There's not so much use for conjugate priors these days tbh. Better to spend the time figuring out posterior predictive checks. Any intuitions you have about marginal effects will probably be helpful too. And the loose equivalence between priors on parameters and regularization may also be helpful.
November 7, 2025 at 11:04 PM
There's not so much use for conjugate priors these days tbh. Better to spend the time figuring out posterior predictive checks. Any intuitions you have about marginal effects will probably be helpful too. And the loose equivalence between priors on parameters and regularization may also be helpful.
If there had been no border crisis with 100ks of migrants -- many from outside the Americas -- released into the interior, I think political rhetoric surrounding regularization of existing long-term migrants/Dreamers and more permissive rules for Mexicans/CentAms would have been *wayyy* less heated.
November 10, 2025 at 5:07 PM
If there had been no border crisis with 100ks of migrants -- many from outside the Americas -- released into the interior, I think political rhetoric surrounding regularization of existing long-term migrants/Dreamers and more permissive rules for Mexicans/CentAms would have been *wayyy* less heated.
New @ecapuab.bsky.social working paper by professor Mario Carillo @mariocarillo.bsky.social "The Effect of Migrant Regularization on Labor Exploitation"
econpapers.repec.org/paper/uabwpr...
econpapers.repec.org/paper/uabwpr...
EconPapers: The Effect of Migrant Regularization on Labor Exploitation
By Francesco Amodio, Elia Benveniste, Mario F. Carillo and Marc Riudavets-Barcons; Abstract: This paper shows that granting migrants legal status reduces labor exploitation. We study Spain's 2005 larg...
econpapers.repec.org
November 4, 2025 at 12:16 PM
New @ecapuab.bsky.social working paper by professor Mario Carillo @mariocarillo.bsky.social "The Effect of Migrant Regularization on Labor Exploitation"
econpapers.repec.org/paper/uabwpr...
econpapers.repec.org/paper/uabwpr...
(5/n) With this, we can run coarse-grained Langevin dynamics directly, without the need for any priors or force labels.
This works across biomolecular systems including fast-folding proteins like Chignolin and BBA.
Here is a comparison with and without our regularization:
This works across biomolecular systems including fast-folding proteins like Chignolin and BBA.
Here is a comparison with and without our regularization:
November 6, 2025 at 2:41 PM
(5/n) With this, we can run coarse-grained Langevin dynamics directly, without the need for any priors or force labels.
This works across biomolecular systems including fast-folding proteins like Chignolin and BBA.
Here is a comparison with and without our regularization:
This works across biomolecular systems including fast-folding proteins like Chignolin and BBA.
Here is a comparison with and without our regularization:
I'm also really curious what's happening here. A push for regularization has been a firm part of many advocacy campaigns, particularly for those on precarious temporary permits. We'll see who is included!
November 4, 2025 at 10:12 PM
I'm also really curious what's happening here. A push for regularization has been a firm part of many advocacy campaigns, particularly for those on precarious temporary permits. We'll see who is included!
