Dean P. Foster

Statistical Relational Learning at U Penn (2008)

We consider the contribution of parameters to the stochastic complexity. The stochastic complexity of a class of models is the length of a universal, one-part code representing this class. It...

Alexandrin Popescul, Dean P. Foster, Lyle H. Ungar

We do statistical relational learning by incrementally extracting data from a relational database, and computing features of that data which are then used in a classical discriminative statistical...

Efficient Feature Selection in the Presence of Multiple Feature Classes (2008)

Dhillon, Paramveer Singh, Foster, Dean P, Ungar, Lyle H

We present an information theoretic approach to feature selection when the data possesses feature classes. Feature classes are pervasive in real data. For example, in gene expression data, the genes...

Information Consistency of Nonparametric Gaussian Process Methods (2008)

Matthias W. Seeger, Sham M. Kakade, Dean P. Foster

Abstract — This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version will be superseded. Bayesian nonparametric models...

That Is Competitive with Bayes Experts (2008)

Unsupervised distance metric learning using predictability (2008)

Gupta, Abhishek A., Foster, Dean P., Ungar, Lyle H.

Distance-based learning methods, like clustering and SVMs, are dependent on good distance metrics. This paper does unsupervised metric learning in the context of clustering. We seek transformations...

Calibration via Regression (2008)

Multi-View Regression via Canonical Correlation Analysis (2008)

Abstract — In the online prediction setting, the concept of calibration entails having the empirical (conditional) frequencies match the claimed predicted probabilities. This contrasts with more...

Sham M. Kakade, Dean P. Foster

Abstract. In the multi-view regression problem, we have a regression problem where the input variable (which is a real vector) can be partitioned into two different views, where it is assumed that...

Unsupervised distance metric learning using (2008)

Abhishek A. Gupta, Dean P. Foster, Lyle H. Ungar

predictability

Abstract (2008)

Sham M. Kakade, Matthias W. Seeger, Dean P. Foster

We present a competitive analysis of some non-parametric Bayesian algorithms in a worst-case online learning setting, where no probabilistic assumptions about the generation of the data are made. We...

Information Consistency of Nonparametric Gaussian Process Methods (2008)

Seeger, Matthias, Kakade, Sham, Foster, Dean P.

Bayesian nonparametric models are widely and successfully used for statistical prediction. While posterior consistency properties are well studied...

Dean Foster and Peyton Young explain a dramatic problem in the hedge fund industry that allows hedge fund managers to profit hugely while exposing their investors to great unknown risks.

Dean Foster and Peyton Young explain a dramatic problem in the hedge fund industry that allows hedge fund managers to profit hugely while exposing their investors to great unknown risks.

Dean Foster and Peyton Young explain a dramatic problem in the hedge fund industry that allows hedge fund managers to profit hugely while exposing their investors to great unknown risks.

Sham M. Kakade, Matthias W. Seeger, Dean P. Foster

Dean Foster and Peyton Young explain a dramatic problem in the hedge fund industry that allows hedge fund managers to profit hugely while exposing their investors to great unknown risks.

Abstract (2008)

We present a competitive analysis of some non-parametric Bayesian algorithms in a worst-case online learning setting, where no probabilistic assumptions about the generation of the data are made. We...

Statistical Relational Learning at U Penn (2008)

Alexandrin Popescul, Dean P. Foster, Lyle H. Ungar

We do statistical relational learning by incrementally extracting data from a relational database, and computing features of that data which are then used in a classical discriminative statistical...

Journal of Machine Learning Research 7 (2006) 1861-1885 Submitted 8/05; Revised 6/06; Published 9/06 Streamwise Feature Selection (2008)

Jing Zhou, Dean P. Foster, Robert A. Stine, Lyle H. Ungar, Isabelle Guyon

In streamwise feature selection, new features are sequentially considered for addition to a predictive model. When the space of potential features is large, streamwise feature selection offers many...

for many illuminating discussions, and Elchanan Ben-Porath, Eddie Dekel, (2008)

Dean P. Foster, Sergiu Hart, Tom Ferguson, Martin Hellwig, Eugene K, Paul Klemperer, ...

