User:Enricobertino

Public application

A: An introduction

• Please describe yourself in three sentences, one of them regarding your current studies.

  I’m Enrico Bertino, an Italian student. After a BSc in mathematical engineering in Politecnico di Milano and a MSc in France, I enrolled in the MSc major Statistics with specialization in big data in Politecnico di Milano and I’m now finishing the last year.

  • Which languages do you speak?

    I speak Italian and French (native) and English (fluent)

  • What's your overall background?

    In the past I spent six months as a researcher in fluid dynamics, programming mostly in Octave. For two years now I have been more passionated about applied statistics, with focus on deep learning and Bayesian approach.

• Why do you want to participate in the Google Summer of Code? What do you hope to gain by doing so?

  It is an interesting project, followed by competent mentors and supported by a great organization. I can not ask for more for entering the open-source world :)

  • Please also describe your previous experience with the GSoC, if any.

    It is the first time for me!

• Why are you choosing Octave?

  I am quite a fan of open source projects, open innovation and expertise sharing. In recent months I have been working on a Python framework for natural language processing and I want to open it asap. I like this same spirit under the Octave project.

C: Contact

• Please state the (unique and identical where possible) nick you use on IRC and any other communication channel related to Octave.

  nick: enricobertino

• Which time zone do you live in? Will that change over GSoC duration?

  UTC+1, it will not change

• Please state the timeframe (in UTC+0) when you feel most comfortable working during GSoC. Where are your time buffers?

  I usually code in the time span 8.00 - 19.00 (UTC+0)

E: Coding experience

This part is one of the more important ones in your application. You are allowed to be as verbose as you want, as long as you stay on topic ;-)

• Please describe your experience with C++, Octave or Matlab m-scripts, OpenGL and Qt.

  I have worked with Matlab and Octave for a long time, in particular with numerical methods for PDEs and statistical analysis. I have a considerable experience with C++ since it is the most used language in my master exams.

• Please describe your experience with other programming languages.

  I’m very familiar with Python and R, the two most common languages for statisticians. I have also some experience in Java and PHP.

• Please describe your experience with being in a development team.

  I’m currently working in a tech startup and I have to collaborate with other 5 developers who have different academic profiles and different skills . We have structured the code from scratch and the process can be likened to a package development.

• Please describe the biggest project you have written code for and what you learned by doing so. Also describe your role in that project over time.

  1. Octave: When I was at INRIA in France, I implemented a Godunov’s scheme to resolve stochastic conservation laws in modeling highway traffic. I started with the implementation of a deterministic method, fitting the parameters from real data gathered from a gps company. In this phase I learned to deal with data input and storage problems. In the second part of the project I modeled a stochastic version of the method with an optimized multidimensional Monte-Carlo approach, which gave me some expertise about computational optimization.
  2. Python: As part of a machine learning exam, I I have participated with other classmates to a Kaggle competition which consisted in the prediction of the outcomes of shelter animals adoptions. Since we were five in the group, I learned the fundamental basis of team development. We used both mathematical and programming skills and we reached the third place worldwide.
  3. C++: During a master project I had to implement a Biham-Middleton-Levine traffic model in C++. Long story short, it consisted in some tricky operations on a matrix. I learned to deal with Eigen, sparse matrices and heritage. At the beginning we coded a fancy structure based on polymorphism, but since the goal of the project was only the rapidity of the code, we rewrite every class in order to optimize the computational time. We implemented also two parallelized versions of the code, with OpenMP and MPI, and the results were highly satisfactory.

  The first project was private, but I pushed the other two on a github repository.

• Please state the commits and patches you already contributed to Octave.

  In order to familiarize with Octave patches, I posted a patch for nnet package (link patch). I’m currently working on some Pytave bugs.