Imaging-based models are being used to assess #BrainAge, but can they predict #NeurologicalDisease? @ritterkerstin.bsky.social &co show that simpler models, with lower age prediction accuracy, are paradoxically more sensitive to disease-related brain changes @plosbiology.org 🧪 plos.io/47NcIew
![Feature importance for brain-age prediction using Ridge regression. The heat maps illustrate the top-10 features for both the accuracy-optimized model (top panel) and the biomarker-optimized model (bottom panel), with color intensity representing the relative SHAP values. The x-axis denotes increasing regularization strength [α]. The accuracy-optimized model (α ~ 10^3), trained for maximum age prediction accuracy, highlights features such as the volume of the pons, gray-white contrast in the inferior parietal region, and volume of the cerebrospinal fluid (CSF). Conversely, the biomarker-optimized model (α ~ 10^5), trained to enhance the brain-age gap effect size for a majority of conditions vs. controls, emphasizes features like the volume of gray matter, volume of peripheral cortical gray matter, and mean intensity of the third ventricle. The dashed vertical lines indicate the regularization strength [α] at which each model was optimized. This comparison underscores the distinct feature importances between models focused on accuracy vs. those optimized for sensitivity to disease-relevant changes, supporting the manuscript’s thesis that traditional accuracy-optimized models may not provide the best biomarkers for disease detection.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:5522ztebtekoor5efelihqhb/bafkreidbug5gkmsmbisx2efl6ibzvquaozjtdfdimluid5jw7u7h23pdqe@jpeg)
October 29, 2025 at 2:07 PM
Imaging-based models are being used to assess #BrainAge, but can they predict #NeurologicalDisease? @ritterkerstin.bsky.social &co show that simpler models, with lower age prediction accuracy, are paradoxically more sensitive to disease-related brain changes @plosbiology.org 🧪 plos.io/47NcIew
I need to get used to the fact that Neural Networks with Dropout regularization can have a better validation loss than training loss. #machineLearning 🤖🧠
October 27, 2025 at 8:27 AM
I need to get used to the fact that Neural Networks with Dropout regularization can have a better validation loss than training loss. #machineLearning 🤖🧠
pfft, ML people are way ahead of you - call it weight decay, or, if you want to be really old school Tikhonov regularization.
October 25, 2025 at 3:18 AM
pfft, ML people are way ahead of you - call it weight decay, or, if you want to be really old school Tikhonov regularization.
🧠 Regularization, Action, and Attractors in the Dynamical “Bayesian” Brain
direct.mit.edu/jocn/article...
(still uncorrected proofs, but they should post the corrected one soon--also OA is forthcoming, for now PDF at brainandexperience.org/pdf/10.1162-...)
direct.mit.edu/jocn/article...
(still uncorrected proofs, but they should post the corrected one soon--also OA is forthcoming, for now PDF at brainandexperience.org/pdf/10.1162-...)
Regularization, Action, and Attractors in the Dynamical “Bayesian” Brain
Abstract. The idea that the brain is a probabilistic (Bayesian) inference machine, continuously trying to figure out the hidden causes of its inputs, has become very influential in cognitive (neuro)sc...
direct.mit.edu
October 22, 2025 at 8:59 AM
🧠 Regularization, Action, and Attractors in the Dynamical “Bayesian” Brain
direct.mit.edu/jocn/article...
(still uncorrected proofs, but they should post the corrected one soon--also OA is forthcoming, for now PDF at brainandexperience.org/pdf/10.1162-...)
direct.mit.edu/jocn/article...
(still uncorrected proofs, but they should post the corrected one soon--also OA is forthcoming, for now PDF at brainandexperience.org/pdf/10.1162-...)
Singular Value-based Atmospheric Tomography with Fourier Domain Regularization (SAFR). Lukas Weissinger et. al. https://arxiv.org/abs/2510.19542
October 23, 2025 at 4:56 AM
Singular Value-based Atmospheric Tomography with Fourier Domain Regularization (SAFR). Lukas Weissinger et. al. https://arxiv.org/abs/2510.19542
"Maybe the inability to memorize is a kind of regularization" About LLMs, but also middle-aged brains.
A kinder than usual way to look at it, I like it.
x.com/karpathy/sta...
A kinder than usual way to look at it, I like it.
x.com/karpathy/sta...
Andrej Karpathy on X: "My pleasure to come on Dwarkesh last week, I thought the questions and conversation were really good. I re-watched the pod just now too. First of all, yes I know, and I'm sorry that I speak so fast :). It's to my detriment because sometimes my speaking thread out-executes my" / X
My pleasure to come on Dwarkesh last week, I thought the questions and conversation were really good. I re-watched the pod just now too. First of all, yes I know, and I'm sorry that I speak so fast :). It's to my detriment because sometimes my speaking thread out-executes my
x.com
October 19, 2025 at 5:50 PM
"Maybe the inability to memorize is a kind of regularization" About LLMs, but also middle-aged brains.
A kinder than usual way to look at it, I like it.
x.com/karpathy/sta...
A kinder than usual way to look at it, I like it.
x.com/karpathy/sta...
