Ziang Niu @MaxwellAng1

edgePython 0.2.6 released. Thanks to everyone who is sharing issues via Github or just letting me know of desired features / improvements. github.com/pachterlab/edg…

Congratulations to the R Core Team on receiving the 2026 Rousseeuw Prize for Statistics. For over 30 years, R Core has maintained one of the most important pieces of infrastructure in statistics. Little recognition, thousands of volunteer hours, and enormous impact. Thank you!

As I have pointed out many times publicly, single cell foundation model performance will scale with the number of perturbations, not the number of cells. We barely have ~100k perturbations in the public domain and it is reasonable to expect we need millions to truly go OOD. Factor in also: modalities, time points, combinations etc. Short term, there is little to do using public data. Best medium term option, consider semi-mechanistic modelling: arxiv.org/abs/2501.19178 Long term, …, wait for the press release ;)

Fabian Theis @fabian_theis

3 weeks ago

Do single-cell foundation models obey scaling laws? A somewhat thought-provoking new Nature Methods study by the Crawford lab suggests that, for current single-cell foundation models, the answer may be “not really.” Across a broad range of architectures and downstream tasks,

10 53 237 29K 185

I see the idea of empirical Bayes is shining there!

Nima Alidoust @nalidoust

3 weeks ago

Today we release Rhaister, an elegant statistical model that predicts drug phenotypes in new contexts w/ accuracies comparable to experimental assays. And dropping Emerald Bay, a 2M cell dataset measuring long time-course phenotypes across 1000s of drug-cell line interactions.

15 55 256 62K 167

1/ We built the Genomics × AI blog so the genomics + ML community can share work fast and actually discuss it — incremental results, negative results, tutorials — without waiting on a publisher. Posts are live, more landing over the coming weeks: genomicsxai.github.io

A flaw in Person-Δ may be overstating progress in single-cell perturbation prediction models. Pearson warned about this in the 19th century: reusing the same controls induces spurious correlation. Split the controls, and much of the claimed prediction power fades. Link below 👇

In AI-guided discovery, models often turn huge candidate pools into shortlists for costly validation. We ask: can we put an error budget on AI-generated shortlists before running the experiment? For example: • Can we keep failed hits below 10%? • How many candidates should we test to get enough true positives? • How far down the list can we go before expecting too many false positives? • If we already have a fixed top-K list, how many are likely wrong? 📢 Excited to share TxConformal, a framework to turn AI scores into shortlists with controlled/estimated false positives, even in tasks where new candidates differ from past experimental data. This is joint work with amazing @KexinHuang5 @jure @EmmanuelCandes , in collaboration with Genentech @nate_diamant @gabo_scalia. We test it across proteins, genetic perturbations, regulatory DNA, clinical trials, ADMET, and antibacterial virtual screening. In a prospective A. baumannii screen at Genentech, TxConformal estimated 80.3 false positives before wet-lab validation; the experiment found 91, within the 90% CI. Preprint: biorxiv.org/content/10.648… Code: github.com/ying531/TxConf… 🧵[1/n] 👇

@YingJin531 Very nice work! I think the loop of AI assisted simulation-validation scientific process is forming! Statistics is still very useful at the end of simulation/generation to control the size of wrongly chosen scientific targets and at efficiently planning validation experiments.

Excited to share our ICML 2026 Hypothesis Testing Workshop in Seoul, this July! @icmlconf 🎉This workshop aims to bring together researchers developing modern hypothesis testing methodology and applying it to machine learning problems such as robustness, distribution shift, security, medicine, and LLM evaluation. In other words, if you care about how we make ML claims rigorous, this workshop is for you. We now have four confirmed speakers: Arthur Gretton @ArthurGretton, Yao Xie @yaoxie21851119, Bo Li @uiuc_aisecure, and Yisong Yue @yisongyue. The organizing team includes Xiuyuan Cheng (Duke), Feng Liu @AlexFengLiu1, Lester Mackey @LesterMackey, Shayak Sen @shayaksen, Danica J. Sutherland @d_j_sutherland, and Nathaniel Xu (UBC). 📌 Submission deadline: 10 May 2026 📌 Notification: 26 May 2026 📌 Camera-ready: 17 June 2026 📌 Workshop date: July 10 or 11, 2026 (TBA) 🚩Check more information below! 🔗Website: testing.ml 🔗Submission Portal: openreview.net/group?id=ICML.… We’re also recruiting PC members/reviewers. 🔗 Reviewer interest form: docs.google.com/forms/d/e/1FAI… 🏁Please feel free to share this with colleagues, collaborators, and students who may be interested. #ICML #ICML26

1/ Happy to release StatsClaw — an open-source multi-agent workflow for building statistical software with AI. w/ @Maple_Optboy Site: statsclaw.ai Paper: bit.ly/statsclaw

@linstonwin @captgouda24 @ben_golub So true.

Last month, a final-year CSE student DM’d me in panic. “Applied to 60+ summer internships. Not a single test link or interview call.” We spent 2 evenings rewriting his applications with Claude. Result: 6 interview invites in the next 10 days (5 startups and Microsoft). Here are the exact 7 prompts that turned it around👇

Yesterday I shared a Claude skill for academic slides. Now, the underlying guide — no AI needed, works for anyone. 📄 Best Practices for Academic & Analytical Presentations (Free PDF) bit.ly/4cp3QPe → Action titles, structured argument, exhibit discipline, citations

Roman Vershynin: A friendly proof of the Berry-Esseen theorem arxiv.org/abs/2602.06234 arxiv.org/pdf/2602.06234 arxiv.org/html/2602.06234

The UCL IMSS Annual Lecture will take place on the 27th April with a keynote from @LesterMackey. The theme is 'Computational Statistics and Machine Learning', and we will have talks from Alessandro Barp, Paula Cordero Encinar & Po-Ling Loh. imss2026.github.io @stats_UCL

Semiparametric KSD test: unifying score and distance-based approaches for goodness-of-fit testing ift.tt/RL4P8ID

🧠 Can Omni Large Language Models (OLLMs) truly reason the same across audio, vision, and text modality? 🚀 Introducing XModBench: a tri-modal multiple-choice benchmark (text ↔ vision ↔ audio) designed to test cross-modal consistency and capability in omni-modal LLMs. 🔍 Covering 60,828 QA pairs across five task families (perception, spatial reasoning, temporal reasoning, linguistic, external knowledge). 🤯Results show: even top models like Gemini 2.5 Pro struggle: (i) < 60% accuracy on spatial & temporal reasoning (ii) Sharp performance drop when the same content is given as audio rather than text (iii) Systematic imbalance — reasoning is far less consistent under vision/audio contexts than text 📘 Paper: arxiv.org/pdf/2510.15148 🌐 Project: xingruiwang.github.io/projects/XModB… 🤗 Huggingface: huggingface.co/datasets/RyanW…

@JasonSun10 Congratulations Zhuo!!!

Proud work with @RenZhimei

arXiv math.ST Statistics Theory @mathSTb

a year ago

Ziang Niu, Zhimei Ren: Assumption-lean weak limits and tests for two-stage adaptive e... arxiv.org/abs/2505.10747 arxiv.org/pdf/2505.10747 arxiv.org/html/2505.10747

0 0 0 489 0