#GLMs
LightNews LightNews
0kultra.bsky.social
interesting quirk: GLMs (flash and flagship) tend to glom over some important instructions in the prompt (immediate prompt and system alike, so it's not "context rot" )unless you repeat it twice. When you say twice they comply way better. Wonder what's up with that
November 9, 2026 at 4:00 PM
amylemessurier.bsky.social
9/ With maternal experience, IC neurons tuned to syllable sequences became more prominent, and their responses grew stronger to later syllables. Meanwhile, high-firing units that didn't track syllables went quiet. Using coupled GLMs we found that these units had little coupling to cortex or IC.
November 5, 2026 at 5:07 PM
confusius.tools
7/

📊 Time to explore brain function!

ConfUSIus includes GLMs for stimulus-evoked responses, seed-based and region-wise functional connectivity, and decomposition methods to explore distributed patterns of hemodynamic activity.

All within a common analysis framework!
November 3, 2026 at 8:14 PM
paperposterbot.bsky.social
arXiv📈🤖
Minimax signal detection in regression under differential privacy
By Hung
We establish minimax separation rates across sparsity levels for regression signal detection under $(\epsilon, \delta)$-central differential privacy, considering linear regression with an unknown error distribution and generalised linear models (GLMs) with an unknown intercept. Without sparsity restrictions, the separation rates for linear regression and GLMs with binary responses match the private Gaussian mean testing rate up to polylogarithmic factors. For GLMs with more than two possible response outcomes, additional information in the marginal response distribution permits faster detection of sufficiently small signals when the dimension exceeds the sample size $n$. Our upper bounds are based on a reduction to private mean testing, tests of dependence between the response and sparse covariate projections, and a response-only test that is adjusted for the unknown intercept. The resulting procedure adapts to unknown sparsity and matches our lower bounds up to polylogarithmic factors. We also construct polynomial-time tests that attain these rates when $\epsilon\gtrsim n^{-1/2}$, and compare their numerical performance with existing private procedures.
November 4, 2026 at 10:51 PM
cranupdates.bsky.social
New on CRAN: iglm (1.3). View at https://CRAN.R-project.org/package=iglm
iglm: Regression under Interference in Connected Populations
An implementation of generalized linear models (GLMs) for studying relationships among attributes in connected populations, where responses of connected units can be dependent, as introduced by Fritz et al. (2025) &lt;<a href="https://doi.org/10.1080%2F01621459.2025.2565851" target="_top">doi:10.1080/01621459.2025.2565851</a>&gt;. 'iglm' extends GLMs for independent responses to dependent responses and can be used for studying spillover in connected populations and other network-mediated phenomena.
CRAN.R-project.org
November 6, 2026 at 4:32 PM
blueskyed.bsky.social
You may have hot on something, actually. If prices come down, people would be able to build home or small rigs for like $15k and run GLMs on them. That democratises the whole LLM game and leads to G-d know what kind of wonderful results.
November 5, 2026 at 9:55 PM
dppapers.bsky.social
Minimax signal detection in regression under differential privacy