New on arXiv: Why noise matters in learned regularization!
Data-driven regularizers fail without noise structure. 🎯
Tikhonov, Lavrentiev, quadratic — not equal under non-white noise.
🤫 Spoiler: Optimal reconstruction needs proper noise modeling.
👉 Regularizer choice = structure and noise.
Data-driven regularizers fail without noise structure. 🎯
Tikhonov, Lavrentiev, quadratic — not equal under non-white noise.
🤫 Spoiler: Optimal reconstruction needs proper noise modeling.
👉 Regularizer choice = structure and noise.
Sebastian Banert, Christoph Brauer, Dirk Lorenz, Lionel Tondji
Why the noise model matters: A performance gap in learned regularization
https://arxiv.org/abs/2510.12521
Why the noise model matters: A performance gap in learned regularization
https://arxiv.org/abs/2510.12521
October 15, 2025 at 6:40 AM
New on arXiv: Why noise matters in learned regularization!
Data-driven regularizers fail without noise structure. 🎯
Tikhonov, Lavrentiev, quadratic — not equal under non-white noise.
🤫 Spoiler: Optimal reconstruction needs proper noise modeling.
👉 Regularizer choice = structure and noise.
Data-driven regularizers fail without noise structure. 🎯
Tikhonov, Lavrentiev, quadratic — not equal under non-white noise.
🤫 Spoiler: Optimal reconstruction needs proper noise modeling.
👉 Regularizer choice = structure and noise.
7/8 The takeaway for the public: consider training choices like entropy regularization can make systems more robust so fewer restarts and less costly retraining when the world shifts. This means your learning systems are more durable and efficient.
October 7, 2025 at 6:34 PM
7/8 The takeaway for the public: consider training choices like entropy regularization can make systems more robust so fewer restarts and less costly retraining when the world shifts. This means your learning systems are more durable and efficient.
How well can we map its surface without making any assumptions (often called regularization in the field of image reconstruction)? We inject different levels of noise into our simulated data and optimize to find the best-fitting map. At high signal-to-noise we basically recover a perfect map! (9/N)
October 1, 2025 at 7:42 AM
How well can we map its surface without making any assumptions (often called regularization in the field of image reconstruction)? We inject different levels of noise into our simulated data and optimize to find the best-fitting map. At high signal-to-noise we basically recover a perfect map! (9/N)
me, thinking about immigration these days:
September 26, 2025 at 10:11 PM
me, thinking about immigration these days:
I tell students: You don’t need to memorize every ML algorithm. But you do need to understand:
— loss functions
— overfitting
— regularization
Master the “why,” not just the “how.”
#DataScience #MachineLearning #Statistics #AI #RStats
— loss functions
— overfitting
— regularization
Master the “why,” not just the “how.”
#DataScience #MachineLearning #Statistics #AI #RStats
September 18, 2025 at 6:08 PM
I tell students: You don’t need to memorize every ML algorithm. But you do need to understand:
— loss functions
— overfitting
— regularization
Master the “why,” not just the “how.”
#DataScience #MachineLearning #Statistics #AI #RStats
— loss functions
— overfitting
— regularization
Master the “why,” not just the “how.”
#DataScience #MachineLearning #Statistics #AI #RStats
there’s no linguistics work on this since 1953 btw. wiktionary & others claim it’s an over-regularization following “eaten” but I have a theory rn that it’s to avoid confusion with “drunk,” as in alcohol, which seems not that different from the tomar/beber regional distinction (in motivation)
September 19, 2025 at 3:44 AM
there’s no linguistics work on this since 1953 btw. wiktionary & others claim it’s an over-regularization following “eaten” but I have a theory rn that it’s to avoid confusion with “drunk,” as in alcohol, which seems not that different from the tomar/beber regional distinction (in motivation)
“Equal rights”: thousands of immigrant workers demand equality and regularization
“Direitos iguais”: milhares de trabalhadores imigrantes exigem igualdade e regularização
Milhares de imigrantes manifestaram-se em Lisboa para exigir documentos e o fim das injustiças na regularização. Detenções ilegais e problemas com sistema de sinalização do espaço Schengen são dois pr...