We propose a measure of riskiness of “gambles ” (risky assets) that is objective: it depends only on the gamble. The measure is based on identifying for every gamble the critical wealth level...

The Hedge Fund Game: Incentives, Excess Returns, and Piggy‐Backing (2008)

Dean P. Foster, H. Peyton Young, Joe Perkins, Kislaya Prasad, Tarun Ramadorai, Neil Shepard, ...

We show that it is very difficult to structure incentive schemes that distinguish between unskilled hedge fund managers, who cannot generate excess returns, and highly skilled managers who can...

A Martingale Test for Alpha (2008)

Dean P. Foster, Robert Stine, H. Peyton Young

We present a new method for testing whether a fund manager’s track record allows us to infer that he is able to beat the market with high probability or is just plain lucky. The test is based on...

Characterizing the generalization performance of model selection strategies (2007)

Lyle H. Ungar, Dean P. Foster

Abstract: We investigate the structure of model selection problems via the bias/variance decomposition. In particular, we characterize the essential structure of a model selection task by the bias...

Workshop on Learning Statistical Models from Relational Data, IJCAI-2003 Statistical Relational Learning at U Penn (2007)

Alexandrin Popescul, Dean P. Foster, Lyle H. Ungar

We do statistical relational learning by incrementally extracting data from a relational database, and computing features of that data which are then used in a classical discriminative statistical...

Local Asymptotics and the Minimum Description Length (2007)

Dean P. Foster, Rakesh Vohra

Common approximations for the minimum description length (MDL) criterion im-ply that the cost of adding a parameter to a model fit to n observations is about (1/2) log n bits. While effective for...

1 (2007)

At each point in time a decision maker must choose a decision. The payoff in a period from the decision chosen depends on the decision as well as the state of the world that obtains at that time. The...

Learning, hypothesis testing, and Nash equilibrium (2007)

Adaptive Variable Selection with Bayesian Oracles ∗ (2007)

Although there exist learning processes for which the empirical distribution of play comes close to Nash equilibrium, it is an open question whether the players themselves can learn to play...

Honest Confidence Intervals for the Error Variance in Stepwise Regression (2007)

We analyze the performance of adaptive variable selection with the aid of a Bayesian oracle. A Bayesian oracle supplies the statistician with a distribution for the unknown model parameters, here the...

Draft. The current version is available from www-stat.wharton.upenn.edu/bob (2007)

An honest confidence interval for the error variance in a stepwise regression is a one-sided interval that adjusts for the effects of variable selection. The endpoint of this interval may be many...

Dean P. Foster, Howard Karloff, Yuval Rabani, Yiftach Ravid, Sundar Vishwanathan

We analyze the performance of adaptive variable selection with the aid of a Bayesian oracle. A Bayesian oracle supplies the statistician with a distribution for the unknown model parameters, here the...

z (2007)

A layered graph is a connected graph whose vertices are partitioned into sets L 0 = fsg; L 1; L 2;:::, and whose edges, which have nonnegative integral weights, run between consecutive layers. Its...

Workshop on Learning Statistical Models from Relational Data, IJCAI-2003 (2007)

Statistical Relational Learning, Alexandrin Popescul, Dean P. Foster, Lyle H. Ungar

ide powerful modeling component but are often limited to a "flat" file propositional domain representation where potential features are fixed-size attribute vectors. Often the manual...

Alpha-investing: A procedure for sequential control of expected (2007)

Early Retirement Using Leveraged Investments (2007)

false discoveries

Dean P. Foster, Sham M. Kakade, Orit Ronen

We study the problem of how one should invest for retirement, focusing on the risk/return tradeoff that must be made over time. In particular, we are interested in the use of financial instruments...