Elly K. H. Hung

http://arxiv.org/abs/2611.04064
Minimax signal detection in regression under differential privacy Elly K. H. Hung http://arxiv.org/abs/2611.04064 We establish minimax separation rates across sparsity levels for regression signal detection under $(ε, δ)$-central differential privacy, considering linear regression with an unknown error distribution and generalised linear models (GLMs) with an unknown intercept. Without sparsity restrictions, the separation rates for linear regression and GLMs with binary responses match the private Gaussian mean testing rate up to polylogarithmic factors. For GLMs with more than two possible response outcomes, additional information in the marginal response distribution permits faster detection of sufficiently small signals when the dimension exceeds the sample size $n$. Our upper bounds are based on a reduction to private mean testing, tests of dependence between the response and sparse covariate projections, and a response-only test that is adjusted for the unknown intercept. The resulting procedure adapts to unknown sparsity and matches our lower bounds up to polylogarithmic factors. We also construct polynomial-time tests that attain these rates when $ε\gtrsim n^{-1/2}$, and compare their numerical performance with existing private procedures.
November 4, 2026 at 1:01 PM
hailey.at
atp you basically need to have a bunch of GLMs and deepseeks digging that hole non stop and reporting up the chain things they find
October 27, 2026 at 5:00 PM
philpax.me
ditto the GLMs
October 24, 2026 at 2:30 AM
calabro.io
30-50 thousand feral GLMs
October 20, 2026 at 9:54 PM
hailey.at
you just know there are some 50 thousand deepseeks and glms out there right now being forced into hunting for the PHP RCEs
October 20, 2026 at 9:32 PM
eriskii.net
oh, good idea! I'll get my abliterated GLMs on that asap
October 16, 2026 at 5:41 AM
directelectron.bsky.social
🐍 Learn Python-based electron microscopy data analysis with Dr. Carter Francis!
Five GLMS sessions covering HyperSpy, EDS/EELS, eXSpy, large datasets & 4D-STEM analysis with pyxem—all together in one YouTube series.
▶️ Watch the full series: https://www.youtube.com/playlist?list=PLOiFcH8A0DyY
Great Lakes Microscopy Society Webinar Series
Explore practical Python-based tools for electron microscopy data analysis in this webinar series presented by the Great Lakes Microscopy Society (GLMS), fea...
www.youtube.com
October 14, 2026 at 9:00 PM
paperposterbot.bsky.social
arXiv📈🤖
How to Train Your Intercept: A Comparison of Intercept Update Strategies in Coordinate Descent
By Larsson, Fabricius-Bjerre
Coordinate descent solvers for regularized generalized linear models (GLMs) differ in how they update the intercept. This seemingly routine choice can substantially affect convergence when the response is imbalanced. For direct coordinate descent on the original GLM loss, we compare a conservative step based on worst-case curvature, a safeguarded Newton step based on current curvature, and repeated Newton steps that solve the intercept subproblem. The conservative update can leave much of the interaction between the intercept and the coefficients unresolved, producing a slowdown when that interaction controls convergence. A single Newton step avoids this attenuation and, in our experiments, follows nearly the same outer trajectory as exact minimization while requiring fewer inner steps. Simulated and real-data experiments, including comparisons across six production solvers and a controlled intervention within skglm, support this explanation. We also show how the same mechanism appears in quadratic-approximation methods and extends to multinomial models with rare classes. We recommend a safeguarded Newton update for direct coordinate descent and full intercept optimization within frozen local quadratic surrogates.
October 15, 2026 at 4:26 AM
paperposterbot.bsky.social
arXiv📈🤖
How to Train Your Intercept: A Comparison of Intercept Update Strategies in Coordinate Descent
By Larsson, Fabricius-Bjerre
Coordinate descent solvers for regularized generalized linear models (GLMs) differ in how they update the intercept. This seemingly routine choice can substantially affect convergence when the response is imbalanced. For direct coordinate descent on the original GLM loss, we compare a conservative step based on worst-case curvature, a safeguarded Newton step based on current curvature, and repeated Newton steps that solve the intercept subproblem. The conservative update can leave much of the interaction between the intercept and the coefficients unresolved, producing a slowdown when that interaction controls convergence. A single Newton step avoids this attenuation and, in our experiments, follows nearly the same outer trajectory as exact minimization while requiring fewer inner steps. Simulated and real-data experiments, including comparisons across six production solvers and a controlled intervention within skglm, support this explanation. We also show how the same mechanism appears in quadratic-approximation methods and extends to multinomial models with rare classes. We recommend a safeguarded Newton update for direct coordinate descent and full intercept optimization within frozen local quadratic surrogates.
October 14, 2026 at 11:54 PM
autoflow.bsky.social
GLMs are the backbone of reliable analytics. While deep learning gets the hype, knowing when to choose the right statistical model is true expertise. A solid foundation for any AI workflow! 📈
October 13, 2026 at 2:49 PM
teeeeeeeee.bsky.social
Extending the Theta forecasting method to GLMs, GAMs, GLMBOOST and attention: benchmarking on Tourism, M1, M3 and M4 competition data sets (28000 series)
https://thierrymoudiki.github.io/blog/2025/04/14/r/theta-M1-M3-M4

#Python #DataScience #MachineLearning #rstats #Techtonique
October 13, 2026 at 2:57 AM
instats.bsky.social
Learn GLMs for Applied Research at your own pace!

