Rare event prediction on time series that change structure mid-stream? [D]

<!-- SC_OFF --><div class="md"><p>Just a Student trying to build some projects and this one caught my attention.</p> <p>An automated system for detecting statistically significant

<!-- SC_OFF --><div class="md"><p>I am responsible for a research project that is supposed to train a GPT-like model (Transformer-decoder) with 100M, 250M and 500M model

<!-- SC_OFF --><div class="md"><p>Most of the content about TSFMs:</p> <ul> <li>Either overhypes their true potential,</li> <li>Or highlights weaknesses that are either irrelevant

<!-- SC_OFF --><div class="md"><p>I have per-frame head counts from P2PNet running on crowd video clips. Counts are stable but noisy (±10%). I need to predict density 5-10 frames

<!-- SC_OFF --><div class="md"><p>Split conformal prediction gives you a clean statement: with a calibration set exchangeable with future data, your prediction sets cover the true

<!-- SC_OFF --><div class="md"><p>I built a calibrated predictive model on point-in-time fundamentals and two results surprised me enough to write up. Sharing the project and the