arXiv subjects

Vitaly Kuznetsov

Publications and source records attributed to Vitaly Kuznetsov.

12 recordsLinked to original sources

arXiv2020

Discrepancy-Based Algorithms for Non-Stationary Rested Bandits

We study the multi-armed bandit problem where the rewards are realizations of general non-stationary stochastic processes, a setting that generalizes many existing lines of work and analyses. In particular, we present a theoretical analysis and derive regret guarantees for rested bandits in which the reward distribution of each arm changes only when we pull that arm. Remarkably, our regret bounds are logarithmic in the number of rounds under several natural conditions. We introduce a new algorithm based on classical UCB ideas combined with the notion of weighted discrepancy, a useful tool for measuring the non-stationarity of a stochastic process. We show that the notion of discrepancy can be used to design very general algorithms and a unified framework for the analysis of multi-armed rested bandit problems with non-stationary rewards. In particular, we show that we can recover the regret guarantees of many specific instances of bandit problems with non-stationary rewards that have been studied in the literature. We also provide experiments demonstrating that our algorithms can enjoy a significant improvement in practice compared to standard benchmarks.

cs.LG↗

arXiv2019

AdaNet: A Scalable and Flexible Framework for Automatically Learning Ensembles

AdaNet is a lightweight TensorFlow-based (Abadi et al., 2015) framework for automatically learning high-quality ensembles with minimal expert intervention. Our framework is inspired by the AdaNet algorithm (Cortes et al., 2017) which learns the structure of a neural network as an ensemble of subnetworks. We designed it to: (1) integrate with the existing TensorFlow ecosystem, (2) offer sensible default search spaces to perform well on novel datasets, (3) present a flexible API to utilize expert information when available, and (4) efficiently accelerate training with distributed CPU, GPU, and TPU hardware. The code is open-source and available at: https://github.com/tensorflow/adanet.

cs.LG↗

arXiv2019

Online Non-Additive Path Learning under Full and Partial Information

We study the problem of online path learning with non-additive gains, which is a central problem appearing in several applications, including ensemble structured prediction. We present new online algorithms for path learning with non-additive count-based gains for the three settings of full information, semi-bandit and full bandit with very favorable regret guarantees. A key component of our algorithms is the definition and computation of an intermediate context-dependent automaton that enables us to use existing algorithms designed for additive gains. We further apply our methods to the important application of ensemble structured prediction. Finally, beyond count-based gains, we give an efficient implementation of the EXP3 algorithm for the full bandit setting with an arbitrary (non-additive) gain.

cs.LG↗

arXiv2019

Foundations of Sequence-to-Sequence Modeling for Time Series

The availability of large amounts of time series data, paired with the performance of deep-learning algorithms on a broad class of problems, has recently led to significant interest in the use of sequence-to-sequence models for time series forecasting. We provide the first theoretical analysis of this time series forecasting framework. We include a comparison of sequence-to-sequence modeling to classical time series models, and as such our theory can serve as a quantitative guide for practitioners choosing between different modeling methodologies.

cs.LG↗

arXiv2018

Theory and Algorithms for Forecasting Time Series

We present data-dependent learning bounds for the general scenario of non-stationary non-mixing stochastic processes. Our learning guarantees are expressed in terms of a data-dependent measure of sequential complexity and a discrepancy measure that can be estimated from data under some mild assumptions. We also also provide novel analysis of stable time series forecasting algorithm using this new notion of discrepancy that we introduce. We use our learning bounds to devise new algorithms for non-stationary time series forecasting for which we report some preliminary experimental results.

cs.LG↗

arXiv2017

AdaNet: Adaptive Structural Learning of Artificial Neural Networks

We present new algorithms for adaptively learning artificial neural networks. Our algorithms (AdaNet) adaptively learn both the structure of the network and its weights. They are based on a solid theoretical analysis, including data-dependent generalization guarantees that we prove and discuss in detail. We report the results of large-scale experiments with one of our algorithms on several binary classification tasks extracted from the CIFAR-10 dataset. The results demonstrate that our algorithm can automatically learn network structures with very competitive performance accuracies when compared with those achieved for neural networks found by standard approaches.

cs.LG↗

arXiv2016

Structured Prediction Theory Based on Factor Graph Complexity

We present a general theoretical analysis of structured prediction with a series of new results. We give new data-dependent margin guarantees for structured prediction for a very wide family of loss functions and a general family of hypotheses, with an arbitrary factor graph decomposition. These are the tightest margin bounds known for both standard multi-class and general structured prediction problems. Our guarantees are expressed in terms of a data-dependent complexity measure, factor graph complexity, which we show can be estimated from data and bounded in terms of familiar quantities. We further extend our theory by leveraging the principle of Voted Risk Minimization (VRM) and show that learning is possible even with complex factor graphs. We present new learning bounds for this advanced setting, which we use to design two new algorithms, Voted Conditional Random Field (VCRF) and Voted Structured Boosting (StructBoost). These algorithms can make use of complex features and factor graphs and yet benefit from favorable learning guarantees. We also report the results of experiments with VCRF on several datasets to validate our theory.

stat.ML↗

arXiv2015

Voted Kernel Regularization

This paper presents an algorithm, Voted Kernel Regularization , that provides the flexibility of using potentially very complex kernel functions such as predictors based on much higher-degree polynomial kernels, while benefitting from strong learning guarantees. The success of our algorithm arises from derived bounds that suggest a new regularization penalty in terms of the Rademacher complexities of the corresponding families of kernel maps. In a series of experiments we demonstrate the improved performance of our algorithm as compared to baselines. Furthermore, the algorithm enjoys several favorable properties. The optimization problem is convex, it allows for learning with non-PDS kernels, and the solutions are highly sparse, resulting in improved classification speed and memory requirements.

cs.LG↗

arXiv2013

Nested Recursions, Simultaneous Parameters and Tree Superpositions

We apply a tree-based methodology to solve new, very broadly defined families of nested recursions of the general form R(n)=sum_{i=1}^k R(n-a_i-sum_{j=1}^p R(n-b_{ij})), where a_i are integers, b_{ij} are natural numbers, and k,p are natural numbers that we use to denote "arity" and "order," respectively, and with some specified initial conditions. The key idea of the tree-based solution method is to associate such recursions with infinite labelled trees in a natural way so that the solution to the recursions solves a counting question relating to the corresponding trees. We characterize certain recursion families within R(n) by introducing "simultaneous parameters" that appear both within the recursion itself and that also specify structural properties of the corresponding tree. First, we extend and unify recently discovered results concerning two families of arity k=2, order p=1 recursions. Next, we investigate the solution of nested recursion families by taking linear combinations of solution sequence frequencies for simpler nested recursions, which correspond to superpositions of the associated trees; this leads us to identify and solve two new recursion families for arity k=2 and general order p. Finally, we extend these results to general arity k>2. We conclude with several related open problems.

math.CO↗

arXiv2012

Nested recursions with ceiling function solutions

Consider a nested, non-homogeneous recursion R(n) defined by R(n) = \sum_{i=1}^k R(n-s_i-\sum_{j=1}^{p_i} R(n-a_ij)) + nu, with c initial conditions R(1) = xi_1 > 0,R(2)=xi_2 > 0, ..., R(c)=xi_c > 0, where the parameters are integers satisfying k > 0, p_i > 0 and a_ij > 0. We develop an algorithm to answer the following question: for an arbitrary rational number r/q, is there any set of values for k, p_i, s_i, a_ij and nu such that the ceiling function ceiling{rn/q} is the unique solution generated by R(n) with appropriate initial conditions? We apply this algorithm to explore those ceiling functions that appear as solutions to R(n). The pattern that emerges from this empirical investigation leads us to the following general result: every ceiling function of the form ceiling{n/q}$ is the solution of infinitely many such recursions. Further, the empirical evidence suggests that the converse conjecture is true: if ceiling{rn/q} is the solution generated by any recursion R(n) of the form above, then r=1. We also use our ceiling function methodology to derive the first known connection between the recursion R(n) and a natural generalization of Conway's recursion.

math.CO↗

arXiv2012

A combinatorial approach for solving certain nested recursions with non-slow solutions

We define the generalized Golomb triangular recursion by g_{j,s,lambda}(n) = g_{j,s,lambda}(n - s - g_{j,s,lambda}(n-j)) + λj. For particular choices of the initial conditions, we show that the solution of the recursion is a non-slow monotone sequence for which we can provide a combinatorial interpretation in terms of a weighted count of the leaves of a certain labeled infinite tree. We discover that more than one such tree interpretation is possible, leading to different choices of the initial conditions and alternative solutions that are closely related. In the case lambda=1 the initial conditions for these alternative tree interpretations coincide and we derive explicit closed forms for the solution sequence and its frequency function.

math.CO↗

arXiv2011

Sums of Ceiling Functions Solve Nested Recursions

It is known that, for given integers s \geq 0 and j > 0, the nested recursion R(n) = R(n - s - R(n - j)) + R(n - 2j - s - R(n - 3j)) has a closed form solution for which a combinatorial interpretation exists in terms of an infinite, labeled tree. For s = 0, we show that this solution sequence has a closed form as the sum of ceiling functions C(n). Further, given appropriate initial conditions, we derive necessary and sufficient conditions on the parameters s1, a1, s2 and a2 so that C(n) solves the nested recursion R(n) = R(n - s1 - R(n - a1)) + R(n- s2 - R(n - a2)).

math.CO↗