Shahriar Noroozizadeh

17/ It is time to stop evaluating in the dark. Come test your models on the whole elephant! w/ Xiaobin Shen*,@jeremycweiss, George H. Chen Paper: arxiv.org/abs/2603.05483 Code: github.com/Shahriarnz14/S… Data: huggingface.co/datasets/snoro… ICLR Page: iclr.cc/virtual/2026/p… #ICLR2026 #CausalInference #SurvivalAnalysis #MachineLearning

16/ TL;DR: • We unify survival HTE methods into 3 families. • We benchmark 53 variants on 40 synthetic, 10 semi-synthetic, and 2 real datasets. • We explicitly test violations of ignorability, positivity, and ignorable censoring. • No method wins everywhere.

1/ Excited to present our #ICLR2026 paper today at the poster session! Come say hi: Pavilion 3 P3-#101 (10:30 AM – 1:00 PM 🇧🇷). “SURVHTE-BENCH: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival Analysis” Here is why we built it, and why it matters 🧵👇

Since you’ve blocked comments on your other post about Iranians "celebrating" war, I’ll respond here. To analyze why some people appear to celebrate being bombed, it’s important to seriously grapple with where that reaction comes from. When tens of thousands are killed in peaceful protests (Jan 8-9, 2026), when every non-violent path is met with bullets, prisons, and silence, centering Pahlavi propaganda as the main explanation feels at best selective. One might think that years of executions, torture, crushed uprisings, and the systematic annihilation of non-violent avenues would rank slightly higher in explaining why some people feel pushed toward catastrophic outcomes, but alas priorities differ, I suppose. For the record, I don’t welcome any war, but I can understand the desperation Iranians are expressing without positioning myself on some higher moral ground.

We should never abstract war, dematerialize it, or dehumanize those who suffer from it. According to official reports from Iran, U.S.-Israeli attacks struck a girl’s elementary school, and at least 40 students were killed.

10/ Come chat with us at the poster session today! (@jeremycweiss, @SayantanKumar94) 💻 Code: github.com/Shahriarnz14/T… 📄Paper: arxiv.org/abs/2504.10340 #AAAI2026 #TemporalAI #HealthcareAI #MachineLearning

9/ TL;DR: 1. Narrative text > Structured data alone. 2. But Text Order ≠ Time Order (The Trap!). 3. Prompting ≠ Forecasting (You need to train!).

1/ Show a doctor a raw spreadsheet of EHR codes. Without context, they’ll struggle to forecast what happens next. 📉 The critical signal lives in the Narrative Notes. That's where the patient's true trajectory is. But for AI, training only on these notes can be flawed. Why?🧵👇