The Experts below are selected from a list of 3345 Experts worldwide ranked by ideXlab platform
Robert A Beauregard - One of the best experts on this subject based on the ideXlab platform.
-
Urban Population Loss in historical perspective: United States, 1820 – 2000
Environment and Planning A, 2020Co-Authors: Robert A BeauregardAbstract:Employing an historical perspective, the author mounts a quantitative and theoretical assessment of Population Loss in the large cities of the United States. Three periods are considered: one prior to 1920 when large city Population Loss was aberrant; a second which captures the severe decline of the decades after World War II, and a third that encompasses the more recent shrinkage of cities. Population Loss is measured in terms of prevalence, severity, and persistence and is also analyzed geographically. The author further identifies factors affecting Population Loss which are common and unique to each period. Although Population Loss has diminished, a number of cities are locked into trajectories of chronic Loss, suggesting that a new phase of urbanization has yet to materialize.
-
Federal policy and postwar urban decline: A case of government complicity?
Housing Policy Debate, 2020Co-Authors: Robert A BeauregardAbstract:Many urban commentators have implicated the federal government in the decline of central cities in the decades just after World War II. They claim that federal policies disproportionately favored suburban development over much needed urban redevelopment and exacerbated the deconcentration and decentralization of people and capital. Close scrutiny reveals flaws in this argument and four of them are examined in this article: the core premise that suburban growth and Population Loss in the central cities are inversely related, the lack of attention to the actual chronology of events, the failure to address the geographic incidence of Population Loss from the central cities, and the deemphasizing of the role of the private sector, often acting with government support. The article concludes with a brief reflection on the tenacity of the complicity claim.
-
urban Population Loss in historical perspective united states 1820 2000
Environment and Planning A, 2009Co-Authors: Robert A BeauregardAbstract:Employing an historical perspective, the author mounts a quantitative and theoretical assessment of Population Loss in the large cities of the United States. Three periods are considered: one prior to 1920 when large city Population Loss was aberrant; a second which captures the severe decline of the decades after World War II, and a third that encompasses the more recent shrinkage of cities. Population Loss is measured in terms of prevalence, severity, and persistence and is also analyzed geographically. The author further identifies factors affecting Population Loss which are common and unique to each period. Although Population Loss has diminished, a number of cities are locked into trajectories of chronic Loss, suggesting that a new phase of urbanization has yet to materialize.
-
aberrant cities urban Population Loss in the united states 1820 1930
Urban Geography, 2003Co-Authors: Robert A BeauregardAbstract:Our understanding of Population Loss from U.S. cities draws primarily from the fate of industrial centers in the decades following World War II. Quite numerous, those cities cast off residents at unprecedented and sustained rates. Prior to this time, few large cities had ended a decade smaller in Population size than they began. In order to broaden and deepen our knowledge of why some cities and not others lose Population, this paper analyzes cities that shed Population in the 19th century. Using Census data and capsule stories developed from city biographies, the paper explores both contextuating and precipitating factors. These findings subsequently become the basis for reflecting anew on urban decline since the mid-20th century.
Abhradeep Thakurta - One of the best experts on this subject based on the ideXlab platform.
-
NeurIPS - Private Stochastic Convex Optimization with Optimal Rates
2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d.~samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
-
Private Stochastic Convex Optimization with Optimal Rates.
arXiv: Learning, 2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d. samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
Raef Bassily - One of the best experts on this subject based on the ideXlab platform.
-
NeurIPS - Private Stochastic Convex Optimization with Optimal Rates
2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d.~samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
-
Private Stochastic Convex Optimization with Optimal Rates.
arXiv: Learning, 2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d. samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
Vitaly Feldman - One of the best experts on this subject based on the ideXlab platform.
-
NeurIPS - Private Stochastic Convex Optimization with Optimal Rates
2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d.~samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
-
Private Stochastic Convex Optimization with Optimal Rates.
arXiv: Learning, 2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d. samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
Kunal Talwar - One of the best experts on this subject based on the ideXlab platform.
-
NeurIPS - Private Stochastic Convex Optimization with Optimal Rates
2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d.~samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.
-
Private Stochastic Convex Optimization with Optimal Rates.
arXiv: Learning, 2019Co-Authors: Raef Bassily, Vitaly Feldman, Kunal Talwar, Abhradeep ThakurtaAbstract:We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the Population Loss given i.i.d. samples from a distribution over convex and Lipschitz Loss functions. A long line of existing work on private convex optimization focuses on the empirical Loss and derives asymptotically tight bounds on the excess empirical Loss. However a significant gap exists in the known bounds for the Population Loss. We show that, up to logarithmic factors, the optimal excess Population Loss for DP algorithms is equal to the larger of the optimal non-private excess Population Loss, and the optimal excess empirical Loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.