www.esquerda.net
September 17, 2025 at 9:21 PM
“Equal rights”: thousands of immigrant workers demand equality and regularization
Michael Potter, Stefano Maxenti, Michael Everett
Robust Survival Analysis with Adversarial Regularization. (arXiv:2312.16019v1 [stat.ML])
http://arxiv.org/abs/2312.16019
Robust Survival Analysis with Adversarial Regularization. (arXiv:2312.16019v1 [stat.ML])
http://arxiv.org/abs/2312.16019
December 27, 2023 at 5:00 AM
Michael Potter, Stefano Maxenti, Michael Everett
Robust Survival Analysis with Adversarial Regularization. (arXiv:2312.16019v1 [stat.ML])
http://arxiv.org/abs/2312.16019
Robust Survival Analysis with Adversarial Regularization. (arXiv:2312.16019v1 [stat.ML])
http://arxiv.org/abs/2312.16019
various regularization terms, we derive corresponding closed-form solutions while maintaining the $L$-smooth adaptable property always holds for any $L\ge 1$. Numerical experiments, on graph regularized clustering and sparse [7/8 of https://arxiv.org/abs/2503.02386v1]
March 5, 2025 at 6:17 AM
various regularization terms, we derive corresponding closed-form solutions while maintaining the $L$-smooth adaptable property always holds for any $L\ge 1$. Numerical experiments, on graph regularized clustering and sparse [7/8 of https://arxiv.org/abs/2503.02386v1]
Ziyang Wu, Jingyuan Zhang, Druv Pai, XuDong Wang, Chandan Singh, Jianwei Yang, Jianfeng Gao, Yi Ma
Simplifying DINO via Coding Rate Regularization
https://arxiv.org/abs/2502.10385
Simplifying DINO via Coding Rate Regularization
https://arxiv.org/abs/2502.10385
February 17, 2025 at 6:19 AM
Ziyang Wu, Jingyuan Zhang, Druv Pai, XuDong Wang, Chandan Singh, Jianwei Yang, Jianfeng Gao, Yi Ma
Simplifying DINO via Coding Rate Regularization
https://arxiv.org/abs/2502.10385
Simplifying DINO via Coding Rate Regularization
https://arxiv.org/abs/2502.10385
"Finally, by introducing several modifications (query specific regularization, disjoint encoders etc.), we are able to improve efficiency, achieving latency on par with BM25 under the same computing constraints."
github.com/naver/splade
github.com/naver/splade
GitHub - naver/splade: SPLADE: sparse neural search (SIGIR21, SIGIR22)
SPLADE: sparse neural search (SIGIR21, SIGIR22). Contribute to naver/splade development by creating an account on GitHub.
github.com
June 6, 2024 at 2:18 PM
"Finally, by introducing several modifications (query specific regularization, disjoint encoders etc.), we are able to improve efficiency, achieving latency on par with BM25 under the same computing constraints."
github.com/naver/splade
github.com/naver/splade
That really shouldn't matter. Bonus if you have the expired passport, but what's really important is that you're probably in the system, which is the important part.
Go here:
soniadiazmexico.com/other-visas/
Scroll past most of it until you see Regularization. Read up on it.
Go here:
soniadiazmexico.com/other-visas/
Scroll past most of it until you see Regularization. Read up on it.
February 18, 2025 at 3:37 AM
That really shouldn't matter. Bonus if you have the expired passport, but what's really important is that you're probably in the system, which is the important part.
Go here:
soniadiazmexico.com/other-visas/
Scroll past most of it until you see Regularization. Read up on it.
Go here:
soniadiazmexico.com/other-visas/
Scroll past most of it until you see Regularization. Read up on it.
It is not a universal amnesty but a particular regularization for very established people and part of our communities. It could help up to 500,000 adults and 50,000 children.
June 19, 2024 at 4:37 PM
It is not a universal amnesty but a particular regularization for very established people and part of our communities. It could help up to 500,000 adults and 50,000 children.