Benjamin Lubin

Navigating heterogeneous processors with market mechanisms

Specialization of datacenter resources brings performance and energy improvements in response to the growing scale and diversity of cloud applications. Yet heterogeneous hardware adds complexity and volatility to latency-sensitive applications. A resource allocation mechanism that leverages architectural principles can overcome both of these obstacles. We integrate research in heterogeneous architectures with recent advances in multi-agent systems. Embedding architectural insight into proxies that bid on behalf of applications, a market effectively allocates hardware to applications with diverse preferences and valuations. Exploring a space of heterogeneous datacenter configurations, which mix server-class Xeon and mobile-class Atom processors, we find an optimal heterogeneous balance that improves both welfare and energy-efficiency. We further design and evaluate twelve design points along the Xeon-to-Atom spectrum, and find that a mix of three processor architectures achieves a 12× reduction in response time violations relative to equal-power homogeneous systems.

ICE: an iterative combinatorial exchange

We present the first design for a fully expressive iterative combinatorial exchange (ICE). The exchange incorporates a tree-based bidding language that is concise and expressive for CEs. Bidders specify lower and upper bounds on their value for different trades. These bounds allow price discovery and useful preference elicitation in early rounds, and allow termination with an efficient trade despite partial information on bidder valuations. All computation in the exchange is carefully optimized to exploit the structure of the bid-trees and to avoid enumerating trades. A proxied interpretation of a revealed-preference activity rule ensures progress across rounds. A VCG-based payment scheme that has been shown to mitigate opportunities for bargaining and strategic behavior is used to determine final payments. The exchange is fully implemented and in a validation phase.

ICE: an expressive iterative combinatorial exchange

We present the design and analysis of the first fully expressive, iterative combinatorial exchange (ICE). The exchange incorporates a tree-based bidding language (TBBL) that is concise and expressive for CEs. Bidders specify lower and upper bounds in TBBL on their value for different trades and refine these bounds across rounds. These bounds allow price discovery and useful preference elicitation in early rounds, and allow termination with an efficient trade despite partial information on bidder valuations. All computation in the exchange is carefully optimized to exploit the structure of the bid-trees and to avoid enumerating trades. A proxied interpretation of a revealed-preference activity rule, coupled with simple linear prices, ensures progress across rounds. The exchange is fully implemented, and we give results demonstrating several aspects of its scalability and economic properties with simulated bidding strategies.

Strategies for anticipating risk in heterogeneous system design

Heterogeneous design presents an opportunity to improve energy efficiency but raises a challenge in resource management. Prior design methodologies aim for performance and efficiency, yet a deployed system may miss these targets due to run-time effects, which we denote as risk. We propose design strategies that explicitly aim to mitigate risk. We introduce new processor selection criteria, such as the coefficient of variation in performance, to produce heterogeneous configurations that balance performance risks and efficiency rewards. Out of the tens of strategies we consider, risk-aware approaches account for more than 70% of the strategies that produce systems with the best service quality. Applying these risk-mitigating strategies to heterogeneous datacenter design can produce a system that violates response time targets 50% less often.

Market mechanisms for managing datacenters with heterogeneous microarchitectures

Specialization of datacenter resources brings performance and energy improvements in response to the growing scale and diversity of cloud applications. Yet heterogeneous hardware adds complexity and volatility to latency-sensitive applications. A resource allocation mechanism that leverages architectural principles can overcome both of these obstacles. We integrate research in heterogeneous architectures with recent advances in multi-agent systems. Embedding architectural insight into proxies that bid on behalf of applications, a market effectively allocates hardware to applications with diverse preferences and valuations. Exploring a space of heterogeneous datacenter configurations, which mix server-class Xeon and mobile-class Atom processors, we find an optimal heterogeneous balance that improves both welfare and energy-efficiency. We further design and evaluate twelve design points along the Xeon-to-Atom spectrum, and find that a mix of three processor architectures achieves a 12× reduction in response time violations relative to equal-power homogeneous systems.

Access is Not Enough: Characteristics of Physicians Who Treat Medicaid Patients.

Background:Access to physicians is a major concern for Medicaid programs. However, little is known about relationships between physician participation in Medicaid and the individual-level and practice-level characteristics of physicians. Methods:We used the 2011 Massachusetts All Payer Claims Database, containing all commercial and Medicaid claims; we linked with data on physician characteristics. We measured Medicaid participation intensity (fraction of the physician’s patient panel with Medicaid) for primary care physicians (PCPs) and medical specialists. We measured influence of physicians within a patient referral network using eigenvector centrality. We used regression models to associate Medicaid intensity with physician individual-level and practice-level characteristics. Findings:About 92.6% of physicians treated at least 1 Medicaid patient, but the median physician’s panel contained only 5.7% Medicaid patients. Medicaid intensity was associated with physician training and influence for PCPs and specialists. For medical specialists, a 1 percentage point increase in Medicaid intensity was associated with a lower probability of being board certified (−0.22 percentage points; 95% CI, −0.30, −0.14), lower probability of attending a domestic medical school (−0.14 percentage points; 95% CI, −0.22, −0.05), having attended a less well-ranked domestic medical school (0.23 ranks; 95% CI, 0.15, 0.30), and having slightly less influence in the referral network. PCPs displayed similar results but high Medicaid intensity physicians had substantially less influence in the referral network. Conclusions:Medicaid participation intensity shows substantial variation across physicians, indicating limits of binary participation measures. Physicians with more Medicaid patients had characteristics often perceived by patients to be of lower quality.

Combinatorial Auctions via Machine Learning-based Preference Elicitation

Combinatorial auctions (CAs) are used to allocate multiple items among bidders with complex valuations. Since the value space grows exponentially in the number of items, it is impossible for bidders to report their full value function even in medium-sized settings. Prior work has shown that current designs often fail to elicit the most relevant values of the bidders, thus leading to inefficiencies. We address this problem by introducing a machine learning-based elicitation algorithm to identify which values to query from the bidders. Based on this elicitation paradigm we design a new CA mechanism we call PVM, where payments are determined so that bidders’ incentives are aligned with allocative efficiency. We validate PVM experimentally in several spectrum auction domains, and we show that it achieves high allocative efficiency even when only few values are elicited from the bidders.

The Role of Organizational Affiliations in Physician Patient-Sharing Relationships:

Provider consolidation may enable improved care coordination, but raises concerns about lack of competition. Physician patient-sharing relationships play a key role in constructing patient care teams, but it is unknown how organization affiliations affect these. We use the Massachusetts All Payer Claims Database to examine whether patient-sharing relationships are associated with sharing a practice site, medical group, and/or physician contracting network. Physicians were 17 percentage points more likely to have a patient-sharing relationship if they shared a practice site and 4 percentage points more likely if they shared a medical group, as compared with sharing no affiliation. However, there was no detectable increased probability of a patient-sharing relationship within the same physician contracting network. Our finding that physician patient-sharing relationships are concentrated within organizational boundaries at practice site and medical group levels helps illuminate referral incentives and provide insight into the role of organizational affiliations in patient care team construction.

Computing Bayes-Nash Equilibria in Combinatorial Auctions with Continuous Value and Action Spaces

Combinatorial auctions (CAs) are widely used in practice, which is why understanding their incentive properties is an important problem. However, finding Bayes-Nash equilibria (BNEs) of CAs analytically is tedious, and prior algorithmic work has only considered limited solution concepts (e.g. restricted action spaces). In this paper, we present a fast, general algorithm for computing symmetric pure ε-BNEs in CAs with continuous values and actions. In contrast to prior work, we separate the search phase (for finding the BNE) from the verification step (for estimating the ε), and always consider the full (continuous) action space in the best response computation. We evaluate our method in the well-studied LLG domain, against a benchmark of 16 CAs for which analytical BNEs are known. In all cases, our algorithm converges quickly, matching the known results with high precision. Furthermore, for CAs with quasi-linear utility functions and independently distributed valuations, we derive a theoretical bound on ε. Finally, we introduce the new Multi-Minded LLLLGG domain with eight goods and six bidders, and apply our algorithm to finding an equilibrium in this domain. Our algorithm is the first to find an accurate BNE in a CA of this size.

The Impact of Partial-Year Enrollment on the Accuracy of Risk Adjustment Systems: A Framework and Evidence

Accurate risk adjustment facilitates health-care market competition. Risk adjustment typically aims to predict annual costs of individuals enrolled in an insurance plan for a full year. However, partial-year enrollment is common and poses a challenge to risk adjustment, since diagnoses are observed with lower probability when an individual is observed for a shorter time. Because of missed diagnoses, risk-adjustment systems will underpay for partial-year enrollees, as compared with full-year enrollees with similar underlying health status and usage patterns. We derive a new adjustment for partial-year enrollment in which payments are scaled up for partial-year enrollees’ observed diagnoses, which improves upon existing methods. We simulate the role of missed diagnoses using a sample of commercially insured individuals and the 2014 Marketplace risk-adjustment algorithm and find the expected spending of six-month enrollees is underpredicted by 19 percent. We then examine whether there are systematically different care usage patterns for partial-year enrollees in this data, which can offset or amplify underprediction due to missed diagnoses. Accounting for differential spending patterns of partial-year enrollees does not substantially change the underprediction for six-month enrollees. However, one-month enrollees use systematically less than one-twelfth the care of full-year enrollees, partially offsetting the missed-diagnosis effect.