Multi-view regression via canonical correlation analysis (2007)

Sham M. Kakade, Dean P. Foster

Abstract. In the multi-view regression problem, we have a regression problem where the input variable (which is a real vector) can be partitioned into two different views, where it is assumed that...

Acknowledgements: We are indebted to Andrew Lo, Tom Norman, Andrew (2007)

Dean P. Foster, H. Peyton Young, Joe Perkins, Tarun Ramadorai, Krishna Ramaswamy, Neil Shephard

We show that it is extremely difficult to devise incentive schemes that distinguish between fund managers who cannot deliver excess returns from those who can, unless investors have specific...

Regret testing: learning to play Nash equilibrium without knowing you have an opponent (2006)

Dean P. Foster; Wharton School, University Of Pennsylvania, H. Peyton Young; Johns Hopkins University And University Of Oxford

[This item is a preserved copy. To view the original, visit http://econtheory.org/] A learning rule is uncoupled if a player does not condition his strategy on the opponent's payoffs. It is radically...

Streamwise Feature Selection (2006)

Zhou, Jing, Foster, Dean P, Stine, Robert A, Ungar, Lyle H.

In streamwise feature selection, new features are sequentially considered for addition to a predictive model. When the space of potential features is large, streamwise feature selection offers many...

Running title: Regret Testing (2006)

Honest Confidence Intervals for the Error Variance in Stepwise Regression (2006)

A learning rule is uncoupled if a player does not condition his strategy on the opponent’s payo¤s. It is radically uncoupled if a player does not condition his strategy on the opponent’s actions...

Modified 9/20/05 Being Warren Buffett: A Classroom Simulation of Risk and Wealth when Investing in the Stock Market (2005)

An honest confidence interval for the error variance in a stepwise regression is a one-sided interval that adjusts for the effects of variable selection. The endpoint of this interval may be many...

Dean P. Foster, Robert A Stine

Students who are new to Statistics and its role in modern Finance have a hard time making the connection between variance and risk. To link these, we developed a classroom simulation in which groups...

Streaming Feature Selection using IIC (2005)

Lyle Ungar And, Lyle H. Ungar, Jing Zhou, Dean P. Foster, Bob A. Stine

In Streaming Feature Selection (SFS), new features are sequentially considered for addition to a predictive model. When the space of potential features is large, SFS offers many advantages over...

Deterministic calibration and Nash equilibrium (2004)

Sham M. Kakade, Dean P. Foster

Abstract. We provide a natural learning process in which the joint frequency of empirical play converges into the set of convex combinations of Nash equilibria. In this process, all players...

Variable Selection in Data Mining: (2004)

Building Predictive Model, Dean P. Foster, Robert A. Stine

We develop and illustrate a methodology for fitting models to large, complex data sets. The methodology uses standard regression techniques that make few assumptions about the structure of the data....

Variable Selection in Data Mining: (2004)

Building Predictive Model, Dean P. Foster, Robert A. Stine

We predict the onset of personal bankruptcy using least squares regression. Although well publicized, only 2,244 bankruptcies occur in our data set of 2.9 million months of credit-card activity. We...

Variable selection in data mining: Building a predictive model for bankruptcy (2004)

Regret testing: A simple payoff-based procedure for learning Nash equilibrium (2004)

We predict the onset of personal bankruptcy using least squares regression. Although well publicized, only 2,244 bankruptcies occur in our data set of 2.9 million months of credit-card activity. We...

Being Warren Buffett: A classroom simulation of financial risk (2004)

constructive comments on an earlier draft. 1 2 A learning rule is uncoupled if a player does not condition his strategy on the opponent’s payoffs. It is radically uncoupled if the player does not...

Dean P. Foster, Robert A Stine

Students in business and other areas who are new to Statistics have a hard time making the connection between variance and risk. To convey the connection, we developed a classroom simulation. In the...

