
13:35–14:05 — Crystallographic Representations and Materials Encoding
Speaker (Academia): N.N.
· Crystal structure encoding: space groups and atomic environments
· Local vs global descriptors for materials property prediction
14:05–14:35 — Representing Electrochemical and Spectroscopic Data
Speaker (Academia): N.N.
· Feature extraction from cyclic voltammetry and impedance spectra
· Encoding time-series and signal-based experimental data
14:35–14:45 — Break
14:45–15:15 — Data Preprocessing: Normalisation, Imputation, and Augmentation
Speaker (Industry): N.N.
· Handling missing and noisy experimental data
· Normalisation techniques and scaling strategies
· Data augmentation for small and expensive datasets
15:15–15:45 — Learned Representations vs Hand-Crafted Descriptors
Speaker (Industry): N.N.
· Representation learning in neural networks
· Trade-offs between interpretability and expressiveness
· Integration with downstream ML models
15:45–16:15 — Representation Design Principles: Invariance, Locality, and Smoothness
Speaker (Policy/Applied Research): N.N.
· Physical constraints in descriptor design
· Ensuring robustness and transferability across datasets
· Linking representation to model architecture and symmetry
16:15–16:55 — Panel Discussion & Strategic Alignment
All Speakers + Moderator
· How critical is representation compared to model choice?
· Balancing physical interpretability with predictive performance
· Standardisation of descriptors across scientific domains
· Best practices for small-data scientific environments
Concluding remarks
Registration
Deadlines:
Early-Bird: until July 15th, 2026
Registration : until September 5th, 2026
The registration takes place online by filling out the form.
The registration and access to the virtual room for workshop is effective only if the fees payment has been confirmed.
Please note that INVIRTA MEMBERS can purchase all our workshops and services with 10% discount . If you desire to get more information on our membership please press here .
Prices* :
Industry: 150 € (early bird**: 125 €)
Academy: 125 € (early bird**: 100 €)
Students***: 90 € (early bird**: 75 €)
* Note that prices are subject to VAT (for European countries, except Germany, provide the VAT number of your organisation to be within the VAT reverse framework payment).
** Until November 15th, 2020
*** Copy of the registration certificate (Ph.D, undergraduates)
Special price offers for Workshop packages
Our Workshop packages are valid for any workshops offered by our academy and can be used anytime. If you want to use one of our packages press here.