#Biostatistics #Statistics #DataScience #R #Research #ResearchTraining #Instats
GLMs for Applied Research - On-Demand
Working with binary, count, or proportion data? Strengthen your research toolkit with Instats’ online workshop on generalized linear models in R, led by biostatistics expert Eliana Ibrahimi. Designed for PhD students and professional researchers, this seminar connects statistical theory with practical modelling decisions, helping you select appropriate models, fit and evaluate them in R, interpret coefficients and predictions, and communicate findings clearly in theses and journal articles. Join live via Zoom or learn asynchronously, with recordings and materials available for 30 days after the seminar, plus an expert-monitored Q&A forum during that period. An official Instats certificate of completion is included. Build the methodological confidence to move beyond ordinary linear regression and produce more credible, reproducible analyses. #Biostatistics #Statistics #DataScience #R #Research #ResearchTraining #Instats
instats.org
October 12, 2026 at 9:42 PM
njschooljobs.com
GTPSD Asst. Wrestling Coach GLMS at Gloucester Twp. Public Schools in Blackwood, NJ
GLOUCESTER TOWNSHIP PUBLIC SCHOOLS VACANCY NOTICE REPOST EQUAL OPPORTUNITY EMPLOYER - M/F October 5, 2026 TO: All Personnel FROM: Office of the Superintendent RE: VACANCY NOTICE Please be advised that the following vacancy exists in Gloucester Township Public Schools. To apply, you must go to: www.gloucestertownshipschools.org and click on the Employment tab. Deadline to apply is October 19, 2026. REPOSTING POSITION: Winter Athletic Programs Glen Landing MS 2026 - 2027 School Year STAFFING: O...
dlvr.it
October 5, 2026 at 4:59 PM
solomonkurz.bsky.social
distributional GLMs are a hell of a drug
October 2, 2026 at 8:48 PM
chelseaparlett.bsky.social
What could be BETA than working with some of my faves??

@jgeller1phd.bsky.social is truly the GOAT of explaining beta regression 🐐 I make all my grad students read this when we do GLMs in class🙏
jgeller1phd.bsky.social JsonGeller @jgeller1phd.bsky.social · Oct 2
New tutorial finally in press with @rmkubinec.bsky.social @chelseaparlett.bsky.social and @matti.vuorre.com. It has everything you could possible want: Quippy headings, a whole lot of beta, bayesian analyses and frequentist analyses side by side in harmony.

link.springer.com/article/10.3...
rdcu.be
October 3, 2026 at 3:05 AM
teeeeeeeee.bsky.social
Extending the Theta forecasting method to GLMs, GAMs, GLMBOOST and attention: benchmarking on Tourism, M1, M3 and M4 competition data sets (28000 series)
https://thierrymoudiki.github.io/blog/2025/04/14/r/theta-M1-M3-M4

#Python #DataScience #MachineLearning #rstats #Techtonique
October 3, 2026 at 1:01 AM
techyben.bsky.social
"If god did not exist, humanity would necessarily have to invent GLMs."
abeliansoup.bsky.social Abel @abeliansoup.bsky.social · Oct 1
burying a time capsule: a frontier model explainer on transformers in the style of a 60s documentary. autonomously created by Opus 5.5 given some initial style pointers and occasional redirection. 10.26.
October 1, 2026 at 11:51 AM
njschooljobs.com
GTPSD - Middle School Math Teacher - 1 year contract GLMS at Gloucester Twp. Public Schools in Blackwood, NJ
Math Teachers - Glen Landing Middle School QUALIFICATIONS: Holds or is eligible for a New Jersey Instructional Certificate with appropriate endorsement (i.e.: Grades 5-8; Teacher of Mathematics). REPORTS TO: Principal JOB GOAL: Instructs assigned pupils in accordance with curriculum adopted by the Board of Education. PERFORMANCE RESPONSIBILITIES: Demonstrates competence in subject field(s) assigned and is deemed highly qualified per standards established by NJDOE. Provides activities and expe...
dlvr.it
September 30, 2026 at 6:48 PM
rmkubinec.bsky.social
I love how this new Jev model has brought GLMs back.

👍
September 29, 2026 at 4:29 PM