Hello everyone, It's been a great summer and we're finally at the end of it. Earlier this week, the results from Google Summer of Code were announced, and I'm happy to report that all six of our students this year passed (all with flying colors of course!). Here's a quick summary and status of each project:
Yasmine Dumouchel (mentored by me): Go bindings are being automatically generated successfully and they are tested; there are just a handful more things to work out before merging it into mlpack entirely, then mlpack can be used from Go! Wenhao Huang (mentored by Mikhail Lozhnikov): The CF (collaborative filtering) module gains the ability to remove global effects and to use weighted sums for the rating calculation. Moreover some decomposition techniques like SVD++ and Bias SVD will be merged soon after some final checks. Shikhar Jaiswal (mentored by Marcus Edel): Several Generative adversarial networks (GAN, DCGAN and WGAN) are implemented; the GAN implementation is almost 1.5 times faster than a comparable Tensorflow implementation (CPU); on top of that, restricted Boltzmann machines (RBM's) were implemented including several optimizations for the network code. Atharva Khandait (mentored by Sumedh Ghaisas): Multilayer perceptron and convolutional VAE's (Variational Autoencoder) were implemented, in addition to a reconstruction of the existing loss computation structure and several optimizations and fixes (Sequence layer, Gradient calculation to name a few) for the network code; some bits are still being optimized and should be merged in the future. Manish Kumar (mentored by me): Both LMNN and BoostMetric (distance learning techniques) are implemented; LMNN is very fast and still being optimized further, and BoostMetric should be merged in the future). Haritha Nair (mentored by Marcus Edel): The neural collaborative filtering (NCF) framework was implemented; besides introducing the policy design pattern into the existing CF framework, there are some issues we have to work out before we can entirely merge all additions and modifications. Each of these projects was a great effort, and I'm really excited about seeing these features become parts of subsequent mlpack releases. Thank you to everyone who participated in this program, including all of the students who applied and contributed during the application process, all of the mentors who helped out, and all of the students who were accepted and did great work on their projects. We'll be releasing mlpack 3.1.0 shortly, which will include the code generated from these projects. :) Have a great weekend! -- Ryan Curtin | "If it's something that can be stopped, then just try to stop it!" r...@ratml.org | - Skull Kid _______________________________________________ mlpack mailing list mlpack@lists.mlpack.org http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack