diff options
Diffstat (limited to 'main.tex')
| -rw-r--r-- | main.tex | 6 |
1 files changed, 4 insertions, 2 deletions
@@ -1,6 +1,6 @@ \subsection{D-Optimality Criterion} -Ideally, motivated by the $D$-optimality criterion, we would like to design a mechanism that maximizes \eqref{dcrit} within a good approximation ratio. As \eqref{dcrit} may take arbitrarily small negative values, to define a meaningful approximation we consider with the (equivalent) maximization of $V(S) = f(\det\T{X_S}X_S )$, for some increasing, on-to function $f:\reals_+\to\reals_+$. However, the following lower bound implies that such an optimization goal cannot be attained under the costraints of truthfulness, budget feasibility, and individional rationallity. +Ideally, motivated by the $D$-optimality criterion, we would like to design a mechanism that maximizes \eqref{dcrit} within a good approximation ratio. As \eqref{dcrit} may take arbitrarily small negative values, to define a meaningful approximation we consider the (equivalent) maximization of $V(S) = f(\det\T{X_S}X_S )$, for some increasing, on-to function $f:\reals_+\to\reals_+$. However, the following lower bound implies that such an optimization goal cannot be attained under the costraints of truthfulness, budget feasibility, and individional rationallity. \begin{lemma} For any $M>1$, there is no truthful, budget feasible, individionally rational mechanism for optimal mechanism design with value fuction $V(S) = \det{\T{X_S}X_S}$. @@ -9,7 +9,9 @@ For any $M>1$, there is no truthful, budget feasible, individionally rational me \input{proof_of_lower_bound1} \end{proof} -This motivates us to look at $$V(S) = \log\det(I_d+\T{X_S}X_S).$$ Interesting for many reasons. Experiment with basis points $e_1$,\ldots,$e_d$, already given for free. Connections to Baysian optimal experiment design: we explore this more in Section~\ref{...}. Close to $D$-optimality criterion when number of experiments is large. Important properties $V(\emptyset)=0$, $V$ is submodular, allows us to exploit the arsenal in our disposal to deal with budget feasible mechanism design for submodular functions. +This negative result motivates us to look at the following modified objective: +\begin{align}V(S) = \log\det(I_d+\T{X_S}X_S), \label{modified}\end{align} where $I_d\in \reals^{d\times d}$ is the identity matrix. +One possible interpretation of \eqref{modified} is that, prior to the auction, the experimenter has free access to $d$ experiments whose features form an ortho-normal basis in $\reals^d$. However, \eqref{modified} can also be motivated in the context of \emph{Bayesian experimental design} \cite{chaloner1995bayesian}. In short, the objective \eqref{modified} arises naturally when the experimenter retrieves the model $\beta$ through \emph{ridge regression}, rather than the linear regression \eqref{leastsquares} over the observed data; we explore this connection in Section~\ref{sec:bed}. From a practical standpoint, \eqref{modified} is a good approximation of \eqref{dcrit} when the number of experiments is large. Crucially, \eqref{modified} is submodular and satifies $V(\emptyset) = 0$, allowing us to use the extensive machinery for the optimization of submodular functions, as well as recent results in the context of budget feasible auctions. \subsection{Truthful, Constant Approximation Mechanism} |