Efficient Nonmyopic Active Search
نویسندگان
چکیده
Active search is a learning paradigm with the goal of actively identifying as many members of a given class as possible. Many real-world problems can be cast as an active search, including drug discovery, fraud detection, and product recommendation. Previous work has derived the Bayesian optimal policy for the problem, which is unfortunately intractable due to exponential complexity. In practice, myopic approximations are used instead, only looking a small number (e.g., 1–3) of steps ahead in the decision process. We propose a novel active search policy that always considers the entire remaining budget and is thus nonmyopic, yet remains efficient. Our approach automatically and dynamically balances exploration and exploitation in a manner consistent with the budget, without relying on a tradeoff parameter. We also develop a bounding technique to achieve greater efficiency when using certain natural probability models. Experimental results show superior performance of our method over myopic approximations to the optimal policy.
منابع مشابه
Supplementary Material for Efficient Nonmyopic Active Search
In this section, we present the proof of Theorem 1. We assume that active search policies have access to the correct marginal probabilities f(x;D) = Pr(y = 1 | x,D), for any given point x and labeled data D, which may include “ficticious” observations. Further, the computational cost will be analyzed as the number of calls to f , i.e., f(x;D) has unit cost. Note that the optimal policy operates...
متن کاملNonmyopic Active Learning of Gaussian Processes: An Exploration–Exploitation Approach
When monitoring spatial phenomena, such as the ecological condition of a river, deciding where to make observations is a challenging task. In these settings, a fundamental question is when an active learning, or sequential design, strategy, where locations are selected based on previous measurements, will perform significantly better than sensing at an a priori specified set of locations. For G...
متن کاملNonmyopic -Bayes-Optimal Active Learning of Gaussian Processes
A fundamental issue in active learning of Gaussian processes is that of the explorationexploitation trade-off. This paper presents a novel nonmyopic -Bayes-optimal active learning ( -BAL) approach that jointly and naturally optimizes the trade-off. In contrast, existing works have primarily developed myopic/greedy algorithms or performed exploration and exploitation separately. To perform activ...
متن کاملActive Learning Is Planning: Nonmyopic ε-Bayes-Optimal Active Learning of Gaussian Processes
A fundamental issue in active learning of Gaussian processes is that of the exploration-exploitation trade-off. This paper presents a novel nonmyopic -Bayes-optimal active learning ( -BAL) approach [4] that jointly optimizes the trade-off. In contrast, existing works have primarily developed greedy algorithms or performed exploration and exploitation separately. To perform active learning in re...
متن کاملEfficient Planning of Informative Paths for Multiple Robots
When monitoring spatial phenomena, such as the ecological condition of a river, deciding where to make observations is a challenging task. In these settings, a fundamental question is when an active learning, or sequential design, strategy, where locations are selected based on previous measurements, will perform significantly better than sensing at an a priori specified set of locations. For G...
متن کامل