Universal codes for finite sequences of integers drawn from a monotone distribution (2002)

Dean P. Foster, Robert A. Stine, Abraham J. Wyner

We oer two noiseless codes for blocks of integers X

Universal codes for finite sequences of integers drawn from a monotone distribution (2002)

Dean P. Foster, Robert A. Stine, Abraham J. Wyner

Abstract We offer two noiseless codes for blocks of integers Xn = (X1; : : : ; Xn). We provide explicit bounds on the relative redundancy that are valid for any distribution F in the class of...

On the Impossibility of Predicting the Behavior of Rational Agents (2001)

On the Impossibility of Predicting the Behavior of Rational Agents (2001)

A foundational assumption in economics is that people are rational-- they choose optimal plans of action given their predictions about future states of the world. In games of strategy this means that...

On the Impossibility of Predicting the Behavior of Rational Agents (2001)

We exhibit a class of games that almost surely cannot be learned by perfectly rational players. That is, if players are uncertain about their opponents ’ payoffs, and if all of them are rational,...

Calibration and empirical Bayes variable selection (2000)

A foundational assumption in economics is that people are rational-- they choose optimal plans of action given their predictions about future states of the world. In games of strategy this means that...

I. George, Dean P. Foster

For the problem ofvariable selection for the normal linear model, selection criteria such as AIC, Cp, BIC and RIC have fixed dimensionality penalties. Such criteria are shown to correspond to...

Calibration and empirical Bayes variable selection (2000)

George, EdwardI., Foster, Dean P.

For the problem of variable selection for the normal linear model, selection criteria such as aic, Cp , bic and ric have fixed dimensionality penalties. Such criteria are shown to correspond to...

Accounting for Cognitive Costs in On-line Auction Design (1999)

David C. Parkes, Lyle H. Ungar, Dean P. Foster

Many auction mechanisms, including first and second price ascending and sealed bid auctions, have been proposed and analyzed in the economics literature. We compare the usefulness of different...

Cost and Trust Issues in On-Line Auctions (1998)

Lyle H. Ungar, David C. Parkes, Dean P. Foster

Abstract. Many auction mechanisms, including rst and second-price ascending and sealed-bid auctions, have been proposed and analyzed in the economics literature. We compare the usefulness of di erent...

A Formal Statistical Approach to Collaborative Filtering (1998)

Lyle H. Ungar, Dean P. Foster

Grouping people into clusters based on the items they have purchased allows accurate recommendations of new items for purchase: If you and I have liked many of the same movies, then I will probably...

Clustering methods for collaborative filtering (1998)

Lyle H. Ungar, Dean P. Foster, Ellen Andre, Star Wars, Fred Star Wars, Dean Star Wars, ...

Grouping people into clusters based on the items they have purchased allows accurate recommendations of new items for purchase: if you and I have liked many of the same movies, then I will probably...

Competitive Algorithms For Layered Graph Traversal (1998)

Amos Fiat, Dean P. Foster, Howard Karloff, Yuval Rabani

.<F3.84e+05> A layered graph is a connected graph whose vertices are partitioned into sets<F3.379e+05> L<F2.724e+05> 0<F3.84e+05>...

Cost and Trust Issues in On-Line Auctions (1998)

Lyle H. Ungar, David C. Parkes, Dean P. Foster

Many auction mechanisms, including first and second-price ascending and sealed-bid auctions, have been proposed and analyzed in the economics literature. We compare the usefulness of different...

Clustering Methods for Collaborative Filtering (1998)

Lyle H. Ungar, Dean P. Foster

Grouping people into clusters based on the items they have purchased allows accurate recommendations of new items for purchase: if you and I have liked many of the same movies, then I will probably...

A Formal Statistical Approach to Collaborative Filtering (1998)

Lyle Ungar, Dean P. Foster

Grouping people into clusters based on the items they have purchased allows accurate recommendations of new items for purchase: If you and I have liked many of the same movies, then I will probably...

