Multi-Objective Multi-Task Learning
نویسنده
چکیده
This dissertation presents multi-objective multi-task learning, a new learning framework. Given a fixed sequence of tasks, the learned hypothesis space must minimize multiple objectives. Since these objectives are often in conflict, we cannot find a single best solution, so we analyze a set of solutions. We first propose and analyze a new learning principle, empirically efficient learning. From a sample complexity perspective, following this principle is not much worse than the single-objective multi-task learning case. In the context of empirically efficient learning, algorithms for the new learning frameworks are proposed and evaluated. First, we pose regularization as a multi-objective problem, in which training error must balance the complexity of the hypothesis space. Second, we consider multiple data-dependent loss functions, in which the error rate in one class must balance the error rate in the other class. Finally, we assume that tasks share a clustering structure in which the average loss in one cluster must balance the loss in another cluster. The algorithms are evaluated on synthetic and real datasets. The results motivate the application of multi-objective optimization, indicating that the objectives are in conflict. By controlling the relative performance of the algorithms to generate a tradeoff surface, we can effectively explore the multi-objective nature of the learning problem.
منابع مشابه
MMDT: Multi-Objective Memetic Rule Learning from Decision Tree
In this article, a Multi-Objective Memetic Algorithm (MA) for rule learning is proposed. Prediction accuracy and interpretation are two measures that conflict with each other. In this approach, we consider accuracy and interpretation of rules sets. Additionally, individual classifiers face other problems such as huge sizes, high dimensionality and imbalance classes’ distribution data sets. This...
متن کاملMulti-objective Differential Evolution for the Flow shop Scheduling Problem with a Modified Learning Effect
This paper proposes an effective multi-objective differential evolution algorithm (MDES) to solve a permutation flow shop scheduling problem (PFSSP) with modified Dejong's learning effect. The proposed algorithm combines the basic differential evolution (DE) with local search and borrows the selection operator from NSGA-II to improve the general performance. First the problem is encoded with a...
متن کاملSolving a New Multi-objective Unrelated Parallel Machines Scheduling Problem by Hybrid Teaching-learning Based Optimization
This paper considers a scheduling problem of a set of independent jobs on unrelated parallel machines (UPMs) that minimizesthe maximum completion time (i.e., makespan or ), maximum earliness ( ), and maximum tardiness ( ) simultaneously. Jobs have non-identical due dates, sequence-dependent setup times and machine-dependentprocessing times. A multi-objective mixed-integer linear programmi...
متن کاملMulti-objective and Scalable Heuristic Algorithm for Workflow Task Scheduling in Utility Grids
To use services transparently in a distributed environment, the Utility Grids develop a cyber-infrastructure. The parameters of the Quality of Service such as the allocation-cost and makespan have to be dealt with in order to schedule workflow application tasks in the Utility Grids. Optimization of both target parameters above is a challenge in a distributed environment and may conflict one an...
متن کاملImproving Agent Performance for Multi-Resource Negotiation Using Learning Automata and Case-Based Reasoning
In electronic commerce markets, agents often should acquire multiple resources to fulfil a high-level task. In order to attain such resources they need to compete with each other. In multi-agent environments, in which competition is involved, negotiation would be an interaction between agents in order to reach an agreement on resource allocation and to be coordinated with each other. In recent ...
متن کامل