@maugendre@hachyderm.io

Eric Maugendre about data @maugendre@hachyderm.io

· 1w ago

Replying to

Here is where i come from: If you train your machine on a dataset with too few input variables, you cannot capture the behaviour. (Current analogies may be corporations focused on "indicators" and governments focused on GDP.) When learning statistics, we are taught of the perils of #overfitting: If the number of input variables is similar to the number of data points, your calculations perform better on the training data, but tend to perform worse on out-of-sample data.

2

1

0

Eric Maugendre about data @maugendre@hachyderm.io

· 1w ago

Replying to

In the below fields, #adaptability may already be baked in, if the system has absorbed other contexts: • Staff have a life apart from being managed. • Animals, trees, ecosystems retain lessons from history. • The machine model with many features will promptly react to disruption. Here is a nice illustration by Jonathan Tonkin: https://predirections.substack.com/p/nine-ways-extreme-events-reshape

predirections.substack.com

Nine ways extreme events reshape river biodiversity

How floods, droughts, heatwaves, and compound events are impacting life in rivers

1

0

2

0

Eric Maugendre about data @maugendre@hachyderm.io

· 1w ago

Replying to

In other words: • An excellent fit makes the data practitioner obtain an inference engine that generalizes badly. • Similary, a living entity that only fits its environment may fail to degrade gracefully when the environment changes. • A social organisation that is optimised is not resilient.

1

2

0

1

Eric Maugendre about data @maugendre@hachyderm.io

· 3mo ago

Boosted by @DataKnightmare@mastodon.xyz

Your Apple or Microsoft device has an identifier (or two). The software reports part of what the device is doing. Using a VPN does not protect your from your operating system. Trackers are placed ahead of due process. Regulations are always late, and insufficient. In the EU, @noybeu@mastodon.social is a leading litigator. Examples: https://en.wikipedia.org/wiki/Max_Schrems#Complaints_filed_under_GDPR_in_2018%E2%80%9319 The latest trackers are called DSID and GDID. Here is a nice story: "It’s basically Microsoft self-certifying, promising that the American body overseeing this, the FTC, is independent enough that Europe can trust it." "The redress side is even worse." https://no01.substack.com/p/your-computer-already-told-on-you #privacy #digitalSovereignty #litigation #EUCJ #GDID #GDPR #FTC #Stokes #VPN #computers

en.wikipedia.org

Max Schrems - Wikipedia

6

1

7

0

Eric Maugendre about data @maugendre@hachyderm.io

· 2mo ago

The weathers, they are a-changing. We can make sense of pattern evolution. I am starting a thread about Europe's new #climate conditions. 🧵 (You may read stories in the "ALT" descriptions.) #trend #heatwave #dataViz #chart #loess #linearRegression #statistics #France #football #WorldCup #Maignan #history #coloniality #weather #summer #Europe #climate #climateChange #globalWarming

3

1

4

0

Eric Maugendre about data @maugendre@hachyderm.io

· 5mo ago

Replying to

@alineblankertz@indieweb.social Please accept a support message from here (data scientist and former telco infra planner). I hope that your intended proposal will include free software and infra coops. Example: https://en.wikipedia.org/wiki/French_Data_Network

en.wikipedia.org

French Data Network - Wikipedia

2

0

Eric Maugendre about data @maugendre@hachyderm.io

· 17mo ago

Replying to

You may consider hosting your social data out of the USA. Here is a search tool of Mastodon instances that filters on country: https://mastodon.fediverse.observer/list 🧵 #howTo #EuropeanAlternatives #decoupling #nonAligned #alternatives #nonUS #MastodonMigration #search #discoverability #MastodonGuide #privacy #GDPR #Mastodon #NewHere #Newbie #beginner #FediverseTips #FediTips #FediHelp #MastodonHelp #MastodonTips #MastodonHowTo #resource #resources

mastodon.fediverse.observer

Fediverse Observer checks all sites in the fediverse and gives you an easy way to find a home from a map or list or automatically.

Mastodon Sites Status. Find a Mastodon server to sign up for, find one close to you!

3

0

3

0

Eric Maugendre about data @maugendre@hachyderm.io

· 47mo ago

#introduction
Hello people :-)
I am a French data engineer.
Here i will initiate conversations from texts about #security #MastodonForks #governance #data protection #dataPrivacy #GDPR #discoverability #psychology #sociality #archives #library #retrieval

If you see this message, please boost it so this new account appears in your local timeline (and gets federated if your posts are public).

5

0

15

0

Eric Maugendre about data @maugendre@hachyderm.io

· 28mo ago

Recording your activities on device offers an infostealer. This feature fundamentally changes the relationship between you and your computer.

Kevin Beaumont @GossiTheDog@cyberplace.social wrote about the #Recall announcement: https://doublepulsar.com/how-the-new-microsoft-recall-feature-fundamentally-undermines-windows-security-aa072829f218

Medium

How the new Microsoft Recall feature fundamentally undermines Windows security

Microsoft ship a feature taking constant screenshots of PCs by design.

