AIS Travel Support for the Data Representation Discovery Workshop to be held on August 12-17, 2007 in Orlando, FL

提案编号：ECS-0736687提案标题：AIS对数据表示发现工作坊的旅行支持PI名称：Guyon，IsabellePI机构：Clopinet这笔赠款将资助五名学生参加下一次数据表示发现研讨会的旅费。讲习班的目的是评估领域专门知识在设计有效预测模型方面的附加值。在数据挖掘和机器学习的框架中，预测模型是从训练样本中建立的，以预测给定的结果，例如医疗诊断中患者的健康状况，或者市场营销中的销售预测。所采取的方法是组织一场挑战赛，参与者在来自不同应用领域(营销、手写识别、药物发现、文本分类和生态)的五个数据集上进行竞争。数据以其“原始”表示和特征向量两种形式可用。例如，在药物发现领域，任务是预测给定化合物对艾滋病毒的活性；化学式已作为原始数据提供，量化物理化学分子特性的系数形成特征表示。本次挑战的结果将于2007年8月1日结束，http://clopinet.com/isabelle/Projects/agnostic/将在研讨会上进行分析和讨论。智力优势：确定整合领域知识与使用现成建模软件的有效性，无论是在模型的上市时间和预测性能方面，都对行业至关重要，但没有明显的量化。更广泛的影响：几名学生在比赛中脱颖而出。讲习班将有助于传播他们的成果和软件(以及其他研究人员的成果和软件)，并为他们之间的合作奠定基础。讲习班还将作为一个平台，规划未来的挑战活动并让学生参与这一进程。

Proposal Number: ECS-0736687Proposal Title: AIS Travel Support for the Data Representation Discovery WorkshopPI Name: Guyon, IsabellePI Institution: ClopinetThis grant will fund the travel of five students to participate in the next Data Representation Discovery Workshop. The objective of the workshop is to evaluate the added value of domain expertise in the design of efficient predictive models. In the framework of data mining and machine learning, predictive models are built from training examples to make predictions of given outcomes, e.g. the health status of a patient in medical diagnosis, or the sales projections in marketing. The approach taken is to organize a challenge in which the participants competed on five datasets from various application domains (marketing, handwriting recognition, drug discovery, text categorization, and ecology). The data are made available both in their "raw" representation and as feature vectors. For instance, in the drug discovery domain, the task is to predict the activity of given compounds against the HIV virus; the chemical formulas have been provided as raw data and coefficients quantifying physico-chemical molecular properties formed the feature representation. The outcomes of this challenge ending August 1st, 2007 http://clopinet.com/isabelle/Projects/agnostic/ will be analyzed and discussed at the workshop.Intellectual merit: Determining the effectiveness of incorporating domain knowledge vs. using off-the-shelf modeling software, both in terms of time to market delivery of models and gain in prediction performance, is essential to the Industry, yet no obvious to quantify. Broader impacts:Several students distinguished themselves in the competition. The workshop will help disseminating their results and software (and those of other researchers) and lay the basis of collaborations between them. The workshop will also serve as a platform to plan for future challenge events and involve students in that process.