JKMRC Friday Seminar 2026: FTIR QA/QC with genAI: Application to the Minerals Industry

Value of the Information & Strength of the Argument

The presentation provides a clear and well-structured argument for using VAEs for anomaly detection in FTIR data. The value lies in the practical application: it addresses a real industrial need for automated QA/QC in high-throughput mineral analysis. The speaker effectively explains the limitations of traditional supervised classification and the necessity of an unsupervised approach. The argumentation is solid, building from the problem of high-dimensional data to the solution of using VAEs to learn a compressed representation and quantify deviations. The toy example and the detailed explanation of the VAE architecture help make the concept accessible. The speaker also highlights the agnostic nature of the anomaly detection, which is a strength as it allows for the identification of various issues (e.g., machine failure, mislabeling, novel mineralogy) without prior knowledge. However, the presentation lacks quantitative results or a comparative analysis with other anomaly detection methods, which would strengthen the argument for the effectiveness of the proposed approach.

Scientific Rigor, Source Quality, Title Accuracy

The presentation is scientifically rigorous in its methodology. The speaker, a research fellow, demonstrates a strong understanding of both the geological and computational aspects. The use of a VAE is well-justified, and the explanation of the latent space and reconstruction process is accurate. The sources cited are primarily the speaker’s own work and the collaboration with Rio Tinto, which lends credibility but also limits the breadth of external validation. The title accurately reflects the content, focusing on the application of generative AI to FTIR QA/QC. The seminar format is appropriate for sharing industry-focused research, but it is not a peer-reviewed publication. The description provides a clear abstract and bio, which helps contextualize the work. No external sources are cited beyond the SMI webinar page, which is a limitation for verifying the claims independently.

The title accurately reflects the content: a seminar on using generative AI (VAEs) for quality assurance/quality control of FTIR data in the minerals industry.

8/10

Presentation by a domain expert (Research Fellow at UWA) detailing a specific industrial application of variational autoencoders for anomaly detection in FTIR spectra. The methodology is clearly explained, and the work is grounded in a real industry collaboration with Rio Tinto. However, the presentation is a seminar talk, not a peer-reviewed publication, and lacks detailed quantitative results or external validation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and acknowledgments
• 03:00Overview of the Critical Resources of the Future center and data platform
• 10:00Introduction to the objective logger project and the need for QA/QC
• 15:00Explanation of the anomaly detection problem and toy example
• 20:00Description of FTIR spectroscopy and the high-dimensional data challenge
• 25:00Introduction to variational autoencoders (VAEs) and their architecture
• 30:00Explanation of the latent space and reconstruction process
• 35:00Case study: unmineralized BIF spectra from the Pilbara, data details
• 40:00Training strategy and use of XRF assays and logging interpretations
• 45:00Demonstration of good and bad reconstructions and anomaly detection results

• SMI Webinars — Page listing past and upcoming JKMRC Friday Seminars and other webinars from the Sustainable Minerals Institute.

• SMI Webinars — The seminar is part of the JKMRC Friday Seminars series, which is listed on this page, confirming the event's existence and context.