intro.tex

% $Id: intro.tex,v 1.1.1.1 2007/10/03 22:28:18 hridesh Exp $

\section{Introduction}
With the increasing popularity of Machine Learning (ML) based systems, the verification and validation of such systems are necessary. Although there is a vast amount of research carried out on the non-ML validation framework, there is a few validation based research done on the ML-based systems due to the nature of probabilistic and complexity. Without a proper validation framework, one might ask about the way to hold ML-based systems being accountable to the expectation. For instance, the contract made with the end-user by exposing the trustworthiness of these systems in terms of accuracy. But this metric of trustworthiness is variant even if the whole experimental setup remains the same. In this study, we address such problems and have proposed a searching algorithm based on the users' intent to measure the near-optimal accuracy.

The recent works on this field can be primarily categorized into two sections, verifying a model to be accountable for the assigned task \cite{pulina2010abstraction,gehr2018ai2,du2018techniques,abdul2018trends,zhang2016understanding}  and holding the accountability by making ML models more robust \cite{wang2018formal,katz2017reluplex,jia2019taso}. The prior works have focused on validating the input influence \cite{datta2016algorithmic}, explaining the models to make the black-box system grayer. However, these systems either hold domain knowledge or model operation knowledge as the key to increase the explainability. In this study, we have combined these two type of knowledge and propose a system that can take the input dataset, model operations and users' choice as input and produce a near-optimal accuracy rather than a single value that changes every time an ML model has been trained with same dataset and same experimental setup. 
We leverage the information to propose a user intent based local search-based approach, \emph{ADNN} that verify the model's capability in terms of classification accuracy.
%While there are works on holding an ML-based system accountable for the assigned task, these works are primarily categorized into two divisions, accountability validation and increasing the explainability of such systems.  Our approach is obtaining the range of seed values, bias, and weight to verify the deep neural network (DNN). Specifically, the DNNs randomly generate these values whenever it trains. Moreover, with different sets of seed value, bias, and weight, we can obtain different output values. Therefore, if we can identify the range, from the lowest value to the highest value, of seed values, bias, and weight, we can acquire the range of output value. We will perform our method in multiple DNNs in the same dataset and get the range of the output of these DNNs. By comparing DNNs' out ranges, we can verify whether a DNN is good or bad.
\paragraph{Problem Statement. }
Our proposed work is focused on the image-based deep neural network (DNN) classifier. We have identified multiple pieces of evidence that with the same experimental setup, one DNN model can produce different output evaluation metrics e.g., accuracy. In the learning process, these models begin the operation with random initialization of several parameters e.g., weight, bias. With this randomly initialized value, the learning process continues that ends up producing different results for a different execution. To address such a problem, we are proposing a mining-based distribution learning and 
%specification language-based 
user intent-based approach that helps the user to hold a model structure accountable for the output.
%If we have a trained DNN, which includes model structure, seed value, bias, weight, we can have the same accuracy for multiple training sections. However, if we have only the model structure, the final accuracy can be different because each training section can create each different accuracy. Therefore, in this situation, we cannot identify which model is better just based on the output value like model's accuracy. However, it can be a different story if we have a range of output. For instance, accuracy range is not changed after we retrain the model; therefore, the range a reliable set of values to identify the quality of the DNN.

%When the inputs are transmitted between neurons, the weights are applied to the inputs and passed into an activation function along with the bias. With different set of input, we can obtain different outputs which can be accuracy or loss values. Therefore, After multiple times modifying this input set, we will obtain a range of output values. In a specific input set, if the output value is far from the output range acquired by using previous input sets, we can detect the erro

Our contribution to this study has been the following:
\begin{itemize}
    \item We have manually gone through the \emph{Keras} documentation to identify the randomly initialized parameters.
%\item We have proposed an adaptive simulated annealing approach that produces the optimum value of accuracy.
\item We have proposed a searching algorithm that modifies the weight and bias of the DNN model.
\item We have also proposed a user intent based technique that restricts a searching algorithm to achieve near-optimal accuracy in the aspect of maximum allowable time, trial, and gain in accuracy which can be provided by the user.
	%\item We have proposed a programming language that can validate the trustworthiness of an ML-based system in a closed range.
\end{itemize}