F: Feeling fine

• Please describe (in short) your experience with the following tools:
  • IRC and mailing lists

    I am starting to discover the Octave “ecosystem” and to be comfortable using the mailing list and IRC

  • Mercurial or other source code management systems

    I’m very familiar with Git, just started with Mercurial

  • Mediawiki or other wiki software

    Kind of

  • make, gcc, gdb or other development tools

    Familiar with make and gcc, less with gdb

• What will make you actively stay in our community after this GSoC is over?

  Keeping on with research in university

O: Only out of interest

• Did you ever hear about Octave before?

  Of course

  • If so, when and where? How far have you been involved already?

    Until three years ago I just used Matlab since it was the common choice at the university. After that I heard about Octave when I was in a researcher lab and I start working mostly on it.

• What was the first question concerning Octave you could not find an answer to rather quickly?

  As normal user, I do not remember a question that was difficult to answer.

P: Prerequisites

• Please state the operating system you work with.

  Ubuntu and macOS

  • If you have access to more than one, please state them and the conditions under which you are granted this access.

    Direct access to macOS and ssh access to Ubuntu, always available.

• Please estimate an average time per day you will be able to access
  • an internet connection

    24

  • a computer

    24

  • a computer with your progressing work on

    24

• Please describe the degree to which you can install new software on computers you have access to.

  Full permissions

S: Self-assessment

• Please describe how useful criticism looks from your point of view as committing student.

  Indispensable!

• How autonomous are you when developing:
  • Do you like to discuss changes intensively and not start coding until you know what you want to do?

    In general I like to start coding when everybody shares a final decision.

  • Do you like to code a proof of concept to 'see how it turns out'?

    I love to test things so often I want to have a little version of something which works and then analyze the whole problem.

Y: Your task

• Did you select a task from our list of proposals and ideas?

  Yes, I would like to work on the project: "Neural_Networks_package” Wiki link to the project

• Please provide a rough estimated timeline for your work on the task.

  05/05 - 30/05 (community bonding period)

  Week 1

  - Begin to stay in contact and familiarize with the community, using both the mailing list and the IRC channel.

  - Improve expertise with tools like Mercurial and autotools

  - Read Matlab doc of Neural Network Toolbox classes and basic functions, with a focus on the deep learning part about CNNs

  Week 2

  - Deeply analyze both Python and C++ Tensorflow APIs in order to figure out the best path to follow

  - Install and run Pytave. Read the doc if provided or exchange with the maintainers

  Week 3

  - Test Pytave and figure out there is some bug or missing feature of the specific part that we need.

  - Fix potential bugs or submit new patches

  - Figure out if we need some object programming in Octave (like classdef) and test it

  30/05 - 30/06 (Phase 1)

  Week 4,5

  - Work on the makefile in order to link TF in either Python or C++ and test soma basic nets with TF

  Week 6,7

  - Write all the Octave classes for every layer and use corresponding TF functions. Because of the focus on Matlab Nnet Toolbox, we will start to define the fundamental layers used for CNNs : Convolutional layer, ReLU layer, Normalization layer, Average pooling layer object, Max pooling layer, Fully connected layer, Dropout layer, Softmax layer, Classification output layer, Regression output layer

  Week 8

  - Implement a draft of the training functions (seriesNetwork object, trainNetwork, trainingOptions) without all options and parameters

  01/07 - 28/07 (Phase 2)

  Week 9-10

  - Implement a complete working version of seriesNetwork (including definition of methods like activations, classify and predict)

  Week 11

  - Implement a complete working version of trainNetwork

  Week 12

  - Implement a complete working version of trainingOptions

  28/07 - 25/08 (Final phase)

  Week 13

  - Able the parallelization and analyze CUDA integration

  Week 14-15

  - Implementation of a cool application like deepDreamImage and (if some time is left) more advanced nets like (AlexNet, vgg16, vgg19)

  Week 16

  - As a bonus if some time is left, try to implement ImageDatastore in order to manage seamless the image import

  All tests and documentation will be written during the whole period, simultaneously with every function. Both BIST tests and doc tests will be written, along with Python unit tests.

  Clearly this is a draft, all the steps will become clearer with some discussions with the mentors.