Large Scale Machine Learning

نویسندگان

  • Ronan Collobert
  • Patrick Gallinari
  • Léon Bottou
  • Hélène Paugam-Moisy
  • Samy Bengio
  • Yves Grandvalet
چکیده

Cette thèse aborde de façon générale les algorithmes d'apprentissage, avec un intérêt tout particulier pour les grandes bases de données. Après avoir for-mulé leprobì eme de l'apprentissage demanì ere mathématique, nous présentons plusieurs algorithmes d'apprentissage importants, en particulier les Multi Layer Perceptrons, les Mixture d'Experts ainsi que les Support Vector Machines. Nous considérons ensuite une méthode d'entraˆınement pour les Support Vector Machines , adaptée aux ensembles de données de tailles raisonnables. Cepen-dant, l'entraˆınement d'un tel modèle reste irréalisable sur de très grande bases de données. Inspirés par la stratégie " diviser pour régner " , nous proposons alors un modèle de la famille des Mixture d'Experts, permettant de séparer le probì eme d' apprentissage en sous-probì emes plus simples , tout en gardant de bonnes performances en généralisation. Malgré de très bonnes performances en pratique , cet algorithme n ' en reste pas moins difficilè a utiliser , ` a cause de son nombre important d ' hyper-paramètres. Pour cette raison , nous préférons nous intéresser ensuitè a l ' amélioration de l ' entraˆınement des Multi Layer Percep-trons , bien plus facilesà utiliser , et plus adaptés aux grandes bases de données que les Support Vector Machines. Enfin , nous montrons que l ' idée de la marge qui fait la force des Support Vector Machines peutêtre appliquéè a une cer-taine classe de Multi Layer Perceptrons , ce qui nous m ` enè a un algorithme très rapide et ayant de très bonnes performances en généralisation. Summary This thesis aims to address machine learning in general , with a particular focus on large models and large databases. After introducing the learning problem in a formal way , we first review several important machine learning algorithms , particularly Multi Layer Perceptrons , Mixture of Experts and Support Vector Machines. We then present a training method for Support Vector Machines , adapted to reasonably large datasets. However the training of such a model is still intractable on very large databases. We thus propose a divide and conquer approach based on a kind of Mixture of Experts in order to break up the training problem into small pieces , while keeping good generalization performance. This mixture model can be applied to any kind of existing machine learning algorithm. Even though it performs well in practice the major drawback of this algorithm is the number of hyper-parameters to tune , which makes it …

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Machine Learning and Citizen Science: Opportunities and Challenges of Human-Computer Interaction

Background and Aim: In processing large data, scientists have to perform the tedious task of analyzing hefty bulk of data. Machine learning techniques are a potential solution to this problem. In citizen science, human and artificial intelligence may be unified to facilitate this effort. Considering the ambiguities in machine performance and management of user-generated data, this paper aims to...

متن کامل

Comparative Analysis of Machine Learning Algorithms with Optimization Purposes

The field of optimization and machine learning are increasingly interplayed and optimization in different problems leads to the use of machine learning approaches‎. ‎Machine learning algorithms work in reasonable computational time for specific classes of problems and have important role in extracting knowledge from large amount of data‎. ‎In this paper‎, ‎a methodology has been employed to opt...

متن کامل

A New Play-off Approach in League Championship Algorithm for Solving Large-Scale Support Vector Machine Problems

There are many numerous methods for solving large-scale problems in which some of them are very flexible and efficient in both linear and non-linear cases. League championship algorithm is such algorithm which may be used in the mentioned problems. In the current paper, a new play-off approach will be adapted on league championship algorithm for solving large-scale problems. The proposed algori...

متن کامل

Two-stage fuzzy-stochastic programming for parallel machine scheduling problem with machine deterioration and operator learning effect

This paper deals with the determination of machine numbers and production schedules in manufacturing environments. In this line, a two-stage fuzzy stochastic programming model is discussed with fuzzy processing times where both deterioration and learning effects are evaluated simultaneously. The first stage focuses on the type and number of machines in order to minimize the total costs associat...

متن کامل

Large-Scale Machine Learning for Classification and Search

Large-Scale Machine Learning for Classification and Search

متن کامل

End-to-End Large Scale Machine Learning with KeystoneML

End-to-End Large Scale Machine Learning with KeystoneML

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2004