Biases and artifacts in training data can cause unwelcome behavior in text classifiers (such as shallow pattern matching), leading to lack of generalizability. One solution to this problem is to include users in the loop and leverage their feedback t
o improve models. We propose a novel explanatory debugging pipeline called HILDIF, enabling humans to improve deep text classifiers using influence functions as an explanation method. We experiment on the Natural Language Inference (NLI) task, showing that HILDIF can effectively alleviate artifact problems in fine-tuned BERT models and result in increased model generalizability.
We show that a general algorithm for efficient computation of outside values under the minimum of superior functions framework proposed by Knuth (1977) would yield a sub-exponential time algorithm for SAT, violating the Strong Exponential Time Hypothesis (SETH).
In this paper, we discuss the completely monotonic functions and their relation to some of
the famous special functions such as (Gamma, Kumar, Parabolic cylinder, Gauss
hypergeometric, MacDonald, Whittaker and Generalized Mittag-Leffler) function.
In
addition, the relationship of the completely monotonic integrations with absolute progress
under conditions of convergence such as transformations (Hankel, Lambert, Stieltjes and
Laplace).
We will found other modes of composite functions given in terms of non-negative power
chains and integrative transformations of completely monotonic non-negative functions,
the state of integrative transform functions with a homogeneous nucleus of the first order,
and the logarithmically completely monotonic functions.
The importance of the row of completely monotonic functions that are associated with the
transformation of the Stieltjes defined as a class of special functions regression functions.
Some of the oscillations of these functions resulting from completely monotonic functions
are not decreasing or convex, but most of them are completely monotonic functions.
It is often useful to replace a function with a sequence of smooth functions
approximating the given function to resolve minimizing optimization problems.
The most famous one is the Moreau envelope. Recently the function was organized
using the Br
egman distance h D . It is worth noting that Bregman distance h D is
not a distance in the usual sense of the term. In general, it is not symmetric and it
does not satisfy the triangle inequality
The purpose of the research is to study the convergence of the Moreau envelope
function and the related proximal mapping depends on Bregman Distance for a
function on Banach space. Proved equivalence between Mosco-epi-convergence of
sequence functions and pointwise convergence of Moreau-Bregman envelope We
also studied the strong and weak convergence of resolvent operators According to
the concept of Bregman distance.
In this research paper, we compare the basic cache algorithms in terms of
performance and speed for the purposes of web caching for dynamic content
and hard disk buffering purposes, by studying the traditional algorithms in
this field, in order to
determine the utilization of the basic algorithms in disk
storage in the field of web caching. The results shows that algorithms with
replacement functions that rely on basic indicators (such as LRU, LFU) give
better results in storage for storage purposes in hard drives, while web
caching algorithms need additional benchmarks for replacement work to get
high performance indicators, Web Cache algorithms also show lower
performance then that hard drive, so the need to constantly develop the Web
cache algorithm.
This research aims to conduct a descriptive and econometric
analysis of the costs functions of rain fed barley in the first, second and
third stabilization zones in Al - Hasakah governorate, and to determine
the optimum sizes for production and th
e profit-maximizing size. Data
were collected through a questionnaire for rain fed barley farmers in the
study area for the average season (2015 / 2016-2016 / 2017).
In this work , The fifth order non-polynomial spline functions is
used to solve linear volterra integral equations with weakly
singular kernel .
Numerical examples are presented to illustrate the applications
of this method and to compare the computed results with
other numerical methods.
This research tries to concentrate on an essential and important issue in symbolic
logic, which is the calculus of propositions in Rassell's logic.
By studying this issue we will be able to understand propositions in Rassell's logic
and its relati
onship with facts because of the great relationship between them. also we are
going to know his opinion of Aristote's Categorical proposition.
This research aims to study the issue of simple truth functions which are treated by
calculus of propositions and deals with it as axiomatic and clarify the difference between
material implication and formal implication.
Finally it studies the axiomatic in calculus of propositions depending on the book
principles of mathematics.
In this paper , we will discuss the way of construction of lyapunov
function for some of linear stochastic difference equations
We will use the general method of constructions of lyapunov
function for stochastic difference equations and we will ob
tain a
sufficient conditions of asymptotic mean square stability of zero
solution for one of linear stochastic difference equations with
constant coefficient ,By using of some main theorems and
definitions for asymptotic mean square stability for linear
stochastic difference equations .
In this paper we study some basic properties of the Moreau-Yosida function with two variables , and generalize the results of related to study of the convergence for sequence of convex-concave functions and the sequence of Moreau-Yosida function corr
esponding , and the basic theorem that we proved is : for any sequence of convex-concave functions , if they are convergent of the Moreau-Yosida distance then the sequence of Moreau-Yosida function corresponding will be convergent to the concept of Mosco-epi/hypo graph convergence .