1

0

1

0

Eric Maugendre about data @maugendre@hachyderm.io

· 1w ago

Replying to

System thinking and inference with context (continued) For statisticians, here is a nice video that promotes double descent: https://www.youtube.com/watch?v=z64a7USuGX0 For statisticians then, here is a beautiful "Escape Dimensions Theory": https://bsky.app/profile/flavioh.bsky.social/post/3mnwg6kx3b22l

What the Books Get Wrong about AI [Double Descent]

0

1

0

Eric Maugendre about data @maugendre@hachyderm.io

· 1w ago

Replying to

… Systematic use of informed studies leads to more precise, but more biased estimates. Critically comparison of informed and skeptical priors can provide more nuanced and solid understanding of our findings. https://bsky.app/profile/fusaroli.eurosky.social/post/3lau7cygksy2h …

Bluesky Social

Riccardo Fusaroli (@fusaroli.eurosky.social)

Should we use findings from previous studies and meta-analyses to shape our statistical inferences (aka informed priors)? What are the advantages and issues? Strap on for a loooong thread (link to a video of the talk at the end) 1/

0

Eric Maugendre about data @maugendre@hachyderm.io

· 1w ago

Replying to

… "In real life, we weigh the anticipated consequences of the decisions that we are about to make. That approach is much more rational than limiting the percentage of making the error of one kind in an artificial (null hypothesis) setting or using a measure of evidence for each model as the weight." Longford (2005) http://www.stat.columbia.edu/~gelman/stuff_for_blog/longford.pdf #bias #stats #nullHypothesis #probability #probabilities #pValues #statistics #statisticalLiteracy #inference #modelling #modeling #regression #linearRegression

stat.columbia.edu

0

1

0

Eric Maugendre about data @maugendre@hachyderm.io

· 5d ago

Replying to

A recent journal article attempts extending biological evolution from #ML! Here is a stylized diagram of "double descent" by the paper author Steve Frank. The first descent (decrease in error on out-of-sample data) occurs as you add input variables to machine training. Generalization error enhances due to overfitting as you continue. A second descent *may* occur if you add even more input variables to machine training.

0

Eric Maugendre about data @maugendre@hachyderm.io

· 5d ago

Replying to

Why would that be (according to Steven Frank)? "The excess dimensions create large subspaces of equivalent training performance but varying generalization ability. Within these subspaces, learning dynamics preferentially reach simple smooth solutions that capture the general structure of the problem rather than its idiosyncratic details". Similarly for proteins: https://bsky.app/profile/john-mcbride.bsky.social/post/3mpycngozbs2j

Bluesky Social

John McBride (@john-mcbride.bsky.social)

We found the same explanation works for why proteins are bigger than they need to be for their functions. More amino acids allows more low-cost paths --> greater evolvability So, there are two explanations that are mutually compatible with evolution and deep learning. 😎 https://academic.oup.com/mbe/article/39/11/msac217/6754178

0

Eric Maugendre about data @maugendre@hachyderm.io

· 5d ago

Replying to

How to predict if you do not know what the driver is? "Conditional Feature Importance (CFI; Strobl et al. (2008)) is a classical variable importance measure that accounts for the relationship between the studied feature and the others." "We want to avoid making strong assumptions about the relationship between inputs and outputs, as it can be complex. However, we assume that the relationship among the input covariates is simple, typically because they originate from the same generative process (e.g. a measurement device). Then, it can be much more efficient and accurate to study the relationship between the j-th covariate, Xj, given the rest, X−j, rather than directly regressing y on X−j. This approach is also advantageous in settings where labeling data is expensive, as there are typically more available samples for the input pair (X−j, Xj) than for the pair (X−j, y)." Angel Reyero-Lobo, Pierre Neuvial, Bertrand Thirion (2026). https://arxiv.org/abs/2501.17520 #stats #statistics #probabilities #machineLearning #featureSelection

arXiv.org

Conditional Feature Importance revisited: Double Robustness, Efficiency and Inference

Conditional Feature Importance (CFI) is a classical variable importance measure that accounts for the relationship between the studied feature and the others. However, CFI has not yet been studied from a theoretical perspective because the conditional sampling step has generally been considered a purely practical problem. In this article, we demonstrate that the recent Conditional Permutation Importance (CPI) is indeed a valid implementation of this concept. Under the conditional null hypothesis

0

Eric Maugendre about data @maugendre@hachyderm.io

· 47mo ago

😃 #Microsoft and #GitHub are asked to tell how training their #machineLearning model with copylefted code is “fair use”:
❤️ The Software Freedom Conservancy, Inc. participates in the lawsuit against #Copilot. They explain: https://sfconservancy.org/news/2022/nov/04/class-action-lawsuit-filing-copilot/
________________________________
#extensions #dev #freesoftware #openSource #FLOSS #FOSS #OSS #licensing #forks #forking #governance #dataProtection #dataPrivacy #GithubLawsuit #Copyleft #publicData #openData #artificialIntelligence #softwareDevelopment

Software Freedom Conservancy

On the filing of the Class Action Law Suit over GitHub

Many of you are inquiring about a lawsuit filed yesterday afternoon by two “J Doe” Plaintiffs regarding the serious and ongoing GitHub Copilot problem which we have been working on for the last 18 months. This issue is dire and important, but includes many complex issues that intersect FOSS license compliance with moral questions of software freedom and the future of machine learning in human endeavor. Complex issues need careful, diligent, and community-oriented consideration and response.

0

4

0