Yair Daon

Lots of fun working on this study. We take observed disease incidence and infer (well, refute) causal relations. @UriObolski @krisparag1

@MikeE_3_14 @LoveForReall @ClaudeEShannon @claudeshannon @claudes24060374 @perseus_sword

יש אהבות שלא עוברות אף פעם - חבל שהמחבורות שלי הלכו לאיבוד במעברים הרבים קרדיט : @LoveForReall

@biumedicine @Mucus_Man @NirQvit @Stanford 🏆🏆🏆🥇🥇🥇👑👑👑🤩🤩🤩💪🏻💪🏻💪🏻

🎉 Congrats to Prof. @NirQvit on his promotion to Associate Professor at the Faculty of Medicine in the Galilee! A chemist advancing ground breaking discoveries for #heartdisease and more, with a @Stanford background and global collaborations. A true scientific leader!

@alec_helbling Seems off tho. Area on right should be ~half as left but doesn’t look so. My guess: instead of halving area they halved each axis —> area on right is ~1/4 of left.

Quite an evocative visualization from the NYT about NSF budget cuts. It gives the impression that science funding has been shattered.

There are many moving pieces when turning a project into a machine learning conference paper, and best practices/nuances no one writes up. I made a comprehensive paper writing checklist for my mentees and am sharing a public version below - hopefully it's useful, esp for NeurIPS!

Oldies but goldies: Benettin, Galgani, Giorgilli & Strelcyn. Lyapunov Characteristic Exponents for smooth dynamical systems and for hamiltonian systems; a method for computing all of them. Identify chaotic behavior and quantify dynamical stochasticity. Format: @gabrielpeyre

@VickersBiostats @seanluomdphd Isn't the prevalence assumed known? I.e. experts and practitioners (unlike myself) know what the prevalence is and can gauge usefulness of a new test based on said prevalence.

@seanluomdphd Specificity is about 90%. For sensitivity of 90%, if prevalence is 1%, NPV would be 99.9% and PPV 47.6%, which would be excellent test characteristics. You can't look at sensitivity and know whether a test is useful or not.

NEJM has not one, but two studies on colorectal cancer screening that report only sensitivity and specificity of a new test. I have zero idea how to interpret the results: are they good tests or bad? #ClinicallyRelevantStatistics nejm.org/doi/full/10.10…

@soumitrashukla9 @SamFriedmanSoc @Daniel_Laurison Super interesting, thanks for sharing!!

1/N The “class ceiling” in elite US jobs is real. @SamFriedmanSoc & @Daniel_Laurison find that even when people from working-class backgrounds reach top professions, they still earn about $20,000 less per year than privileged-origin colleagues... 🧵

@GilladRosen גּוּרה שהיא הכי טובה

הרדמנו את החתול אחרי 21 שנה איתו אז ברשותכם אשמח אם תוכלו להגיב פה תמונות חמודות של החיות שלכם 🙏🏻

I am recruiting graduate students for various projects, including identifying early warning signals to, and inferring root causes of epidemics. Contact me at firstname.lastname@gmail.com @biumedicine

Hi Twitter! I am recruiting graduate students for various projects, including identifying early warning signals to epidemics and inferring root causes of epidemics from observational data alone! @biumedicine

INTRODUCING: pxl-srt It sorts pixels in an image by color (i implemented and deployed this 20min after seeing this tweet)

@EBirati and myself are recruiting a PhD student for identifying cardiac failure modes! Project requires a mix of machine learning, insights from dynamical systems theory and deep understanding of how the human heart operates. Apply at firstname.lastname@gmail.com @biumedcine

2025 resolution: get more citations at any cost Next paper: (via Philip Cohen. The culprit paper: mpra.ub.uni-muenchen.de/122925/)

@EBirati and myself are recruiting a PhD student for a high impact computational study of heart failures! You will apply machine learning techniques to high resolution multi-channel ECG data, creating data-driven theory of heart failure dynamics. @biumedicine

@Bayesian_Analys Bonus: I also prove that replacing iid noise with correlated noise mitigates clusterization in the generic model (but this is kinda known).

@Bayesian_Analys Moreover, clusterization is generic (fig below; all you need to understand is that clusterization fraction ≈ 1). Luckily, all this bad stuff happens for linear problems with iid Gaussian noise. Hence I suggest you avoid linearization. Alternatives in arxiv.org/abs/2007.12032

Great news from @Bayesian_Analys! My study on clusterization in D-optimal designs was accepted for publication. For inverse problems, optimal sensor locations may be clustered (pic below). I asked why this is so, is clusterization generic, and what are the implications. 1/n