From Deep Learning to Scientific Discovery: Toward Reliable and Trustworthy AI
Researchers are invited to a focused session on dependable AI for science. The Department of Chemistry and the Northern West Virginia Local Section of the American Chemical Society will host this presentation at 6 p.m., Jan. 21, in Clark Hall, Room 208.
The presenter is Prashnna Gyawali, assistant professor in the Lane Department of Computer Science and Electrical Engineering.
Event details
- Title: From Deep Learning to Scientific Discovery: Toward Reliable and Trustworthy AI
- Host: Department of Chemistry and the Northern West Virginia Local Section of the American Chemical Society
- Presenter: Prashnna Gyawali, assistant professor, Lane Department of Computer Science and Electrical Engineering
- Time: 6 p.m., Jan. 21
- Location: Clark Hall, Room 208
Why this matters for your work
As models move closer to the bench, reliability and trust are non-negotiable. This session centers on how to build AI you can test, explain, and defend in peer review.
Expect a practical lens on moving from neural nets to scientific results that hold up under replication.
Key concerns to bring to the discussion
- Data quality: controls, drift, and documentation
- Model trust: calibration, uncertainty estimates, and interpretability
- Reproducibility: versioning, pipelines, and audit trails
- Integration: pairing predictions with experimental design and validation
Make the most of the session
- Bring one active research question where AI could speed analysis or guide experiments.
- Outline how you would verify model outputs in your lab (ground truth, orthogonal methods, or blinded tests).
- List failure modes you worry about (overfitting, bias, sample leakage) and ask how to test for them.
Further reading and tools
- NIST AI Risk Management Framework for evaluating AI reliability and risk.
- AI courses by job to build the exact skills your role demands.
Put this on your calendar and plan to arrive a few minutes early. If you're working at the edge of chemistry, materials, biology, or engineering, this conversation will help you tighten your methods and get cleaner results from your models.
Your membership also unlocks:
