arXiv subjects

Yi Su

Publications and source records attributed to Yi Su.

At least 91 records · Page 5Linked to original sources

arXiv2020

Doubly robust off-policy evaluation with shrinkage

We propose a new framework for designing estimators for off-policy evaluation in contextual bandits. Our approach is based on the asymptotically optimal doubly robust estimator, but we shrink the importance weights to minimize a bound on the mean squared error, which results in a better bias-variance tradeoff in finite samples. We use this optimization-based framework to obtain three estimators: (a) a weight-clipping estimator, (b) a new weight-shrinkage estimator, and (c) the first shrinkage-based estimator for combinatorial action sets. Extensive experiments in both standard and combinatorial bandit benchmark problems show that our estimators are highly adaptive and typically outperform state-of-the-art methods.

cs.LG↗

arXiv2020

Adaptive Estimator Selection for Off-Policy Evaluation

We develop a generic data-driven method for estimator selection in off-policy policy evaluation settings. We establish a strong performance guarantee for the method, showing that it is competitive with the oracle estimator, up to a constant factor. Via in-depth case studies in contextual bandits and reinforcement learning, we demonstrate the generality and applicability of the method. We also perform comprehensive experiments, demonstrating the empirical efficacy of our approach and comparing with related approaches. In both case studies, our method compares favorably with existing methods.

cs.LG↗

arXiv2020

Off-policy Bandits with Deficient Support

Learning effective contextual-bandit policies from past actions of a deployed system is highly desirable in many settings (e.g. voice assistants, recommendation, search), since it enables the reuse of large amounts of log data. State-of-the-art methods for such off-policy learning, however, are based on inverse propensity score (IPS) weighting. A key theoretical requirement of IPS weighting is that the policy that logged the data has "full support", which typically translates into requiring non-zero probability for any action in any context. Unfortunately, many real-world systems produce support deficient data, especially when the action space is large, and we show how existing methods can fail catastrophically. To overcome this gap between theory and applications, we identify three approaches that provide various guarantees for IPS-based learning despite the inherent limitations of support-deficient data: restricting the action space, reward extrapolation, and restricting the policy space. We systematically analyze the statistical and computational properties of these three approaches, and we empirically evaluate their effectiveness. In addition to providing the first systematic analysis of support-deficiency in contextual-bandit learning, we conclude with recommendations that provide practical guidance.

cs.LG↗

arXiv2020

Accurate Tumor Tissue Region Detection with Accelerated Deep Convolutional Neural Networks

Manual annotation of pathology slides for cancer diagnosis is laborious and repetitive. Therefore, much effort has been devoted to develop computer vision solutions. Our approach, (FLASH), is based on a Deep Convolutional Neural Network (DCNN) architecture. It reduces computational costs and is faster than typical deep learning approaches by two orders of magnitude, making high throughput processing a possibility. In computer vision approaches using deep learning methods, the input image is subdivided into patches which are separately passed through the neural network. Features extracted from these patches are used by the classifier to annotate the corresponding region. Our approach aggregates all the extracted features into a single matrix before passing them to the classifier. Previously, the features are extracted from overlapping patches. Aggregating the features eliminates the need for processing overlapping patches, which reduces the computations required. DCCN and FLASH demonstrate high sensitivity (~ 0.96), good precision (~0.78) and high F1 scores (~0.84). The average time taken to process each sample for FLASH and DCNN is 96.6 seconds and 9489.20 seconds, respectively. Our approach was approximately 100 times faster than the original DCNN approach while simultaneously preserving high accuracy and precision.

cs.CV↗

arXiv2019

CAB: Continuous Adaptive Blending Estimator for Policy Evaluation and Learning

The ability to perform offline A/B-testing and off-policy learning using logged contextual bandit feedback is highly desirable in a broad range of applications, including recommender systems, search engines, ad placement, and personalized health care. Both offline A/B-testing and off-policy learning require a counterfactual estimator that evaluates how some new policy would have performed, if it had been used instead of the logging policy. In this paper, we identify a family of counterfactual estimators which subsumes most such estimators proposed to date. Our analysis of this family identifies a new estimator - called Continuous Adaptive Blending (CAB) - which enjoys many advantageous theoretical and practical properties. In particular, it can be substantially less biased than clipped Inverse Propensity Score (IPS) weighting and the Direct Method, and it can have less variance than Doubly Robust and IPS estimators. In addition, it is sub-differentiable such that it can be used for learning, unlike the SWITCH estimator. Experimental results show that CAB provides excellent evaluation accuracy and outperforms other counterfactual estimators in terms of learning performance.

cs.LG↗

arXiv2019

Model Adaptation via Model Interpolation and Boosting for Web Search Ranking

This paper explores two classes of model adaptation methods for Web search ranking: Model Interpolation and error-driven learning approaches based on a boosting algorithm. The results show that model interpolation, though simple, achieves the best results on all the open test sets where the test data is very different from the training data. The tree-based boosting algorithm achieves the best performance on most of the closed test sets where the test data and the training data are similar, but its performance drops significantly on the open test sets due to the instability of trees. Several methods are explored to improve the robustness of the algorithm, with limited success.

cs.LG↗

arXiv2018

Network estimation via graphon with node features

Estimating the probabilities of linkages in a network has gained increasing interest in recent years. One popular model for network analysis is the exchangeable graph model (ExGM) characterized by a two-dimensional function known as a graphon. Estimating an underlying graphon becomes the key of such analysis. Several nonparametric estimation methods have been proposed, and some are provably consistent. However, if certain useful features of the nodes (e.g., age and schools in social network context) are available, none of these methods was designed to incorporate this source of information to help with the estimation. This paper develops a consistent graphon estimation method that integrates the information from both the adjacency matrix itself and node features. We show that properly leveraging the features can improve the estimation. A cross-validation method is proposed to automatically select the tuning parameter of the method.

stat.ME↗

arXiv2018

Modeling Multidimensional User Relevance in IR using Vector Spaces

It has been shown that relevance judgment of documents is influenced by multiple factors beyond topicality. Some multidimensional user relevance models (MURM) proposed in literature have investigated the impact of different dimensions of relevance on user judgment. Our hypothesis is that a user might give more importance to certain relevance dimensions in a session which might change dynamically as the session progresses. This motivates the need to capture the weights of different relevance dimensions using feedback and build a model to rank documents for subsequent queries according to these weights. We propose a geometric model inspired by the mathematical framework of Quantum theory to capture the user's importance given to each dimension of relevance and test our hypothesis on data from a web search engine and TREC Session track.

cs.IR↗

arXiv2017

Combinatorics of symmetric plabic graphs

A plabic graph is a planar bicolored graph embedded in a disk, which satisfies some combinatorial conditions. Postnikov's boundary measurement map takes the space of positive edge weights of a plabic graph $G$ to a positroid cell in some totally nonnegative Grassmannian. In this note, we investigate plabic graphs which are symmetric about a line of reflection, up to reversing the colors of vertices. These symmetric plabic graphs arise naturally in the study of total positivity for the Lagrangian Grassmannian. We characterize various combinatorial objects associated with symmetric plabic graphs, and describe the subset of a Grassmannian which can be realized by symmetric weightings of symmetric plabic graphs.

math.CO↗

arXiv2017

A Novel Multi-task Deep Learning Model for Skin Lesion Segmentation and Classification

In this study, a multi-task deep neural network is proposed for skin lesion analysis. The proposed multi-task learning model solves different tasks (e.g., lesion segmentation and two independent binary lesion classifications) at the same time by exploiting commonalities and differences across tasks. This results in improved learning efficiency and potential prediction accuracy for the task-specific models, when compared to training the individual models separately. The proposed multi-task deep learning model is trained and evaluated on the dermoscopic image sets from the International Skin Imaging Collaboration (ISIC) 2017 Challenge - Skin Lesion Analysis towards Melanoma Detection, which consists of 2000 training samples and 150 evaluation samples. The experimental results show that the proposed multi-task deep learning model achieves promising performances on skin lesion segmentation and classification. The average value of Jaccard index for lesion segmentation is 0.724, while the average values of area under the receiver operating characteristic curve (AUC) on two individual lesion classifications are 0.880 and 0.972, respectively.

cs.CV↗

arXiv2016

A weighted edge-count two-sample test for multivariate and object data

Two-sample tests for multivariate data and non-Euclidean data are widely used in many fields. Parametric tests are mostly restrained to certain types of data that meets the assumptions of the parametric models. In this paper, we study a nonparametric testing procedure that utilizes graphs representing the similarity among observations. It can be applied to any data types as long as an informative similarity measure on the sample space can be defined. The classic test based on a similarity graph has a problem when the two sample sizes are different. We solve the problem by applying appropriate weights to different components of the classic test statistic. The new test exhibits substantial power gains in simulation studies. Its asymptotic permutation null distribution is derived and shown to work well under finite samples, facilitating its application to large datasets. The new test is illustrated through an analysis on a real dataset of network data.

stat.ME↗

arXiv2014

Electrical Lie Algebra of Classical Types

We investigate the structure of electrical Lie algebras of finite Dynkin type. These Lie algebras were introduced by Lam-Pylyavskyy in the study of \textit{circular planar electrical networks}. The corresponding Lie group acts on such networks via some combinatorial operations studied by Curtis-Ingerman-Morrow and Colin de Verdière-Gitler-Vertigan. Lam-Pylyavskyy studied the electrical Lie algebra of type $A$ of even rank in detail, and gave a conjecture for the dimension of electrical Lie algebras of finite Dynkin types. We prove this conjecture for all classical Dynkin types, that is, $A$, $B$, $C$, and $D$. Furthermore, we are able to explicitly describe the structure of the corresponding electrical Lie algebras as the semisimple product of the symplectic Lie algebra with its finite dimensional irreducible representations.

math.CO↗

arXiv2010

Induced subgraphs in sparse random graphs with given degree sequence

For any $S\subset [n]$, we compute the probability that the subgraph of $\mathcal{G}_{n,d}$ induced by $S$ is a given graph $H$ on the vertex set $S$. The result holds for any $d=o(n^{1/3})$ and is further extended to $\mathcal{G}_{\bf d}$, the probability space of random graphs with a given degree sequence $\bf d$.

math.CO↗

arXiv2010

The lattice of integer flows of a regular matroid

For a finite multigraph G, let Λ(G) denote the lattice of integer flows of G -- this is a finitely generated free abelian group with an integer-valued positive definite bilinear form. Bacher, de la Harpe, and Nagnibeda show that if G and H are 2-isomorphic graphs then Λ(G) and Λ(H) are isometric, and remark that they were unable to find a pair of nonisomorphic 3-connected graphs for which the corresponding lattices are isometric. We explain this by examining the lattice Λ(M) of integer flows of any regular matroid M. Let M_\bullet be the minor of M obtained by contracting all co-loops. We show that Λ(M) and Λ(N) are isometric if and only if M_\bullet and N_\bullet are isomorphic.

math.CO↗

arXiv2010

Structural Solutions For Additively Coupled Sum Constrained Games

We propose and analyze a broad family of games played by resource-constrained players, which are characterized by the following central features: 1) each user has a multi-dimensional action space, subject to a single sum resource constraint; 2) each user's utility in a particular dimension depends on an additive coupling between the user's action in the same dimension and the actions of the other users; and 3) each user's total utility is the sum of the utilities obtained in each dimension. Familiar examples of such multi-user environments in communication systems include power control over frequency-selective Gaussian interference channels and flow control in Jackson networks. In settings where users cannot exchange messages in real-time, we study how users can adjust their actions based on their local observations. We derive sufficient conditions under which a unique Nash equilibrium exists and the best-response algorithm converges globally and linearly to the Nash equilibrium. In settings where users can exchange messages in real-time, we focus on user choices that optimize the overall utility. We provide the convergence conditions of two distributed action update mechanisms, gradient play and Jacobi update.

cs.GT↗

arXiv2009

Dynamic Conjectures in Random Access Networks Using Bio-inspired Learning

This paper considers a conjecture-based distributed learning approach that enables autonomous nodes to independently optimize their transmission probabilities in random access networks. We model the interaction among multiple self-interested nodes as a game. It is well-known that the Nash equilibria in this game result in zero throughput for all the nodes if they take myopic best-response, thereby leading to a network collapse. This paper enables nodes to behave as intelligent entities which can proactively gather information, form internal conjectures on how their competitors would react to their actions, and update their beliefs according to their local observations. In this way, nodes are capable to autonomously "learn" the behavior of their competitors, optimize their own actions, and eventually cultivate reciprocity in the random access network. To characterize the steady-state outcome, the conjectural equilibrium is introduced. Inspired by the biological phenomena of "derivative action" and "gradient dynamics", two distributed conjecture-based action update mechanisms are proposed to stabilize the random access network. The sufficient conditions that guarantee the proposed conjecture-based learning algorithms to converge are derived. Moreover, it is shown that all the achievable operating points in the throughput region are essentially stable conjectural equilibria corresponding to different conjectures. We investigate how the conjectural equilibrium can be selected in heterogeneous networks and how the proposed methods can be extended to ad-hoc networks. Simulations verify that the system performance significantly outperforms existing protocols, such as IEEE 802.11 DCF protocol and the PMAC protocol, in terms of throughput, fairness, convergence, and stability.

cs.GT↗

arXiv2009

Linearly Coupled Communication Games

This paper discusses a special type of multi-user communication scenario, in which users' utilities are linearly impacted by their competitors' actions. First, we explicitly characterize the Nash equilibrium and Pareto boundary of the achievable utility region. Second, the price of anarchy incurred by the non-collaborative Nash strategy is quantified. Third, to improve the performance in the non-cooperative scenarios, we investigate the properties of an alternative solution concept named conjectural equilibrium, in which individual users compensate for their lack of information by forming internal beliefs about their competitors. The global convergence of the best response and Jacobi update dynamics that achieve various conjectural equilibria are analyzed. It is shown that the Pareto boundaries of the investigated linearly coupled games can be sustained as stable conjectural equilibria if the belief functions are properly initialized. The investigated models apply to a variety of realistic applications encountered in the multiple access design, including wireless random access and flow control.

cs.GT↗

arXiv2008

A New Perspective on Multi-user Power Control Games in Interference Channels

This paper considers the problem of how to allocate power among competing users sharing a frequency-selective interference channel. We model the interaction between selfish users as a non-cooperative game. As opposed to the existing iterative water-filling algorithm that studies the myopic users, this paper studies how a foresighted user, who knows the channel state information and response strategies of its competing users, should optimize its transmission strategy. To characterize this multi-user interaction, the Stackelberg equilibrium is introduced, and the existence of this equilibrium for the investigated non-cooperative game is shown. We analyze this interaction in more detail using a simple two-user example, where the foresighted user determines its transmission strategy by solving as a bi-level program which allows him to account for the myopic user's response. It is analytically shown that a foresighted user can improve its performance, if it has the necessary information about its competitors. Since the optimal solution of Stackelberg equilibrium is computationally prohibitive, we propose a practical low-complexity approach based on Lagrangian duality theory. Numerical simulations verify the performance improvements. Possible ways to acquire the required information and to extend the formulation to more than two users are also discussed.

cs.GT↗