Hello Shubham, Thanks for getting in touch. The Augmented Recurrent Neural Networks project could be an exciting way to spend a summer.
> I would like to work on the topic "Augmented RNNs" which is closely related to > the work I am currently doing. I would like to know if we want the > implementation for any specific task such as Machine Translation (where > Seq2Seq > models can be useful). Is there any priority order of the tasks mentioned? > Regarding the implementation, does it also involve CUDA programming? As an > initial step, I have installed the mlpack and started looking more into the > code > structure. None of the mentioned models is necessary domain specific, but I agree you can draw a link between the use of external memory and e.g. language models. If you like you can go into a specific domain like Machine Translation that makes use of external memory when you implement the ideas. One interesting paper in that direction is: "Improving Neural Language Models with a Continuous Cache" by E. Grave et al. The project does not include any CUDA programming, we use armadillo as linear algebra library which can be build against NVBLAS which is a GPU- accelerated implementation of BLAS. Also, you don't have to implement all of the mentioned models, depending on the model choose one or two that you like to work on, you can also suggest other memory augmented models that you like to work on over the summer. > Please help me to get started. Definitely a good first step to take is to study the mlpack codebase closely and become familiar with its API, so that you can design a flexible and generic network structure that can be used interchangeable with existing mlpack classes. I hope this is helpful. Thanks, Marcus > On 13 Mar 2017, at 15:51, Shubham Agarwal <[email protected]> wrote: > > Hi, > > I am a second year Master student in Data Science. Currently, I am doing an > internship at Xerox Research Center Europe (XRCE), Grenoble, France under Dr. > Marc Dymetman on Deep RNN for Sequence Prediction. I graduated from > Mathematics and Computing at IIT Delhi and thereafter worked as Data > Scientist at TrulyMadly Matchmakers Pvt. Ltd, an online dating startup. > > I would like to work on the topic "Augmented RNNs" which is closely related > to the work I am currently doing. I would like to know if we want the > implementation for any specific task such as Machine Translation (where > Seq2Seq models can be useful). Is there any priority order of the tasks > mentioned? Regarding the implementation, does it also involve CUDA > programming? As an initial step, I have installed the mlpack and started > looking more into the code structure. > > Please help me to get started. > > Thank you. > > Best, > > Shubham Agarwal > Master 2 Data Science > Master of Science in Industrial and Applied Maths (MSIAM) > Ensimag Grenoble INP > _______________________________________________ > mlpack mailing list > [email protected] > http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
_______________________________________________ mlpack mailing list [email protected] http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
