Guest
Sergey Barannikov, Senior Research Scientist at Skoltech AI Center (PhD, UC Berkeley), kicks off the SMILES‑2025 morning in Harbin with a deep dive into topological data analysis. Highlights: • an intuitive introduction to persistent homology and barcodes that reveal hidden “holes” in data; • how topological features diagnose and prevent mode collapse in GANs, autoencoders and VAEs, and help monitor transformer training; • real‑world use cases: mitigating hallucinations in large language models and detecting synthetic or anomalous data. A must‑watch for researchers seeking interpretable quality metrics for generative models and practitioners working with high‑dimensional datasets.
