There's a problem though. By clipping the tails, this approach introduces significant (statistical) bias. Indeed, as we analyze in a paper with @vkerdos @thejonullman (
https://
arxiv.org/abs/2002.09464), we get minimax rates for estimation by *balancing* the bias & variance from noise. 3/n
DATA
-
Balancing Bias and Variance in Statistical Estimation Methods
By
–
-
Private Mean Estimation: Clipping, Noise, and Differential Privacy
By
–
By now, we understand private mean estimation pretty well, especially in 1D. It turns out that one of the simplest algorithms, taking the empirical mean of the clipped samples (to restrict sensitivity) and adding noise (to introduce privacy) works pretty well. 2/n
-
Bias-Variance-Privacy Trilemma in Statistical Estimation
By
–
"A Bias-Variance-Privacy Trilemma for Statistical Estimation," with @argymouz
, Matthew Regehr, @vkerdos, @shortstein
, and @thejonullman
. https://
arxiv.org/abs/2301.13334 Private estimators MUST be biased! 1/n -
Stable Diffusion Overfitting: Realistic Image Generation Scenarios
By
–
Y por último, en ningún momento he negado que en Stable Diffusion, al igual que sucede en la mayoría de modelos de Deep Learning no vaya a haber overfitting. Lo que me interesa es saber qué tan real es un escenario donde un usuario U genere una imagen X* muy parecida al dato X.
-
Overfitting in AI Model Training: Statistical Analysis
By
–
Ahora el 2º tweet, donde hablamos de si hay o no overfitting – y no del objetivo del entrenamiento – El porcentaje de 109 imágenes de 175.000.000 puedes calcularlo, pero es ínfimo, mucho más bajo que ganar la lotería de navidad…
-
Training objectives: data reconstruction versus distribution learning
By
–
Como comenté, y de donde sale ese tweet, el objetivo de entrenar para "reconstruir una copia de los datos de entrenamientos" es apuntar a plagiar esos datos. Y no es lo mismo que el objetivo al que se apunta con estos modelos generativos: "aprender la distribución de los datos".
-
Co-Chief Scientist discusses content creation and source attribution in LLMs
By
–
Interview with our Co-Chief Scientist, @YoavLevine
: "When someone creates content, they want to be comfortable to put their name on it. And in certain use cases, the ability to connect the content to sources really facilitates this” https://
bdtechtalks.com/2023/01/30/ai2
1labs-llms-ralm/
… via @bdtechtalks -
Cloud Analytics and Automation Saving Lives Against Disease
By
–
You don't need to wear a cape to save lives. @IDDOnews does heroic feats with #CloudAnalytics. Join them at #AlterNext — Saving Lives with Cloud Analytics — to hear how they fight malaria, Ebola, and COVID-19 with the power of #AnalyticAutomation: http://
ow.ly/PM5S50MEXUp -
Data Strategy and Lakehouse Implementation for Real-Time Business Insights
By
–
Investing in your #data strategy should be at the top of your list Our CRO Ron Gabrisko & CFO Dave Conte discuss how implementing #Lakehouse provides a 360-degree view of your business for real-time decision-making
-
Data Connectivity Solutions for Industry 4.0 and Sustainability
By
–
Get insights on how to overcome the data connectivity challenges in OT and IT. Read “ A Data-Driven Approach to Sustainability in Industry 4.0 Using MQTT” http://
ow.ly/Ffqc50MlJvU @GregorianCT1 @BigDataMinded #sponsored #hivemq_iiot #industry40 #mqtt #digitaltransformation #data