Ofir Pele: "Interpolated Discretized Embedding of Single Objects and Object Pairs for Classification, Semi Metric Learning and Distance Approximation"

×

Error message

  • Deprecated function: Creation of dynamic property LdapUserConf::$createLDAPAccounts is deprecated in LdapUserConf->load() (line 265 of /var/lib/drupal7/modules/ldap/ldap_user/LdapUserConf.class.php).
  • Deprecated function: Creation of dynamic property LdapUserConf::$createLDAPAccountsAdminApproval is deprecated in LdapUserConf->load() (line 266 of /var/lib/drupal7/modules/ldap/ldap_user/LdapUserConf.class.php).

We present a method for the approximation and learning of arbitrary continuous distance functions and semimetrics. Previous work have been limited to Euclidean metrics or other families of metrics (e.g., the Earth Mover's Distance). One usage of this method is to learn a function that given two objects returns whether they match (e.g., descriptor matching learning). Another usage is directly learning a distance function (e.g., perceptual color difference learning). The method can also be used for approximation of distances and for single object classification or regression tasks. The time complexity of computing the distance function is relatively efficient: $O(n\log(nc))$, where n is the dimension of the input vectors and c is a parameter that determines the distance function complexity.

Date and Time: 
Thursday, May 26, 2016 - 13:30 to 14:30
Speaker: 
Ofir Pele
Location: 
IDC, C.110
Speaker Bio: 

<a href="http://www.ariel.ac.il/sites/ofirpele/">Ofir Pele</a>,  Department of Computer Science and Mathematics,  Ariel University