Asymptotic calibration (1998)

FOSTER, DEAN P., VOHRA, RAKESH V.

Can we forecast the probability of an arbitrary sequence of events happening so that the stated probability of an event happening is close to its empirical probability? We can view this prediction...

Precision and accuracy of judgmental estimation (1997)

DRAFT An Information Theoretic Comparison of Model Selection Criteria (1997)

Journal of Behavioral Decision Making, Vol. 10, 21±32 (1997) Whereas probabilistic calibration has been a central normative concept of accuracy in previous research on interval estimates, we suggest...

Calibrated learning and correlated equilibrium (1997)

Information theory offers a coherent perspective on model selection. As in Rissanen’s original application of information theory to model selection, our perspective arises from viewing a model as a...

Dean P. Foster, Rakesh V. Vohra

Suppose two players meet each other in a repeated game where: 1. each uses a learning rule with the property that it is a calibrated forecast of the others plays, and 2. each plays a best response to...

Characterizing the Generalization Performance of Model Selection Strategies (1997)

Dale Schuurmans, Lyle H. Ungar, Dean P. Foster

: We investigate the structure of model selection problems via the bias/variance decomposition. In particular, we characterize the essential structure of a model selection task by the bias and...

Characterizing the Generalization Performance of Model Selection Strategies (1997)

Dale Schuurmans, Lyle H. Ungar, Dean P. Foster

We investigate the structure of model selection problems via the bias/variance decomposition. In particular, we characterize the essential aspects of a model selection task by the bias and variance...

ARTICLE NO. GA970595 Calibrated Learning and Correlated Equilibrium (1996)

Dean P. Foster, Rakesh V. Vohra

Suppose two players repeatedly meet each other to play a game where 1. each uses a learning rule with the property that it is a calibrated forecast of the other’s plays, and 2. each plays a myopic...

Continuous record asymptotics for rolling sample variance estimators (1996)

Dean P. Foster, Dan B. Nelson

It is widely known that conditional covariances of asset returns change over time. Researchers doing empirical work have adopted many strategies for accommodating conditional heteroskedasticity....

ARTICLE NO. GA970626 On the Nonconvergence of Fictitious Play in Coordination Games (1995)

Variable selection in Data Mining: Building a Predictive Model for Brankruptcy (0000)

It is shown by example that learning rules of the fictitious play type fail to converge in certain kinds of coordination games. Variants of fictitious play in which past actions are eventually...

Foster, Dean P.

We predict the onset of personal bankruptcy using least squares regression. Although well publicized, only 2,244 bankruptcies occur in our dataset of 2.9 million months of credit-card activity. We...

Regret Testing Leads to Nash Equilibrium

Dean P Foster, Peyton Young

On the impossibility of predicting the behavior of rational agents

On the impossibility of predicting the behavior of rational agents

A foundational assumption in economics is that people are rational: they choose optimal plans of action given their predictions about future states of the world. In games of strategy this means that...

On the Impossibility of Predicting the Behavior of Rational Agents

A foundational assumption in economics is that people are rational: they choose optimal plans of action given their predictions about future states of the world. In games of strategy this means that...

Calibration, Expected Utility and Local Optimality

A foundational assumption in economics is that people are rational -- they choose optimal plans of action given their predictions about future states of the world. In games of strategy this means...

Dean P. Foster, Rakesh V. Vohra

We propose a framework for reconciling frequentist and subjectivist views of probability. In an environment with repeated trails we show that beliefs about the possible states of nature can be...

An Operational Measure of Riskiness

Dean P. Foster, Sergiu Hart

We define the riskiness of a gamble g as that unique number R(g) such that no-bankruptcy is guaranteed if and only if one never accepts gambles whose riskiness exceeds the current wealth.

Being Warren Buffett: A Classroom Simulation of Risk and Wealth When Investing in the Stock Market

Foster, Dean P., Stine, Robert A.

