Ram Akella

Optimal control of production rate in a failure prone manufacturing system

Consider a manufacturing system producing a single commodity. The manufacturing system can be in one of two states: functional and failed. It moves back and forth between these two states as a continuous time Markov chain, with mean time between failures = 1/q1, and mean time to repair 1/q2. When functional, the manufacturing system can produce at up to a maximum rate d. When failed, it cannot produce the commodity at all.

Single-Period Multiproduct Inventory Models with Substitution

We study a single period multiproduct inventory problem with substitution and proportional costs and revenues. We consider N products and N demand classes with full downward substitution, i.e., excess demand for class i can be satisfied using product j for i ≥ j. We first discuss a two-stage profit maximization formulation for the multiproduct substitution problem. We show that a greedy allocation policy is optimal. We use this to write the expected profits and its first partials explicitly. This in turn enables us to prove additional properties of the profit function and several interesting properties of the optimal solution. In a limited computational study using two products, we illustrate the benefits of solving for the optimal quantities when substitution is considered at the ordering stage over similar computations without considering substitution while ordering. Specifically, we show that the benefits are higher with high demand variability, low substitution cost, low profit margins (or low price to cost ratio), high salvage values, and similarity of products in terms of prices and costs.

Supply management in assembly systems with random yield and random demand

Abstract In this paper, we consider an assembly system where a firm faces random demand for a finished product which is assembled using two critical components. The components are procured from the suppliers who, due to production yield losses, deliver a random fraction of the order quantity. We formulate the exact cost function where the decision variables are the target level of finished products to assemble, and the order quantity of the components from the suppliers. Since the exact cost function is analytically complex to solve, we introduce a modified cost function and derive bounds on the difference in the objective function values. Using the modified cost function, we determine the combined component ordering and production (assembly) decisions for the firm. The benefit of coordinating ordering and assembly decisions is numerically demonstrated by comparing the results with two heuristic policies commonly used in practice. In an extension to the model, we consider the case when the firm has the added option of ordering both the components in a set from a joint supplier. First, we consider the case when the joint supplier is reliable in delivery and obtain dominance conditions on the suppliers to be chosen. The maximum price a firm would be willing to pay to ensure reliable supply of components is determined. Later, we consider the uncertainty in the deliveries from the joint supplier and determine conditions under which there is no diversification, that is, either the individual suppliers are used, or the joint supplier is used, but never both.

Incorporating diversity and density in active learning for relevance feedback

Relevance feedback, which uses the terms in relevant documents to enrich the users initial query, is an effective method for improving retrieval performance. An associated key research problem is the following: Which documents to present to the user so that the users feedback on the documents can significantly impact relevance feedback performance. This paper views this as an active learning problem and proposes a new algorithm which can efficiently maximize the learning benefits of relevance feedback. This algorithm chooses a set of feedback documents based on relevancy, document diversity and document density. Experimental results show a statistically significant and appreciable improvement in the performance of our new approach over the existing active feedback methods.

Optimal order policies in assembly systems with random demand and random supplier delivery

In this paper we consider an assembly problem where two critical components are required for assembly of the final product, the demand for which is stochastic. The components can be ordered separat...

Reducing Features to Improve Code Change-Based Bug Prediction

Machine learning classifiers have recently emerged as a way to predict the introduction of bugs in changes made to source code files. The classifier is first trained on software history, and then used to predict if an impending change causes a bug. Drawbacks of existing classifier-based bug prediction techniques are insufficient performance for practical use and slow prediction times due to a large number of machine learned features. This paper investigates multiple feature selection techniques that are generally applicable to classification-based bug prediction methods. The techniques discard less important features until optimal classification performance is reached. The total number of features used for training is substantially reduced, often to less than 10 percent of the original. The performance of Naive Bayes and Support Vector Machine (SVM) classifiers when using this technique is characterized on 11 software projects. Naive Bayes using feature selection provides significant improvement in buggy F-measure (21 percent improvement) over prior change classification bug prediction results (by the second and fourth authors [28]). The SVMs improvement in buggy F-measure is 9 percent. Interestingly, an analysis of performance for varying numbers of features shows that strong performance is achieved at even 1 percent of the original number of features.

In-line defect sampling methodology in yield management: an integrated framework

In this paper we provide an integrated framework for designing the optimal defect sampling strategy for wafer inspection, which is crucial in yield management of state-of-the-art technologies. We present a comprehensive cost-based methodology which allows us to achieve the trade-off between the cost of inspection and the cost of yield impact of the undetected defects. We illustrate the effectiveness of our methodology using data from several leading fablines across the world. We demonstrate that this work has already caused a significant change in the sampling practices in these fablines especially in the area of defect data preprocessing (declustering), in-line defect based yield prediction, and optimization of wafer inspection equipment allocation.

Reducing Features to Improve Bug Prediction

Recently, machine learning classifiers have emerged as a way to predict the existence of a bug in a change made to a source code file. The classifier is first trained on software history data, and then used to predict bugs. Two drawbacks of existing classifier-based bug prediction are potentially insufficient accuracy for practical use, and use of a large number of features. These large numbers of features adversely impact scalability and accuracy of the approach. This paper proposes a feature selection technique applicable to classification-based bug prediction. This technique is applied to predict bugs in software changes, and performance of Naive Bayes and Support Vector Machine (SVM) classifiers is characterized.

Control of batch processing systems in semiconductor wafer fabrication facilities

Loading policies for a batch processing machine, i.e. a machine that can process more than one job at a time, when the arrival times of jobs to the machine are uncertain, are described. The motivation for the study is the structure of process flows and the predominance of batch processing systems in a semiconductor wafer fabrication facility. A two stage serial-batch system with the serial stage (e.g. photolithography) feeding the batch (e.g. furnace) is considered. Machines in the serial stage process one job at a time; further, these machines are subject to failure. A control limit policy for loading the batch machine is assumed, i.e. load if the queue length >or=Q, else wait until the number of jobs in queue is at least Q. The basic tradeoffs considered are delay (waiting too long) vs. capacity utilization (loading early with very few jobs). An average cost analysis is done and optimized to compute the critical number Q. In an extension to the basic model, the effect of due dates on the critical number is analyzed. Comparison with simulation results is very encouraging. >

A bayesian logistic regression model for active relevance feedback

Relevance feedback, which traditionally uses the terms in the relevant documents to enrich the users initial query, is an effective method for improving retrieval performance. The traditional relevance feedback algorithms lead to overfitting because of the limited amount of training data and large term space. This paper introduces an online Bayesian logistic regression algorithm to incorporate relevance feedback information. The new approach addresses the overfitting problem by projecting the original feature space onto a more compact set which retains the necessary information. The new set of features consist of the original retrieval score, the distance to the relevant documents and the distance to non-relevant documents. To reduce the human evaluation effort in ascertaining relevance, we introduce a new active learning algorithm based on variance reduction to actively select documents for user evaluation. The new active learning algorithm aims to select feedback documents to reduce the model variance. The variance reduction approach leads to capturing relevance, diversity and uncertainty of the unlabeled documents in a principled manner. These are the critical factors of active learning indicated in previous literature. Experiments with several TREC datasets demonstrate the effectiveness of the proposed approach.