Top Tweets for #InfluenceFunction
How to interpret predictions of #LLMs from a #DataCentricAI perspective?
The @icmlconf #ICML2024 work of @JingtanW @xiaoqiang_98 @ray_qiaorui et al. shows that Shapley offers a more robust & reliable approach for data attribution than #InfluenceFunction #LeaveOneOut (1/n).
You want to know more about the connection between the #influencefunction and the #Deltamethod ? Check out our #preprint we made with @camaringe @MJTSmith @aurelien_belot @mf_schomaker
@icon_lshtm @mirschnitzer @watzilei
All comments are welcome!
The variance of the #influencefunction usually is a good approximation to the variance of your estimator and you can easily calculate the asymptotic confience intervals in #rstats with our examples:
The #influencefunction wuantifies how robust your estimators are to external noise. For example, if you measure the values 1, 1.2, 1.09, 1.03 and then you have a typo and write 101 instead of 1.01 the sample mean becomes too high!
This happens because your IF is not bounded.
If the direction in which you take this directional derivative is that of a potential new value (i.e. you analyze how your estimate changes by adding one measurement) you end up with a famous function: the #influencefunction
¿Quieres saber + de cómo se conectan el #MetodoDelta y la #influencefunction y cómo usarlas para varianzas en #epi? Checa el #preprint que hicimos junto con @WATZILEI y los demás (que no encuentro en Twitter). Comentarios/sugerencias de cómo hacerlo + amigable son bienvenidas
Bueno pues resulta que la varianza de la #influencefunction suele ser una buena aproximación a la varianza de tu estimador por lo que es posible calcular intervalos de confianza asintóticos de la forma clásica siguiendo nuestros ejemplos de código en R:
La #influencefunction cuantifica qué tan robustos son tus estimadores al ruido. Por ejemplo si estás midiendo valores 1, 1.2, 1.09, 1.03 y alguien hace un error de dedo y te pasa el 101 en lugar del 1.01 ¡pum tu promedio se vuelve altísimo!
Esto pasa porque la IF no es acotada
Ahora, si la dirección de la derivada que eliges es la de un nuevo valor (es decir decides analizar cómo cambia tu estimador si agregas una nueva medición) acabas con una función famosa: la #influencefunction.
"The further away (social distance?) the grater is the influence." @GavinWince @PatrickBergy65 @Millie__Weaver @TransOrganicRE
#InfluenceFunction #ShadowGate #Individuation
Wonder how couterfactually augmented data relates to #CounterfactualInference and #influencefunction
Enjoyed chatting w. @samcharrington on @twimlai podcast. We cover the sate of the field, papers I found interesting, recent work in my group on (i) counterfactually-augmented data (https://t.co/OaWuZFXEyq) & (ii) non-ideal approaches to theorizing on fair ML (posting soon) & more
Last Seen Hashtags on Sotwe
페티시
Seen from Korea
nadiajocelyn
Seen from Singapore
5giugno
Seen from Italy
somali #wasmo
Seen from Netherlands
childrensoutingassociation
Seen from United States
incestcaption
Seen from Indonesia
Linalondon
Seen from Italy
สาวใหญ่รับงานระยอง
Seen from Thailand
naturelover
Seen from United States
outdoorcum
Seen from Netherlands
Trends for you
Most Popular Users
Elon Musk 
@elonmusk
240.1M followers
Barack Obama
@barackobama
119.3M followers
Donald J. Trump
@realdonaldtrump
111.6M followers
Cristiano Ronaldo
@cristiano
108.9M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
97.3M followers
NASA
@nasa
92.1M followers
Justin Bieber
@justinbieber
90.6M followers
KATY PERRY
@katyperry
86.8M followers
Taylor Swift
@taylorswift13
80.6M followers
Lady Gaga
@ladygaga
72.1M followers
Kim Kardashian
@kimkardashian
69.4M followers
YouTube
@youtube
68.6M followers
Virat Kohli
@imvkohli
68.5M followers
Bill Gates
@billgates
63.4M followers
The Ellen Show
@theellenshow
62.5M followers
CNN
@cnn
61.9M followers
Neymar Jr
@neymarjr
61M followers
X
@x
60.9M followers
CNN Breaking News
@cnnbrk
59.9M followers