The Hedge Fund Game: Incentives, Excess Returns, and Piggy-Backing

Peyton Young, Dean P Foster

We show that it is very difficult to structure incentive schemes that distinguish between unskilled hedge fund managers, who cannot generate excess returns, and highly skilled managers who can...

The Hedge Fund Game

This paper examines theoretical properties of incentive contracts in the hedge fund industry. We show that it is very difficult to structure incentive payments that distinguish between unskilled...

Variable selection in Data Mining: Building a Predictive Model for Brankruptcy

Foster, Dean P.

We predict the onset of personal bankruptcy using least squares regression. Although well publicized, only 2,244 bankruptcies occur in our dataset of 2.9 million months of credit-card activity. We...

Learning, hypothesis testing, and Nash equilibrium

Regret in the On-Line Decision Problem

Foster, Dean P., Vohra, Rakesh

A Proof of Calibration via Blackwell's Approachability Theorem

Foster, Dean P.

On the Nonconvergence of Fictitious Play in Coordination Games

Calibrated Learning and Correlated Equilibrium

Foster, Dean P., Vohra, Rakesh V.

"&agr;"-investing: a procedure for sequential control of expected false discoveries

A Proof of Calibration Via Blackwell's Approachability Theorem

"&agr;"-investing is an adaptive sequential methodology that encompasses a large family of procedures for testing multiple hypotheses. All control mFDR, which is the ratio of the expected number of...

An Information Theoretic Comparison of Model Selection Criteria

Over the past few years many proofs of calibration have been presented (Foster and Vohra (1991, 1997), Hart (1995), Fudenberg and Levine (1995), Hart and Mas-Colell (1996)). Does the literature...

Continuous Record Asymptotics for Rolling Sample Variance Estimators.

Information theory offers a coherent perspective on model selection. As in Rissanen's original application of information theory to model selection, our perspective arises from viewing a model as a...

Foster, Dean P, Nelson, Daniel B

It is widely known that conditional covariances of asset returns change over time. Researchers doing empirical work have adopted many strategies for accommodating conditional heteroskedasticity. One...

Asymptotic Filtering Theory for Univariate ARCH Models.

Nelson, Daniel B, Foster, Dean P

Researchers often employ ARCH models to estimate conditional variances and covariances. How successfully can misspecified ARCH models carry out this estimation? This paper employs continuous record...

Filtering and forecasting with misspecified ARCH models II : Making the right forecast with the wrong model

Nelson, Daniel B., Foster, Dean P.

Regret testing: learning to play Nash equilibrium without knowing you have an opponent

A proof of Calibration via Blackwell's Approachability Theorem

A learning rule is uncoupled if a player does not condition his strategy on the opponent's payoffs. It is radically uncoupled if a player does not condition his strategy on the opponent's actions or...

Dean P Foster

Filtering and Forecasting with Misspecified Arch Models II: Making the Right Forecast with the Wrong Model

Daniel B. Nelson, Dean P. Foster

A companion paper (Nelson (1992)) showed that in data observed at high frequencies, an ARCH model may do a good job at estimating conditional variances, even when the ARCH model is severely...

Asypmtotic Filtering Theory for Univariate Arch Models

Daniel B. Nelson, Dean P. Foster

This paper builds on this earlier work by deriving the asymptotic distribution of the measurement error. This allows us to approximate the measurement accuracy of ARCH conditional variance estimates...

Continuous Record Asymptotics for Rolling Sample Variance Estimators

Dean P. Foster, Daniel B. Nelson

It is widely known that conditional covariances of asset returns change over time. Researchers adopt many strategies to accommodate conditional heteroskedasticity. Among the most popular are: (a)...

Variable Selection in Data Mining: Building a Predictive Model for Bankruptcy

We develop and illustrate a methodology for fitting models to large, complex data sets. The methodology uses standard regression techniques that make few assumptions about the structure of the data....

Hedge Fund Wizards